Runtime update: released native 1.0.5, memory flags (VAE tile 256, lazy KV handoff), verified-digest cache, background VAE load, chunked INT8 head quantization, compile paths only with --warm-compile; measurements refreshed
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- .gitattributes +1 -0
- README.md +12 -10
- SHA256SUMS +92 -70
- evaluation/profiles/baseline.json +0 -158
- evaluation/profiles/exact/run-00/config.json +3 -3
- evaluation/profiles/exact/run-00/metrics.json +94 -53
- evaluation/profiles/exact/run-00/plan.json +6 -6
- evaluation/profiles/exact/run-00/plan_manifest.json +1 -1
- evaluation/profiles/exact/run-00/result.json +85 -45
- evaluation/profiles/exact/run-01/config.json +3 -3
- evaluation/profiles/exact/run-01/metrics.json +94 -53
- evaluation/profiles/exact/run-01/plan.json +6 -6
- evaluation/profiles/exact/run-01/plan_manifest.json +1 -1
- evaluation/profiles/exact/run-01/result.json +85 -45
- evaluation/profiles/exact/summary.json +188 -106
- evaluation/profiles/fast-cfg1.5/run-00/config.json +3 -3
- evaluation/profiles/fast-cfg1.5/run-00/metrics.json +94 -54
- evaluation/profiles/fast-cfg1.5/run-00/plan.json +6 -6
- evaluation/profiles/fast-cfg1.5/run-00/plan_manifest.json +1 -1
- evaluation/profiles/fast-cfg1.5/run-00/result.json +83 -44
- evaluation/profiles/fast-cfg1.5/run-01/config.json +3 -3
- evaluation/profiles/fast-cfg1.5/run-01/metrics.json +94 -54
- evaluation/profiles/fast-cfg1.5/run-01/plan.json +6 -6
- evaluation/profiles/fast-cfg1.5/run-01/plan_manifest.json +1 -1
- evaluation/profiles/fast-cfg1.5/run-01/result.json +83 -44
- evaluation/profiles/fast-cfg1.5/summary.json +188 -108
- evaluation/profiles/fast/run-00/config.json +3 -3
- evaluation/profiles/fast/run-00/metrics.json +94 -54
- evaluation/profiles/fast/run-00/plan.json +6 -6
- evaluation/profiles/fast/run-00/plan_manifest.json +1 -1
- evaluation/profiles/fast/run-00/result.json +83 -44
- evaluation/profiles/fast/run-01/config.json +3 -3
- evaluation/profiles/fast/run-01/metrics.json +94 -54
- evaluation/profiles/fast/run-01/plan.json +6 -6
- evaluation/profiles/fast/run-01/plan_manifest.json +1 -1
- evaluation/profiles/fast/run-01/result.json +83 -44
- evaluation/profiles/fast/summary.json +188 -108
- evaluation/profiles/lowmem/run-00/config.json +3 -3
- evaluation/profiles/lowmem/run-00/metrics.json +95 -55
- evaluation/profiles/lowmem/run-00/plan.json +6 -6
- evaluation/profiles/lowmem/run-00/plan_manifest.json +1 -1
- evaluation/profiles/lowmem/run-00/result.json +84 -45
- evaluation/profiles/lowmem/run-01/config.json +3 -3
- evaluation/profiles/lowmem/run-01/metrics.json +95 -55
- evaluation/profiles/lowmem/run-01/plan.json +6 -6
- evaluation/profiles/lowmem/run-01/plan_manifest.json +1 -1
- evaluation/profiles/lowmem/run-01/result.json +84 -45
- evaluation/profiles/lowmem/summary.json +190 -110
- evaluation/profiles/off/run-00/config.json +88 -0
- evaluation/profiles/off/run-00/metrics.json +165 -0
.gitattributes
CHANGED
|
@@ -97,3 +97,4 @@ evaluation/profiles/lowmem/run-00/audio.flac filter=lfs diff=lfs merge=lfs -text
|
|
| 97 |
evaluation/profiles/lowmem/run-01/audio.flac filter=lfs diff=lfs merge=lfs -text
|
| 98 |
evaluation/profiles/turbo/run-00/audio.flac filter=lfs diff=lfs merge=lfs -text
|
| 99 |
evaluation/profiles/turbo/run-01/audio.flac filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
| 97 |
evaluation/profiles/lowmem/run-01/audio.flac filter=lfs diff=lfs merge=lfs -text
|
| 98 |
evaluation/profiles/turbo/run-00/audio.flac filter=lfs diff=lfs merge=lfs -text
|
| 99 |
evaluation/profiles/turbo/run-01/audio.flac filter=lfs diff=lfs merge=lfs -text
|
| 100 |
+
src/wheels/orbitquant-0.9.8-py3-none-any.whl filter=lfs diff=lfs merge=lfs -text
|
README.md
CHANGED
|
@@ -24,6 +24,8 @@ pip install torch==2.10.0 --index-url https://download.pytorch.org/whl/cu128
|
|
| 24 |
python src/run.py --profile fast --fuse-rms-quant --kv-cache-dtype int8 --mode fast --output out
|
| 25 |
```
|
| 26 |
|
|
|
|
|
|
|
| 27 |
| `--mode` | Changes | Logits |
|
| 28 |
|---|---|---|
|
| 29 |
| `off` | reference decode path | reference |
|
|
@@ -32,21 +34,21 @@ python src/run.py --profile fast --fuse-rms-quant --kv-cache-dtype int8 --mode f
|
|
| 32 |
| `lowmem` | + INT8 token embeddings | as `fast` |
|
| 33 |
| `turbo` | `lowmem` with W4A4 output heads | KL 0.016 / 0.0009 |
|
| 34 |
|
| 35 |
-
Requirements: Linux x86_64, Python 3.12, PyTorch 2.10.0 + CUDA 12.8; OrbitQuant 0.9.
|
| 36 |
|
| 37 |
## Measurements
|
| 38 |
|
| 39 |
-
RTX 4090, identical input and seed
|
| 40 |
|
| 41 |
-
|
|
| 42 |
-
|---|---:|---:|---:|---:|---:|
|
| 43 |
-
| `off` | 168.9 s |
|
| 44 |
-
| `exact` | 168.9 s |
|
| 45 |
-
| `fast` | 140.7 s |
|
| 46 |
-
| `lowmem` | 152.2 s |
|
| 47 |
-
| `turbo` | 178.2 s |
|
| 48 |
|
| 49 |
-
`exact`
|
| 50 |
|
| 51 |
## Files
|
| 52 |
|
|
|
|
| 24 |
python src/run.py --profile fast --fuse-rms-quant --kv-cache-dtype int8 --mode fast --output out
|
| 25 |
```
|
| 26 |
|
| 27 |
+
One request per process is the default. A long-lived process serving many requests adds `--warm-compile` (compiled VAE decoder: 0.2 s instead of 0.7 s per request after a one-time 2 s cache load).
|
| 28 |
+
|
| 29 |
| `--mode` | Changes | Logits |
|
| 30 |
|---|---|---|
|
| 31 |
| `off` | reference decode path | reference |
|
|
|
|
| 34 |
| `lowmem` | + INT8 token embeddings | as `fast` |
|
| 35 |
| `turbo` | `lowmem` with W4A4 output heads | KL 0.016 / 0.0009 |
|
| 36 |
|
| 37 |
+
Requirements: Linux x86_64, Python 3.12, PyTorch 2.10.0 + CUDA 12.8; the OrbitQuant 0.9.8 wheel is bundled in `src/wheels`, native kernels 1.0.5 under `src/runtime/kernels`. Kernels: SM89 (validated on RTX 4090) and SM120 (built, not run). The first request compiles the VAE decoder (~25 s, cached afterwards).
|
| 38 |
|
| 39 |
## Measurements
|
| 40 |
|
| 41 |
+
RTX 4090 (48-core host), identical input and seed. "Warm" is the second request of a long-lived process with `--warm-compile`; "process" is one `src/run.py` call from interpreter start to the written files (imports and model load included). Peaks are for the whole process.
|
| 42 |
|
| 43 |
+
| Mode | Audio | Warm request | audio s / wall s | Whole process | NVML peak | torch peak |
|
| 44 |
+
|---|---:|---:|---:|---:|---:|---:|
|
| 45 |
+
| `off` | 168.9 s | 17.94 s | 9.42 | — | 5.39 GiB | 4.31 GiB |
|
| 46 |
+
| `exact` | 168.9 s | 15.85 s | 10.66 | 21.6 s | 5.33 GiB | 4.05 GiB |
|
| 47 |
+
| `fast` | 140.7 s | 12.26 s | 11.47 | 18.0 s | 4.64 GiB | 3.37 GiB |
|
| 48 |
+
| `lowmem` | 152.2 s | 13.49 s | 11.28 | 19.3 s | 4.47 GiB | 3.37 GiB |
|
| 49 |
+
| `turbo` | 178.2 s | 15.57 s | 11.44 | 21.7 s | 4.71 GiB | 3.56 GiB |
|
| 50 |
|
| 51 |
+
`exact` keeps the `off` tokens and logits bit for bit; its waveform differs from `off` only at FP16 rounding in the VAE. INT8/W4A4 heads change sampling at the same seed, so compare those modes by audio seconds per wall second. Of a whole process, about 4 s are Python and library imports plus model load; the rest is generation. Wall time also depends on the host CPU: on a slow 4-core host the same `exact` request took 17.7 s and `off` 28.3 s. Raw data: `evaluation/profiles/`.
|
| 52 |
|
| 53 |
## Files
|
| 54 |
|
SHA256SUMS
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
060985741d20e70613b4c189c7de106cabd3fb2109fbbfb6d705c9d619417dd0 LICENSE
|
| 2 |
-
|
| 3 |
14d3fd9f6fee86b4260b69b0979735b99ffec8c6b4db254567047a4215cd9af3 THIRD_PARTY_NOTICES.md
|
| 4 |
e479ab4f807b93d4c945ae9a14c5269dafb34751ee059d29f9883a5e812f2e56 config.json
|
| 5 |
5e375ad03c84d8118f3b37e56a3d01cf137b5aa1526b5725d91b17857839616d evaluation/ar-fused-fixed.json
|
|
@@ -234,15 +234,15 @@ da90c2dc55ecca5ffd2f7b15d1968180d5744a5b31ade45a7d221041096bdebe src/kernel-sou
|
|
| 234 |
ac54a84ce2bb7a0742a1cd46df398aeb23d83ebde9ed8c7cc0f7d914b59837dd src/kernel-source/yue2-fast/torch-ext/torch_binding.h
|
| 235 |
a368cb32d65829479d15c19451fe3199cf7a2d6510891a9596aa08fc4ed746dd src/kernel-source/yue2-fast/torch-ext/yue2_qkv_fused/__init__.py
|
| 236 |
62d2bd758c8fff1a0d64584a699343076fca9a7880db5d8e19263bc30ae79d7c src/modeling_yue2.py
|
| 237 |
-
|
| 238 |
-
|
| 239 |
-
|
| 240 |
3cb8837d8c883e0bc1dcbfe62481d24a4ededb35acd50b188ae6b1ebb27a19f8 src/runtime/OPTIMIZATIONS.md
|
| 241 |
-
|
| 242 |
026e36da4a42cf2095b75ebacb250cf8e20f5011763b2f0e44247afabdb9d112 src/runtime/fuse_packed.py
|
| 243 |
-
|
| 244 |
c1df0cc119cdfe0471754469458fa83fc718c6f42feca0b79a0d07cb2c068ce5 src/runtime/kernels/base/orbitquant_packed_matmul/_ops.py
|
| 245 |
-
|
| 246 |
a368cb32d65829479d15c19451fe3199cf7a2d6510891a9596aa08fc4ed746dd src/runtime/kernels/fast/torch210-cxx11-cu128-x86_64-linux/__init__.py
|
| 247 |
7352681d880f505a211531aa4b7028830a6f6889f2520353958748c0d3b2d7de src/runtime/kernels/fast/torch210-cxx11-cu128-x86_64-linux/_ops.py
|
| 248 |
b0a4d667dc1e0b1701f0bf9371254c293e30e974892e5681341980c9fe4d00ef src/runtime/kernels/fast/torch210-cxx11-cu128-x86_64-linux/_yue2_qkv_fused_cuda_46ceaaf_dirty.abi3.so
|
|
@@ -277,19 +277,19 @@ ce8fb82c3996d4c93eca1bf3a9405ef77da65c46a45623edc959fcf83bdb778d src/runtime/yu
|
|
| 277 |
3a25793081ed10325225e4e7bd9303243998abecd3a1a5b7a15139d664b70604 src/runtime/yue2/cli.py
|
| 278 |
cb54aab0abaef5c35c87400dce1a65e3614d6d1b8ebdcf5e2b4604943ce99f92 src/runtime/yue2/combined_sampling.py
|
| 279 |
617ee6914d534a234f1ff5d5f4685391eac5dc9f6be667429949eabf1e541517 src/runtime/yue2/compact_sampling.py
|
| 280 |
-
|
| 281 |
e033afd3a06f02a824c3a78b3bdce991d4f8e09a62d0078d715f3a2a1a5827e5 src/runtime/yue2/device_ring_sampling.py
|
| 282 |
e1722959ad0b0bf27eeaaf019cb9f109212a2450f03dd145866b1a0239151f38 src/runtime/yue2/fast.py
|
| 283 |
0f99252c4424bfa43500442649f38aede2be9a8f759c80060e8d82637cf3cd05 src/runtime/yue2/int8_kv.py
|
| 284 |
e6523f113132df6c04878103bca1a8b3de519fefcbc0117e97455f0dd49d1fba src/runtime/yue2/modeling_vae.py
|
| 285 |
62d2bd758c8fff1a0d64584a699343076fca9a7880db5d8e19263bc30ae79d7c src/runtime/yue2/modeling_yue2.py
|
| 286 |
-
|
| 287 |
-
|
| 288 |
8de1fa588607dd0525aee37b652f44b17350689ad9e6ddcba8aaa8aaccf61b24 src/runtime/yue2/progress.py
|
| 289 |
e00d32bf22baac46c7d4dafaeecd4f3d5a4753d651b3dc8f4164b9f3774980ad src/runtime/yue2/protocol.py
|
| 290 |
91e54fcd38a638727ebc05e240d2997097c541eec0cc18693fe429f702f38e78 src/runtime/yue2/quantization.py
|
| 291 |
9f5972207b1abd1930a80087ef611e3b324ff5c79bcdde20bfa5baf011af579d src/runtime/yue2/sampling.py
|
| 292 |
-
|
| 293 |
fc88387f96e5db2020a0c3f5374b82726206026d83159b68b83e4136bb741acc src/runtime/yue2/tokenization_yue2.py
|
| 294 |
0122ab1d9ec3556655db7ab435fdc9fce8373f9354a36e5bafac6637337f3159 src/runtime/yue2_orbit.py
|
| 295 |
7f612c79441b14529c9a7d0f35cb22cc871e3a9b33d24f9a19352cb3e2ff9bdb src/wheels/orbitquant-0.9.7-py3-none-any.whl
|
|
@@ -421,7 +421,7 @@ cfea3b98788df66437cfc7aa314411f5d98c951543d044d66168722fe742921d src/runtime/ke
|
|
| 421 |
0c560f96b857c188c4a8297ff27d1299619957c3459a5e433e132329fbfdf066 src/runtime/kernels/fast-sm120/torch210-cxx11-cu128-x86_64-linux/yue2_qkv_fused/__init__.py
|
| 422 |
8d84fd51f6a2e0cadb8540e50473c76c7d49c146ec7a7893283717f7f397c2ef src/runtime/kernels/fused/torch210-cxx11-cu128-x86_64-linux/__init__.py
|
| 423 |
ba7afaa7f20b76d639172201f3f6e373b292616cd39e9d6869b092454b44ab97 src/runtime/kernels/fused/torch210-cxx11-cu128-x86_64-linux/_ops.py
|
| 424 |
-
|
| 425 |
cb9393248c594e1098b0255cc48e920ddfb34c1556161260867e6772251104d0 src/runtime/kernels/fused/torch210-cxx11-cu128-x86_64-linux/metadata.json
|
| 426 |
0c560f96b857c188c4a8297ff27d1299619957c3459a5e433e132329fbfdf066 src/runtime/kernels/fused/torch210-cxx11-cu128-x86_64-linux/yue2_qkv_fused/__init__.py
|
| 427 |
8d84fd51f6a2e0cadb8540e50473c76c7d49c146ec7a7893283717f7f397c2ef src/runtime/kernels/fused-sm120/torch210-cxx11-cu128-x86_64-linux/__init__.py
|
|
@@ -458,83 +458,105 @@ b0c6b168be534d1b0be139b894bc0b0271b76b7952f6da5d75b4e821fbe42be4 src/runtime/ke
|
|
| 458 |
5c648e010c6192939f8a5829d20f346f9115d7db1cec40e165cce40aef60b9b8 evaluation/kl/kl-kv_parts-32-off0.json
|
| 459 |
f21c00c083e5fd2aa3f28394e3b2b2664b2c80a872dbb969924b06aeb85916dc evaluation/kl/kl-lookahead-off0.json
|
| 460 |
2f65e7918dd8ea20e6971422334c1093297c64a90c9629fdb9df16776c763ce3 evaluation/kl/kl-w4_heads-off0.json
|
| 461 |
-
|
| 462 |
-
|
| 463 |
-
|
| 464 |
-
|
| 465 |
-
8bea84adedb0f8f11df1046c96178364f5a8d13e3e3db3d5b8b8804beb612419 evaluation/profiles/exact/run-00/plan_manifest.json
|
| 466 |
f78032e194094dda24a200bb1204bd5f36b2807035bea32f48e098a919728782 evaluation/profiles/exact/run-00/request.json
|
| 467 |
-
|
| 468 |
027b9b0b33d4c6f837675c864eacac2954c420a18244eb55c6707fbf32b7748d evaluation/profiles/exact/run-00/score.abc
|
| 469 |
-
|
| 470 |
-
|
| 471 |
-
|
| 472 |
-
|
| 473 |
f78032e194094dda24a200bb1204bd5f36b2807035bea32f48e098a919728782 evaluation/profiles/exact/run-01/request.json
|
| 474 |
-
|
| 475 |
027b9b0b33d4c6f837675c864eacac2954c420a18244eb55c6707fbf32b7748d evaluation/profiles/exact/run-01/score.abc
|
| 476 |
-
|
| 477 |
-
|
| 478 |
-
|
| 479 |
-
|
| 480 |
-
|
| 481 |
f78032e194094dda24a200bb1204bd5f36b2807035bea32f48e098a919728782 evaluation/profiles/fast/run-00/request.json
|
| 482 |
-
|
| 483 |
8d7ed2e1687e35072e41ff1b64b852068ff81d496bb8142d83019b53b66498ab evaluation/profiles/fast/run-00/score.abc
|
| 484 |
-
|
| 485 |
-
|
| 486 |
-
|
| 487 |
-
|
| 488 |
f78032e194094dda24a200bb1204bd5f36b2807035bea32f48e098a919728782 evaluation/profiles/fast/run-01/request.json
|
| 489 |
-
|
| 490 |
8d7ed2e1687e35072e41ff1b64b852068ff81d496bb8142d83019b53b66498ab evaluation/profiles/fast/run-01/score.abc
|
| 491 |
-
|
| 492 |
-
|
| 493 |
-
|
| 494 |
-
|
| 495 |
-
|
| 496 |
99f362b30e461a1d6cf7b3aebf08057b8224fbac5b7a160b2db3a82bf89750d0 evaluation/profiles/fast-cfg1.5/run-00/request.json
|
| 497 |
-
|
| 498 |
8d7ed2e1687e35072e41ff1b64b852068ff81d496bb8142d83019b53b66498ab evaluation/profiles/fast-cfg1.5/run-00/score.abc
|
| 499 |
-
|
| 500 |
-
|
| 501 |
-
|
| 502 |
-
|
| 503 |
99f362b30e461a1d6cf7b3aebf08057b8224fbac5b7a160b2db3a82bf89750d0 evaluation/profiles/fast-cfg1.5/run-01/request.json
|
| 504 |
-
|
| 505 |
8d7ed2e1687e35072e41ff1b64b852068ff81d496bb8142d83019b53b66498ab evaluation/profiles/fast-cfg1.5/run-01/score.abc
|
| 506 |
-
|
| 507 |
-
|
| 508 |
-
|
| 509 |
-
|
| 510 |
-
|
| 511 |
f78032e194094dda24a200bb1204bd5f36b2807035bea32f48e098a919728782 evaluation/profiles/lowmem/run-00/request.json
|
| 512 |
-
|
| 513 |
ff4ea2b1fab4dc0e10a8c3347b8a9106151d38845493e5a3e11b847a53606e13 evaluation/profiles/lowmem/run-00/score.abc
|
| 514 |
-
|
| 515 |
-
|
| 516 |
-
|
| 517 |
-
|
| 518 |
f78032e194094dda24a200bb1204bd5f36b2807035bea32f48e098a919728782 evaluation/profiles/lowmem/run-01/request.json
|
| 519 |
-
|
| 520 |
ff4ea2b1fab4dc0e10a8c3347b8a9106151d38845493e5a3e11b847a53606e13 evaluation/profiles/lowmem/run-01/score.abc
|
| 521 |
-
|
| 522 |
-
|
| 523 |
-
|
| 524 |
-
|
| 525 |
-
|
| 526 |
-
|
| 527 |
f78032e194094dda24a200bb1204bd5f36b2807035bea32f48e098a919728782 evaluation/profiles/turbo/run-00/request.json
|
| 528 |
-
|
| 529 |
8be471a9f3ff34b4c4aa3737caf3012af641c722a04f96bc0b9d2a1a779933ea evaluation/profiles/turbo/run-00/score.abc
|
| 530 |
-
|
| 531 |
-
|
| 532 |
-
|
| 533 |
-
|
| 534 |
f78032e194094dda24a200bb1204bd5f36b2807035bea32f48e098a919728782 evaluation/profiles/turbo/run-01/request.json
|
| 535 |
-
|
| 536 |
8be471a9f3ff34b4c4aa3737caf3012af641c722a04f96bc0b9d2a1a779933ea evaluation/profiles/turbo/run-01/score.abc
|
| 537 |
-
|
| 538 |
7e397a366abe73220b0ca693572d0a2e687e31a118b79656bd0ccb4431f24461 src/__pycache__/run.cpython-314.pyc
|
| 539 |
-
|
| 540 |
379eb27253ab160a28a0e5909a234ed8e11fe86e1407b6eba5f7e034e2904eae src/runtime/__pycache__/fast_ops.cpython-314.pyc
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
060985741d20e70613b4c189c7de106cabd3fb2109fbbfb6d705c9d619417dd0 LICENSE
|
| 2 |
+
a8d87558ebab01600493bb0b93bf9ae256c359b922c56ca0c090fdfc6e678d17 README.md
|
| 3 |
14d3fd9f6fee86b4260b69b0979735b99ffec8c6b4db254567047a4215cd9af3 THIRD_PARTY_NOTICES.md
|
| 4 |
e479ab4f807b93d4c945ae9a14c5269dafb34751ee059d29f9883a5e812f2e56 config.json
|
| 5 |
5e375ad03c84d8118f3b37e56a3d01cf137b5aa1526b5725d91b17857839616d evaluation/ar-fused-fixed.json
|
|
|
|
| 234 |
ac54a84ce2bb7a0742a1cd46df398aeb23d83ebde9ed8c7cc0f7d914b59837dd src/kernel-source/yue2-fast/torch-ext/torch_binding.h
|
| 235 |
a368cb32d65829479d15c19451fe3199cf7a2d6510891a9596aa08fc4ed746dd src/kernel-source/yue2-fast/torch-ext/yue2_qkv_fused/__init__.py
|
| 236 |
62d2bd758c8fff1a0d64584a699343076fca9a7880db5d8e19263bc30ae79d7c src/modeling_yue2.py
|
| 237 |
+
88fe8712e74438f710ba427dcb04ea437b99f1eb94a53472eb4228152e14e8c1 src/requirements.txt
|
| 238 |
+
afa3f7a9fe89467c0080ed50be56b40778fd34bc4076576a3041fe9191f4cd02 src/run.py
|
| 239 |
+
6b40a8eb2be03a574454b7d33aeb0c1cda235932a3358929dd529cbe12c160a7 src/runtime-lock.json
|
| 240 |
3cb8837d8c883e0bc1dcbfe62481d24a4ededb35acd50b188ae6b1ebb27a19f8 src/runtime/OPTIMIZATIONS.md
|
| 241 |
+
d930a9142b2705a65e64c48947b3e3b7cc87b8df5c7cab75270a5a34b2b165da src/runtime/fast_ops.py
|
| 242 |
026e36da4a42cf2095b75ebacb250cf8e20f5011763b2f0e44247afabdb9d112 src/runtime/fuse_packed.py
|
| 243 |
+
234232325c8a2573d74002bd5c16c7e0f33105a8a5ad4f88525ca583e363f430 src/runtime/kernels/base/orbitquant_packed_matmul/__init__.py
|
| 244 |
c1df0cc119cdfe0471754469458fa83fc718c6f42feca0b79a0d07cb2c068ce5 src/runtime/kernels/base/orbitquant_packed_matmul/_ops.py
|
| 245 |
+
118d64db97c3200ef785e83b8d703ebc3b480a799bb17a6affb104d1b4faa28d src/runtime/kernels/base/orbitquant_packed_matmul/_orbitquant_packed_matmul_cuda_rel_cu128_t210.abi3.so
|
| 246 |
a368cb32d65829479d15c19451fe3199cf7a2d6510891a9596aa08fc4ed746dd src/runtime/kernels/fast/torch210-cxx11-cu128-x86_64-linux/__init__.py
|
| 247 |
7352681d880f505a211531aa4b7028830a6f6889f2520353958748c0d3b2d7de src/runtime/kernels/fast/torch210-cxx11-cu128-x86_64-linux/_ops.py
|
| 248 |
b0a4d667dc1e0b1701f0bf9371254c293e30e974892e5681341980c9fe4d00ef src/runtime/kernels/fast/torch210-cxx11-cu128-x86_64-linux/_yue2_qkv_fused_cuda_46ceaaf_dirty.abi3.so
|
|
|
|
| 277 |
3a25793081ed10325225e4e7bd9303243998abecd3a1a5b7a15139d664b70604 src/runtime/yue2/cli.py
|
| 278 |
cb54aab0abaef5c35c87400dce1a65e3614d6d1b8ebdcf5e2b4604943ce99f92 src/runtime/yue2/combined_sampling.py
|
| 279 |
617ee6914d534a234f1ff5d5f4685391eac5dc9f6be667429949eabf1e541517 src/runtime/yue2/compact_sampling.py
|
| 280 |
+
22ce26e78e184e29a7d68d2c1f43aaf483182d56f91de3a64733c02718430106 src/runtime/yue2/cuda_graph.py
|
| 281 |
e033afd3a06f02a824c3a78b3bdce991d4f8e09a62d0078d715f3a2a1a5827e5 src/runtime/yue2/device_ring_sampling.py
|
| 282 |
e1722959ad0b0bf27eeaaf019cb9f109212a2450f03dd145866b1a0239151f38 src/runtime/yue2/fast.py
|
| 283 |
0f99252c4424bfa43500442649f38aede2be9a8f759c80060e8d82637cf3cd05 src/runtime/yue2/int8_kv.py
|
| 284 |
e6523f113132df6c04878103bca1a8b3de519fefcbc0117e97455f0dd49d1fba src/runtime/yue2/modeling_vae.py
|
| 285 |
62d2bd758c8fff1a0d64584a699343076fca9a7880db5d8e19263bc30ae79d7c src/runtime/yue2/modeling_yue2.py
|
| 286 |
+
eaf5e30605ab2732c6b1b729031afb84ec3922651c95737154b339d024b49030 src/runtime/yue2/nar.py
|
| 287 |
+
d5da3319d7fdc9822a26b5640b689aa15c847d84fcd4fb27e8cef54379d39eae src/runtime/yue2/pipeline.py
|
| 288 |
8de1fa588607dd0525aee37b652f44b17350689ad9e6ddcba8aaa8aaccf61b24 src/runtime/yue2/progress.py
|
| 289 |
e00d32bf22baac46c7d4dafaeecd4f3d5a4753d651b3dc8f4164b9f3774980ad src/runtime/yue2/protocol.py
|
| 290 |
91e54fcd38a638727ebc05e240d2997097c541eec0cc18693fe429f702f38e78 src/runtime/yue2/quantization.py
|
| 291 |
9f5972207b1abd1930a80087ef611e3b324ff5c79bcdde20bfa5baf011af579d src/runtime/yue2/sampling.py
|
| 292 |
+
c64554d9af073d2fc1e12b155e637c0890f9db13e5c5a76295296dfe3df360a1 src/runtime/yue2/storage.py
|
| 293 |
fc88387f96e5db2020a0c3f5374b82726206026d83159b68b83e4136bb741acc src/runtime/yue2/tokenization_yue2.py
|
| 294 |
0122ab1d9ec3556655db7ab435fdc9fce8373f9354a36e5bafac6637337f3159 src/runtime/yue2_orbit.py
|
| 295 |
7f612c79441b14529c9a7d0f35cb22cc871e3a9b33d24f9a19352cb3e2ff9bdb src/wheels/orbitquant-0.9.7-py3-none-any.whl
|
|
|
|
| 421 |
0c560f96b857c188c4a8297ff27d1299619957c3459a5e433e132329fbfdf066 src/runtime/kernels/fast-sm120/torch210-cxx11-cu128-x86_64-linux/yue2_qkv_fused/__init__.py
|
| 422 |
8d84fd51f6a2e0cadb8540e50473c76c7d49c146ec7a7893283717f7f397c2ef src/runtime/kernels/fused/torch210-cxx11-cu128-x86_64-linux/__init__.py
|
| 423 |
ba7afaa7f20b76d639172201f3f6e373b292616cd39e9d6869b092454b44ab97 src/runtime/kernels/fused/torch210-cxx11-cu128-x86_64-linux/_ops.py
|
| 424 |
+
c7f2a62dc5ff53a2a11dd25f762f5af741ec13c56af580c6030713a8d72e7cb7 src/runtime/kernels/fused/torch210-cxx11-cu128-x86_64-linux/_yue2_qkv_fused_cuda_fg11a.abi3.so
|
| 425 |
cb9393248c594e1098b0255cc48e920ddfb34c1556161260867e6772251104d0 src/runtime/kernels/fused/torch210-cxx11-cu128-x86_64-linux/metadata.json
|
| 426 |
0c560f96b857c188c4a8297ff27d1299619957c3459a5e433e132329fbfdf066 src/runtime/kernels/fused/torch210-cxx11-cu128-x86_64-linux/yue2_qkv_fused/__init__.py
|
| 427 |
8d84fd51f6a2e0cadb8540e50473c76c7d49c146ec7a7893283717f7f397c2ef src/runtime/kernels/fused-sm120/torch210-cxx11-cu128-x86_64-linux/__init__.py
|
|
|
|
| 458 |
5c648e010c6192939f8a5829d20f346f9115d7db1cec40e165cce40aef60b9b8 evaluation/kl/kl-kv_parts-32-off0.json
|
| 459 |
f21c00c083e5fd2aa3f28394e3b2b2664b2c80a872dbb969924b06aeb85916dc evaluation/kl/kl-lookahead-off0.json
|
| 460 |
2f65e7918dd8ea20e6971422334c1093297c64a90c9629fdb9df16776c763ce3 evaluation/kl/kl-w4_heads-off0.json
|
| 461 |
+
9759afacf903510fff557ab5bb8b459b1dca120b1d7ae1e6aaf8f3a7995a6008 evaluation/profiles/exact/run-00/config.json
|
| 462 |
+
d6e8d447f7b7042bf78a000a95a98dd04e844123f17c9a323869bfc343f091f7 evaluation/profiles/exact/run-00/metrics.json
|
| 463 |
+
6e00a465ef2f20ebd78c1f88580e02c481bc1f4e028e3877299acfbd48adcffe evaluation/profiles/exact/run-00/plan.json
|
| 464 |
+
a49450598666b62d5c3fe0e84b23786fb4de7299a70d21fdeaebef74f48480ef evaluation/profiles/exact/run-00/plan_manifest.json
|
|
|
|
| 465 |
f78032e194094dda24a200bb1204bd5f36b2807035bea32f48e098a919728782 evaluation/profiles/exact/run-00/request.json
|
| 466 |
+
98a8ae26b66ae88ae0a48dc1af58fa50f3e2a7e7d118a1c5ef073e5ad8025f28 evaluation/profiles/exact/run-00/result.json
|
| 467 |
027b9b0b33d4c6f837675c864eacac2954c420a18244eb55c6707fbf32b7748d evaluation/profiles/exact/run-00/score.abc
|
| 468 |
+
9759afacf903510fff557ab5bb8b459b1dca120b1d7ae1e6aaf8f3a7995a6008 evaluation/profiles/exact/run-01/config.json
|
| 469 |
+
8a3eb8d3e407ee6da17b83f4df752f54b9a541ea5d7ea043f5dcce3fac7096ed evaluation/profiles/exact/run-01/metrics.json
|
| 470 |
+
8b926548d742c8ceac8733ef5e1d56de6bdcd76bc0df62989869cf431a274af8 evaluation/profiles/exact/run-01/plan.json
|
| 471 |
+
a738569a1d7393ea4d9c30fda391c7271d0c085d41b01b4910182397a725d355 evaluation/profiles/exact/run-01/plan_manifest.json
|
| 472 |
f78032e194094dda24a200bb1204bd5f36b2807035bea32f48e098a919728782 evaluation/profiles/exact/run-01/request.json
|
| 473 |
+
93e4a2c460fafcf247db4ddccc412987a44762d5a8df92c7b07c5c5d43ef25ae evaluation/profiles/exact/run-01/result.json
|
| 474 |
027b9b0b33d4c6f837675c864eacac2954c420a18244eb55c6707fbf32b7748d evaluation/profiles/exact/run-01/score.abc
|
| 475 |
+
1603f701ef4514c81d8b86b8bda6ac4d122e7c495df67cd28e528144fae1956f evaluation/profiles/exact/summary.json
|
| 476 |
+
1ca79e079b519059f4656324481ca68ddd6138f12dd333ba544811f339cb9c40 evaluation/profiles/fast/run-00/config.json
|
| 477 |
+
c519d4a104695435f9a29ada0018e469fb8625fbbaab87a40d79bd0391503f60 evaluation/profiles/fast/run-00/metrics.json
|
| 478 |
+
1979add7ff86e1abccc806a796464b646f2418bdeaa3367c907974dfde143234 evaluation/profiles/fast/run-00/plan.json
|
| 479 |
+
4c2599cdbf93d309218efb019039189bc67984ccb9360d6827ec88a58008e512 evaluation/profiles/fast/run-00/plan_manifest.json
|
| 480 |
f78032e194094dda24a200bb1204bd5f36b2807035bea32f48e098a919728782 evaluation/profiles/fast/run-00/request.json
|
| 481 |
+
efa0e3151c02d857cbe157dce1efd94ec90bed7762efacc718ca40f2ae52f31e evaluation/profiles/fast/run-00/result.json
|
| 482 |
8d7ed2e1687e35072e41ff1b64b852068ff81d496bb8142d83019b53b66498ab evaluation/profiles/fast/run-00/score.abc
|
| 483 |
+
1ca79e079b519059f4656324481ca68ddd6138f12dd333ba544811f339cb9c40 evaluation/profiles/fast/run-01/config.json
|
| 484 |
+
b90398b1550e2ee9724eb88f02131079d8020ab3b91ca4540decba2dfd32b3e7 evaluation/profiles/fast/run-01/metrics.json
|
| 485 |
+
5094ca63eae20eaec46f49a19e87a1c4f2668d5a8a7c799aacf408055659a9c1 evaluation/profiles/fast/run-01/plan.json
|
| 486 |
+
b17db0a80a096b20a98aac9ffba004946f3677da2fb5b84efb3a8b75e838eed2 evaluation/profiles/fast/run-01/plan_manifest.json
|
| 487 |
f78032e194094dda24a200bb1204bd5f36b2807035bea32f48e098a919728782 evaluation/profiles/fast/run-01/request.json
|
| 488 |
+
c368091bcdcaaf6964fba7c47eb5e3ab1158400cd9108f10f580a38cf4209e2c evaluation/profiles/fast/run-01/result.json
|
| 489 |
8d7ed2e1687e35072e41ff1b64b852068ff81d496bb8142d83019b53b66498ab evaluation/profiles/fast/run-01/score.abc
|
| 490 |
+
debb2bdfef443083ac944af7e5602328ef6a4541df947fb72e847bd8bd088b7f evaluation/profiles/fast/summary.json
|
| 491 |
+
4584bf7a27383453ed3fec41d52c3ca5410e0c4ce7893eb0c18be0a6413f6af1 evaluation/profiles/fast-cfg1.5/run-00/config.json
|
| 492 |
+
71445e1fecb3408580640a6e4b95fb5b40fe2757e1ab2232a1250f4197a25e6e evaluation/profiles/fast-cfg1.5/run-00/metrics.json
|
| 493 |
+
f22a94c8a3249590e6bb573c0f05fe1dbd6e6aa2eab6ee2faddb57c5180c86f1 evaluation/profiles/fast-cfg1.5/run-00/plan.json
|
| 494 |
+
ac1fab9a8b5e0377938251e4f97231c7b4c092fd3de0abf176b75272252c398d evaluation/profiles/fast-cfg1.5/run-00/plan_manifest.json
|
| 495 |
99f362b30e461a1d6cf7b3aebf08057b8224fbac5b7a160b2db3a82bf89750d0 evaluation/profiles/fast-cfg1.5/run-00/request.json
|
| 496 |
+
6e6eca998cc6e5ecde12c5535e7b8430363e73248b8ed61c11cb767d5d7a22d0 evaluation/profiles/fast-cfg1.5/run-00/result.json
|
| 497 |
8d7ed2e1687e35072e41ff1b64b852068ff81d496bb8142d83019b53b66498ab evaluation/profiles/fast-cfg1.5/run-00/score.abc
|
| 498 |
+
4584bf7a27383453ed3fec41d52c3ca5410e0c4ce7893eb0c18be0a6413f6af1 evaluation/profiles/fast-cfg1.5/run-01/config.json
|
| 499 |
+
bfb421153bd25912c108ab49f3c48c8d122eefe2abc74a6be08770d469f88ed6 evaluation/profiles/fast-cfg1.5/run-01/metrics.json
|
| 500 |
+
500905e85275ec425ab6f4e717b6a74e56e9584ca12fea060d9f9136e7b9f474 evaluation/profiles/fast-cfg1.5/run-01/plan.json
|
| 501 |
+
c48c7a04343d2998767c80f020bf5cf0d1ca767d0586d8a0dc5ebfc1d226681d evaluation/profiles/fast-cfg1.5/run-01/plan_manifest.json
|
| 502 |
99f362b30e461a1d6cf7b3aebf08057b8224fbac5b7a160b2db3a82bf89750d0 evaluation/profiles/fast-cfg1.5/run-01/request.json
|
| 503 |
+
bec7c051af1703bf83e3efb1b15166270ec8f386b69b71281f9f9cf9c865578c evaluation/profiles/fast-cfg1.5/run-01/result.json
|
| 504 |
8d7ed2e1687e35072e41ff1b64b852068ff81d496bb8142d83019b53b66498ab evaluation/profiles/fast-cfg1.5/run-01/score.abc
|
| 505 |
+
64a432c4cf280e0c93cdf05eb64d4bfb97fe67d284b2f504a9b5e89ff7f4be01 evaluation/profiles/fast-cfg1.5/summary.json
|
| 506 |
+
1ca79e079b519059f4656324481ca68ddd6138f12dd333ba544811f339cb9c40 evaluation/profiles/lowmem/run-00/config.json
|
| 507 |
+
f17d2fce1847fb799fba7c973c8d07caccf620ee98be85658d2994a828c4c32b evaluation/profiles/lowmem/run-00/metrics.json
|
| 508 |
+
b8b2dfaa50594f8fd427345295a13e1c2cd8acef36533d8b61c66f875480328d evaluation/profiles/lowmem/run-00/plan.json
|
| 509 |
+
e8a900437b475ba87e9f5443420b49d3c586511b1062367efb426340be60e381 evaluation/profiles/lowmem/run-00/plan_manifest.json
|
| 510 |
f78032e194094dda24a200bb1204bd5f36b2807035bea32f48e098a919728782 evaluation/profiles/lowmem/run-00/request.json
|
| 511 |
+
8aae3a7041a064638c2dab1a434d7aae2019bf813fd85fc3ec66e38a4c25989b evaluation/profiles/lowmem/run-00/result.json
|
| 512 |
ff4ea2b1fab4dc0e10a8c3347b8a9106151d38845493e5a3e11b847a53606e13 evaluation/profiles/lowmem/run-00/score.abc
|
| 513 |
+
1ca79e079b519059f4656324481ca68ddd6138f12dd333ba544811f339cb9c40 evaluation/profiles/lowmem/run-01/config.json
|
| 514 |
+
028c1a663fee0a77f2f397835fecc5311489baaea5177c2baf94f425513cb4d9 evaluation/profiles/lowmem/run-01/metrics.json
|
| 515 |
+
69335439395268b5c586e26f8ea5caaff53490cfba3b69186bd16abe0e752542 evaluation/profiles/lowmem/run-01/plan.json
|
| 516 |
+
5d1712e55cee8affd69a54d391d2edc20ba768790f28804baf9fd6541d8a8ed6 evaluation/profiles/lowmem/run-01/plan_manifest.json
|
| 517 |
f78032e194094dda24a200bb1204bd5f36b2807035bea32f48e098a919728782 evaluation/profiles/lowmem/run-01/request.json
|
| 518 |
+
170ec86f673c51c65a986b9fd04b1ec93cc8d20ae1a1c6b207ea45f7647cf132 evaluation/profiles/lowmem/run-01/result.json
|
| 519 |
ff4ea2b1fab4dc0e10a8c3347b8a9106151d38845493e5a3e11b847a53606e13 evaluation/profiles/lowmem/run-01/score.abc
|
| 520 |
+
cf8cf83d387f1766c35733eb5956c8c21b6c0e819ccc75904b729d6123967278 evaluation/profiles/lowmem/summary.json
|
| 521 |
+
c1bea33622e3ebf6b508018704d88d4f50393d3e18ec0f08826b0d56c5b88eb3 evaluation/profiles/spectral.json
|
| 522 |
+
1ca79e079b519059f4656324481ca68ddd6138f12dd333ba544811f339cb9c40 evaluation/profiles/turbo/run-00/config.json
|
| 523 |
+
516c78a1959052b8bee0a5aaf9fb0999d5ce6c25d0b4480a848b148ff2694ca5 evaluation/profiles/turbo/run-00/metrics.json
|
| 524 |
+
75510a92e2eb84186617d61427591b80a8c486031a12576f2cb710488e098e81 evaluation/profiles/turbo/run-00/plan.json
|
| 525 |
+
086d2ded9ccf23c9add15a1e7707a1b75860386dd8605dd34f9492a81bf73f8a evaluation/profiles/turbo/run-00/plan_manifest.json
|
| 526 |
f78032e194094dda24a200bb1204bd5f36b2807035bea32f48e098a919728782 evaluation/profiles/turbo/run-00/request.json
|
| 527 |
+
7172fa8d1c9fd617f6fcf5db1dd4129c32d378fd07b4b835846aae233342224d evaluation/profiles/turbo/run-00/result.json
|
| 528 |
8be471a9f3ff34b4c4aa3737caf3012af641c722a04f96bc0b9d2a1a779933ea evaluation/profiles/turbo/run-00/score.abc
|
| 529 |
+
1ca79e079b519059f4656324481ca68ddd6138f12dd333ba544811f339cb9c40 evaluation/profiles/turbo/run-01/config.json
|
| 530 |
+
e8c5eeece4d13780bfb9eb93710ead7bffeb05f2f563df79eb98bcfcd123cc3f evaluation/profiles/turbo/run-01/metrics.json
|
| 531 |
+
f4d9e1e4da302b785581e0e2b0c8dc6bce550ef57f1c1951250e83d8bf45822e evaluation/profiles/turbo/run-01/plan.json
|
| 532 |
+
8c41af3c2ecf704f72efab805be752300dab7548ecf01aa0d48c45843d0e2e6f evaluation/profiles/turbo/run-01/plan_manifest.json
|
| 533 |
f78032e194094dda24a200bb1204bd5f36b2807035bea32f48e098a919728782 evaluation/profiles/turbo/run-01/request.json
|
| 534 |
+
eaa6002907c20da3fa5347e6fa3336aa025346bf67fa5fbb81d652d56b718cf9 evaluation/profiles/turbo/run-01/result.json
|
| 535 |
8be471a9f3ff34b4c4aa3737caf3012af641c722a04f96bc0b9d2a1a779933ea evaluation/profiles/turbo/run-01/score.abc
|
| 536 |
+
1a47870860944489b04fdb252c207d397c74c53b4093477e39a3edf0d29fc558 evaluation/profiles/turbo/summary.json
|
| 537 |
7e397a366abe73220b0ca693572d0a2e687e31a118b79656bd0ccb4431f24461 src/__pycache__/run.cpython-314.pyc
|
| 538 |
+
ad39d33cff152638045ad71bbe64c55745ffd32c12ded98c40144e343f5759fb src/kernel-binaries.json
|
| 539 |
379eb27253ab160a28a0e5909a234ed8e11fe86e1407b6eba5f7e034e2904eae src/runtime/__pycache__/fast_ops.cpython-314.pyc
|
| 540 |
+
684ac369a74ded3017bde80328718b5186bb9bd8f803a1311dba9325f5fae6fb evaluation/profiles/off/run-00/config.json
|
| 541 |
+
73c600f0e77ae1273c96ec4563d4f183bd922c15f44355d22f1286580f1cae68 evaluation/profiles/off/run-00/metrics.json
|
| 542 |
+
6183d80c77115dcb2a254594dcd9b4dd5c56b24fbfa1e7945cf0a2ad7e2f73a3 evaluation/profiles/off/run-00/plan.json
|
| 543 |
+
34cb31e6d049d3361c947e5c7ca93cd6e48c72b3c86c3a98feec63b056545bad evaluation/profiles/off/run-00/plan_manifest.json
|
| 544 |
+
f78032e194094dda24a200bb1204bd5f36b2807035bea32f48e098a919728782 evaluation/profiles/off/run-00/request.json
|
| 545 |
+
8786431384687852d56d128ae5aedec513d26f03b77490c3ccb2ad75a1bb5c6e evaluation/profiles/off/run-00/result.json
|
| 546 |
+
027b9b0b33d4c6f837675c864eacac2954c420a18244eb55c6707fbf32b7748d evaluation/profiles/off/run-00/score.abc
|
| 547 |
+
684ac369a74ded3017bde80328718b5186bb9bd8f803a1311dba9325f5fae6fb evaluation/profiles/off/run-01/config.json
|
| 548 |
+
10dcdf1c0a0db8bdb09cb0d24048187f5a2adcb727786b0f92e4940a64f5f428 evaluation/profiles/off/run-01/metrics.json
|
| 549 |
+
ca74419cf2ab4c68906e8a183d54c765c0ec4b137825d8be9355ad9a3c366ac4 evaluation/profiles/off/run-01/plan.json
|
| 550 |
+
e5040f6bb62669b4afd399a5adfc3073acffb7b648c9d96ee9a4990c3c8ac03d evaluation/profiles/off/run-01/plan_manifest.json
|
| 551 |
+
f78032e194094dda24a200bb1204bd5f36b2807035bea32f48e098a919728782 evaluation/profiles/off/run-01/request.json
|
| 552 |
+
f5d1d7ab06610dd8ff05b2a9aa831098f109f93f28191a1e0161da065ed2edb1 evaluation/profiles/off/run-01/result.json
|
| 553 |
+
027b9b0b33d4c6f837675c864eacac2954c420a18244eb55c6707fbf32b7748d evaluation/profiles/off/run-01/score.abc
|
| 554 |
+
312828e0d018b79da3fc6c910aae9c047bc0580ef4b32e37075e435cc759b6ec evaluation/profiles/off/summary.json
|
| 555 |
+
8cdee3ad737b54434d55d139a5e85db7ec396cc107df4dbfee03383205327e6e evaluation/profiles/single-process-cli.json
|
| 556 |
+
5ffd1a4fb3ab63ecf99d0c7ff0894b2d037c56d0b21ecad130ffd2cd325450bb evaluation/profiles/spectral-exact-variants.json
|
| 557 |
+
331c06a248811f778f9f5ed280101e86bc0dd580e80db85619893148ff885720 src/OPTIMIZATION_IDEAS.md
|
| 558 |
+
e69db09574c063e29b215d7e69feffceacaa1645dc2330e63d50da31c7b09c16 src/ROUND6_EXPERIMENTS.md
|
| 559 |
+
9cf29973e588fbe41f1fe09415f180f5c3b907882f79b4ebda5629fa6a0ff88d src/ROUND7_EXPERIMENTS.md
|
| 560 |
+
1300988fbdfaab67cc7ec2a340ee85771d57e887b5d9911691682019078ccaa5 src/ROUND8_EXPERIMENTS.md
|
| 561 |
+
aa451d1ee3d7f255f0390c9dd81ad90c116305cc828f028a0814b1b01f8fd8f3 src/ROUND9_EXPERIMENTS.md
|
| 562 |
+
98e0d1a1a6b6c8c56761fbbe2ec5b9b68b3d6b0a1d3f115f49e8953a8e15780b src/wheels/orbitquant-0.9.8-py3-none-any.whl
|
evaluation/profiles/baseline.json
DELETED
|
@@ -1,158 +0,0 @@
|
|
| 1 |
-
[
|
| 2 |
-
{
|
| 3 |
-
"tag": "baseline",
|
| 4 |
-
"native": "published",
|
| 5 |
-
"native_sha256": "6fb8e94e30e2c9982da5e3e56c04dabed946fe9dc8ad1785c11cc3cf200217b4",
|
| 6 |
-
"repeat": 0,
|
| 7 |
-
"wall_seconds": 57.41068919800455,
|
| 8 |
-
"audio_seconds": 168.91866666666667,
|
| 9 |
-
"audio_per_wall": 2.942285992841519,
|
| 10 |
-
"nvml_peak_gib": 5.31573486328125,
|
| 11 |
-
"torch_peak_gib": 4.048099517822266,
|
| 12 |
-
"timing": {
|
| 13 |
-
"abc": {
|
| 14 |
-
"seconds": 13.882036672992399,
|
| 15 |
-
"prefill_seconds": 8.619665990991052,
|
| 16 |
-
"ttft_seconds": 8.627110775996698,
|
| 17 |
-
"output_tokens": 1424,
|
| 18 |
-
"content_tokens": 1423,
|
| 19 |
-
"output_tps": 102.57860813538997,
|
| 20 |
-
"prefix_tokens": 423,
|
| 21 |
-
"cfg_branches": 1,
|
| 22 |
-
"sampling": "combined_cuda_graph",
|
| 23 |
-
"execution": "cuda_graph",
|
| 24 |
-
"attention": "int8",
|
| 25 |
-
"combined_graph_replays": 1423,
|
| 26 |
-
"device_history": true
|
| 27 |
-
},
|
| 28 |
-
"semantic": {
|
| 29 |
-
"seconds": 16.691209856973728,
|
| 30 |
-
"prefill_seconds": 6.7202503659937065,
|
| 31 |
-
"ttft_seconds": 6.720333366974955,
|
| 32 |
-
"output_tokens": 4224,
|
| 33 |
-
"content_tokens": 4223,
|
| 34 |
-
"output_tps": 253.06733521388068,
|
| 35 |
-
"prefix_tokens": 1848,
|
| 36 |
-
"cfg_branches": 1,
|
| 37 |
-
"sampling": "combined_cuda_graph",
|
| 38 |
-
"execution": "cuda_graph",
|
| 39 |
-
"attention": "int8",
|
| 40 |
-
"combined_graph_replays": 4223,
|
| 41 |
-
"device_history": true
|
| 42 |
-
},
|
| 43 |
-
"nar_seconds": 9.772851961984998,
|
| 44 |
-
"vae_seconds": 17.034749298007227,
|
| 45 |
-
"load": {
|
| 46 |
-
"resolve_and_integrity_seconds": 9.673675911995815,
|
| 47 |
-
"mot_load_seconds": 1.798799140000483
|
| 48 |
-
},
|
| 49 |
-
"e2e_seconds": 57.40883600001689
|
| 50 |
-
},
|
| 51 |
-
"semantic_sha256": "36a231d034d554c12388e82b4cd4be2598079ae23f0e7fdfb14187e339adcb0c",
|
| 52 |
-
"waveform_sha256": "4580778f3e57663e34b77e329dd1bbf89ad647a6a250dbfbd8df6ffe8372f79f",
|
| 53 |
-
"gpu": "NVIDIA GeForce RTX 4090",
|
| 54 |
-
"runtime": {
|
| 55 |
-
"int8_attention": "ordered",
|
| 56 |
-
"kv_cache_dtype": "int8",
|
| 57 |
-
"captured_sampling": true,
|
| 58 |
-
"fuse_rms_quant": true,
|
| 59 |
-
"orbitquant_version": "0.9.7",
|
| 60 |
-
"profile": "fast",
|
| 61 |
-
"allocator_backend": "native",
|
| 62 |
-
"modules": 224,
|
| 63 |
-
"logical_projections": 392,
|
| 64 |
-
"fused": true,
|
| 65 |
-
"effective_modes": {
|
| 66 |
-
"native_packed_matmul": 224
|
| 67 |
-
},
|
| 68 |
-
"activation_backends": {
|
| 69 |
-
"native_cuda_int8_surrogate": 168,
|
| 70 |
-
"triton_cuda_packed_w4": 56
|
| 71 |
-
},
|
| 72 |
-
"decoded_weight_caches": 0,
|
| 73 |
-
"int8_weight_caches": 0,
|
| 74 |
-
"packed_weight_bytes": 1409286144
|
| 75 |
-
},
|
| 76 |
-
"cfg": 1.0,
|
| 77 |
-
"steps": 8,
|
| 78 |
-
"allocator": "expandable_segments:True"
|
| 79 |
-
},
|
| 80 |
-
{
|
| 81 |
-
"tag": "baseline",
|
| 82 |
-
"native": "published",
|
| 83 |
-
"native_sha256": "6fb8e94e30e2c9982da5e3e56c04dabed946fe9dc8ad1785c11cc3cf200217b4",
|
| 84 |
-
"repeat": 1,
|
| 85 |
-
"wall_seconds": 28.26992048000102,
|
| 86 |
-
"audio_seconds": 168.91866666666667,
|
| 87 |
-
"audio_per_wall": 5.975208412282757,
|
| 88 |
-
"nvml_peak_gib": 5.38409423828125,
|
| 89 |
-
"torch_peak_gib": 4.311957836151123,
|
| 90 |
-
"timing": {
|
| 91 |
-
"abc": {
|
| 92 |
-
"seconds": 4.3758966060122475,
|
| 93 |
-
"prefill_seconds": 0.1942817910166923,
|
| 94 |
-
"ttft_seconds": 0.19432357102050446,
|
| 95 |
-
"output_tokens": 1424,
|
| 96 |
-
"content_tokens": 1423,
|
| 97 |
-
"output_tps": 325.4190233936287,
|
| 98 |
-
"prefix_tokens": 423,
|
| 99 |
-
"cfg_branches": 1,
|
| 100 |
-
"sampling": "combined_cuda_graph",
|
| 101 |
-
"execution": "cuda_graph",
|
| 102 |
-
"attention": "int8",
|
| 103 |
-
"combined_graph_replays": 1423,
|
| 104 |
-
"device_history": true
|
| 105 |
-
},
|
| 106 |
-
"semantic": {
|
| 107 |
-
"seconds": 10.08559686600347,
|
| 108 |
-
"prefill_seconds": 0.14707892900332808,
|
| 109 |
-
"ttft_seconds": 0.1471185589907691,
|
| 110 |
-
"output_tokens": 4224,
|
| 111 |
-
"content_tokens": 4223,
|
| 112 |
-
"output_tps": 418.81507422116573,
|
| 113 |
-
"prefix_tokens": 1848,
|
| 114 |
-
"cfg_branches": 1,
|
| 115 |
-
"sampling": "combined_cuda_graph",
|
| 116 |
-
"execution": "cuda_graph",
|
| 117 |
-
"attention": "int8",
|
| 118 |
-
"combined_graph_replays": 4223,
|
| 119 |
-
"device_history": true
|
| 120 |
-
},
|
| 121 |
-
"nar_seconds": 3.8059511620085686,
|
| 122 |
-
"vae_seconds": 8.396466452017194,
|
| 123 |
-
"load": {
|
| 124 |
-
"resolve_and_integrity_seconds": 9.673675911995815,
|
| 125 |
-
"mot_load_seconds": 1.798799140000483
|
| 126 |
-
},
|
| 127 |
-
"e2e_seconds": 28.266109021002194
|
| 128 |
-
},
|
| 129 |
-
"semantic_sha256": "36a231d034d554c12388e82b4cd4be2598079ae23f0e7fdfb14187e339adcb0c",
|
| 130 |
-
"waveform_sha256": "4580778f3e57663e34b77e329dd1bbf89ad647a6a250dbfbd8df6ffe8372f79f",
|
| 131 |
-
"gpu": "NVIDIA GeForce RTX 4090",
|
| 132 |
-
"runtime": {
|
| 133 |
-
"int8_attention": "ordered",
|
| 134 |
-
"kv_cache_dtype": "int8",
|
| 135 |
-
"captured_sampling": true,
|
| 136 |
-
"fuse_rms_quant": true,
|
| 137 |
-
"orbitquant_version": "0.9.7",
|
| 138 |
-
"profile": "fast",
|
| 139 |
-
"allocator_backend": "native",
|
| 140 |
-
"modules": 224,
|
| 141 |
-
"logical_projections": 392,
|
| 142 |
-
"fused": true,
|
| 143 |
-
"effective_modes": {
|
| 144 |
-
"native_packed_matmul": 224
|
| 145 |
-
},
|
| 146 |
-
"activation_backends": {
|
| 147 |
-
"native_cuda_int8_surrogate": 168,
|
| 148 |
-
"triton_cuda_packed_w4": 56
|
| 149 |
-
},
|
| 150 |
-
"decoded_weight_caches": 0,
|
| 151 |
-
"int8_weight_caches": 0,
|
| 152 |
-
"packed_weight_bytes": 1409286144
|
| 153 |
-
},
|
| 154 |
-
"cfg": 1.0,
|
| 155 |
-
"steps": 8,
|
| 156 |
-
"allocator": "expandable_segments:True"
|
| 157 |
-
}
|
| 158 |
-
]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
evaluation/profiles/exact/run-00/config.json
CHANGED
|
@@ -34,12 +34,12 @@
|
|
| 34 |
"model_dtype": "bfloat16",
|
| 35 |
"vae_dtype": "float32",
|
| 36 |
"vae_decode": "halo_crop",
|
| 37 |
-
"vae_core_frames":
|
| 38 |
"vae_halo_frames": 16,
|
| 39 |
"device": "cuda:0",
|
| 40 |
"memory_budget_gib": 18.0,
|
| 41 |
"offload_ar": false,
|
| 42 |
-
"runtime_sha256": "
|
| 43 |
"decoder_release": "standard",
|
| 44 |
"validation_status": "unvalidated",
|
| 45 |
"runtime_profile": "fast",
|
|
@@ -47,7 +47,7 @@
|
|
| 47 |
"captured_sampling": true,
|
| 48 |
"kv_cache_dtype": "int8",
|
| 49 |
"int8_attention": "ordered",
|
| 50 |
-
"runtime_bundle_sha256": "
|
| 51 |
"vae_compute_dtype": "torch.float16",
|
| 52 |
"expert_staging": false,
|
| 53 |
"orbitquant": {
|
|
|
|
| 34 |
"model_dtype": "bfloat16",
|
| 35 |
"vae_dtype": "float32",
|
| 36 |
"vae_decode": "halo_crop",
|
| 37 |
+
"vae_core_frames": 256,
|
| 38 |
"vae_halo_frames": 16,
|
| 39 |
"device": "cuda:0",
|
| 40 |
"memory_budget_gib": 18.0,
|
| 41 |
"offload_ar": false,
|
| 42 |
+
"runtime_sha256": "c354c01c9b04ef1dad51891857040d20790473fac7cb787042b70f6f9f451643",
|
| 43 |
"decoder_release": "standard",
|
| 44 |
"validation_status": "unvalidated",
|
| 45 |
"runtime_profile": "fast",
|
|
|
|
| 47 |
"captured_sampling": true,
|
| 48 |
"kv_cache_dtype": "int8",
|
| 49 |
"int8_attention": "ordered",
|
| 50 |
+
"runtime_bundle_sha256": "51c8dd94465183f85c9759831fba4318bd89780458923f2cd3a3a5daa1e37952",
|
| 51 |
"vae_compute_dtype": "torch.float16",
|
| 52 |
"expert_staging": false,
|
| 53 |
"orbitquant": {
|
evaluation/profiles/exact/run-00/metrics.json
CHANGED
|
@@ -1,41 +1,42 @@
|
|
| 1 |
{
|
| 2 |
"bundle": "v11",
|
| 3 |
-
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default",
|
| 4 |
"flags": [
|
| 5 |
"abc_crop",
|
| 6 |
"lookahead",
|
| 7 |
"resident",
|
| 8 |
"vae_compile",
|
|
|
|
| 9 |
"vae_mode=default"
|
| 10 |
],
|
| 11 |
-
"native_sha256": "
|
| 12 |
"repeat": 0,
|
| 13 |
-
"wall_seconds":
|
| 14 |
-
"load_seconds":
|
| 15 |
"audio_seconds": 168.91866666666667,
|
| 16 |
-
"audio_per_wall":
|
| 17 |
-
"abc_s":
|
| 18 |
"abc_tokens": 1424,
|
| 19 |
-
"sem_s": 9.
|
| 20 |
"sem_tokens": 4224,
|
| 21 |
-
"abc_ms_per_tok":
|
| 22 |
-
"sem_ms_per_tok": 2.
|
| 23 |
-
"nar_s":
|
| 24 |
-
"vae_s":
|
| 25 |
-
"nvml_peak_gib": 5.
|
| 26 |
-
"torch_peak_gib":
|
| 27 |
"torch_reserved_peak_gib": 4.357421875,
|
| 28 |
"semantic_sha256": "36a231d034d554c12388e82b4cd4be2598079ae23f0e7fdfb14187e339adcb0c",
|
| 29 |
-
"waveform_sha256": "
|
| 30 |
"latent_sha256": "5c566f70f058de65c5f3e8daf3dab9846cb354ad72b26df9680d472f193b6c7d",
|
| 31 |
"timing": {
|
| 32 |
"abc": {
|
| 33 |
-
"seconds":
|
| 34 |
-
"prefill_seconds":
|
| 35 |
-
"ttft_seconds":
|
| 36 |
"output_tokens": 1424,
|
| 37 |
"content_tokens": 1423,
|
| 38 |
-
"output_tps":
|
| 39 |
"prefix_tokens": 423,
|
| 40 |
"cfg_branches": 1,
|
| 41 |
"sampling": "combined_cuda_graph",
|
|
@@ -46,16 +47,16 @@
|
|
| 46 |
"lookahead": true,
|
| 47 |
"unroll2": false,
|
| 48 |
"first_step_seconds": null,
|
| 49 |
-
"capture_seconds": 0.
|
| 50 |
-
"close_seconds": 0.
|
| 51 |
},
|
| 52 |
"semantic": {
|
| 53 |
-
"seconds": 9.
|
| 54 |
-
"prefill_seconds": 0.
|
| 55 |
-
"ttft_seconds": 0.
|
| 56 |
"output_tokens": 4224,
|
| 57 |
"content_tokens": 4223,
|
| 58 |
-
"output_tps":
|
| 59 |
"prefix_tokens": 1848,
|
| 60 |
"cfg_branches": 1,
|
| 61 |
"sampling": "combined_cuda_graph",
|
|
@@ -66,48 +67,88 @@
|
|
| 66 |
"lookahead": true,
|
| 67 |
"unroll2": false,
|
| 68 |
"first_step_seconds": null,
|
| 69 |
-
"capture_seconds": 0.
|
| 70 |
-
"close_seconds": 0.
|
| 71 |
},
|
| 72 |
-
"nar_seconds":
|
| 73 |
-
"vae_seconds":
|
| 74 |
"load": {
|
| 75 |
-
"resolve_and_integrity_seconds":
|
| 76 |
-
"mot_load_seconds":
|
| 77 |
},
|
| 78 |
"stages": {
|
| 79 |
-
"t0->plan_done":
|
| 80 |
-
"plan_done->semantic_done": 9.
|
| 81 |
-
"semantic_done->nar_done":
|
| 82 |
-
"nar_done->vae_done":
|
| 83 |
-
"plan_outside_timer": 0.
|
| 84 |
-
"semantic_outside_timer": 0.
|
| 85 |
-
"semantic_close_seconds": 0.
|
| 86 |
"vae": {
|
| 87 |
-
"load_s":
|
| 88 |
-
"decode_s":
|
| 89 |
-
"post_s":
|
| 90 |
"tiles": [
|
| 91 |
-
|
| 92 |
-
0.
|
| 93 |
-
0.
|
| 94 |
-
0.
|
| 95 |
-
0.
|
| 96 |
-
0.
|
| 97 |
-
0.
|
| 98 |
-
0.
|
| 99 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 100 |
],
|
| 101 |
"cudnn_benchmark": false,
|
| 102 |
"cudnn_deterministic": true,
|
| 103 |
-
"core_frames":
|
| 104 |
"input_device": "cpu",
|
| 105 |
"input_contiguous": false
|
| 106 |
},
|
| 107 |
"nar_prefill_source": "prefill",
|
| 108 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 109 |
},
|
| 110 |
-
"e2e_seconds":
|
| 111 |
},
|
| 112 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 113 |
"cfg": 1.0,
|
|
@@ -117,7 +158,7 @@
|
|
| 117 |
"kv_cache_dtype": "int8",
|
| 118 |
"captured_sampling": true,
|
| 119 |
"fuse_rms_quant": true,
|
| 120 |
-
"orbitquant_version": "0.9.
|
| 121 |
"profile": "fast",
|
| 122 |
"allocator_backend": "native",
|
| 123 |
"modules": 224,
|
|
|
|
| 1 |
{
|
| 2 |
"bundle": "v11",
|
| 3 |
+
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,vae_core=256",
|
| 4 |
"flags": [
|
| 5 |
"abc_crop",
|
| 6 |
"lookahead",
|
| 7 |
"resident",
|
| 8 |
"vae_compile",
|
| 9 |
+
"vae_core=256",
|
| 10 |
"vae_mode=default"
|
| 11 |
],
|
| 12 |
+
"native_sha256": "118d64db97c3200ef785e83b8d703ebc3b480a799bb17a6affb104d1b4faa28d",
|
| 13 |
"repeat": 0,
|
| 14 |
+
"wall_seconds": 18.547393172048032,
|
| 15 |
+
"load_seconds": 2.2395440358668566,
|
| 16 |
"audio_seconds": 168.91866666666667,
|
| 17 |
+
"audio_per_wall": 9.107407445335047,
|
| 18 |
+
"abc_s": 4.272322495467961,
|
| 19 |
"abc_tokens": 1424,
|
| 20 |
+
"sem_s": 9.2354048024863,
|
| 21 |
"sem_tokens": 4224,
|
| 22 |
+
"abc_ms_per_tok": 3.0002264715364895,
|
| 23 |
+
"sem_ms_per_tok": 2.186412121800734,
|
| 24 |
+
"nar_s": 2.741346231661737,
|
| 25 |
+
"vae_s": 2.292730061337352,
|
| 26 |
+
"nvml_peak_gib": 5.326416015625,
|
| 27 |
+
"torch_peak_gib": 3.11099910736084,
|
| 28 |
"torch_reserved_peak_gib": 4.357421875,
|
| 29 |
"semantic_sha256": "36a231d034d554c12388e82b4cd4be2598079ae23f0e7fdfb14187e339adcb0c",
|
| 30 |
+
"waveform_sha256": "e0aeec8e4144ed157e18c54b2e8141a8ba6051343a3ca178cd85bfbc133749c9",
|
| 31 |
"latent_sha256": "5c566f70f058de65c5f3e8daf3dab9846cb354ad72b26df9680d472f193b6c7d",
|
| 32 |
"timing": {
|
| 33 |
"abc": {
|
| 34 |
+
"seconds": 4.272322495467961,
|
| 35 |
+
"prefill_seconds": 0.6354847745969892,
|
| 36 |
+
"ttft_seconds": 0.6366398222744465,
|
| 37 |
"output_tokens": 1424,
|
| 38 |
"content_tokens": 1423,
|
| 39 |
+
"output_tps": 333.3081717287414,
|
| 40 |
"prefix_tokens": 423,
|
| 41 |
"cfg_branches": 1,
|
| 42 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 47 |
"lookahead": true,
|
| 48 |
"unroll2": false,
|
| 49 |
"first_step_seconds": null,
|
| 50 |
+
"capture_seconds": 0.0413797153159976,
|
| 51 |
+
"close_seconds": 0.00031330157071352005
|
| 52 |
},
|
| 53 |
"semantic": {
|
| 54 |
+
"seconds": 9.2354048024863,
|
| 55 |
+
"prefill_seconds": 0.05777517054229975,
|
| 56 |
+
"ttft_seconds": 0.05779500026255846,
|
| 57 |
"output_tokens": 4224,
|
| 58 |
"content_tokens": 4223,
|
| 59 |
+
"output_tps": 457.37031460308486,
|
| 60 |
"prefix_tokens": 1848,
|
| 61 |
"cfg_branches": 1,
|
| 62 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 67 |
"lookahead": true,
|
| 68 |
"unroll2": false,
|
| 69 |
"first_step_seconds": null,
|
| 70 |
+
"capture_seconds": 0.032724469900131226,
|
| 71 |
+
"close_seconds": 0.00040849484503269196
|
| 72 |
},
|
| 73 |
+
"nar_seconds": 2.741346231661737,
|
| 74 |
+
"vae_seconds": 2.292730061337352,
|
| 75 |
"load": {
|
| 76 |
+
"resolve_and_integrity_seconds": 0.20317224133759737,
|
| 77 |
+
"mot_load_seconds": 0.6788896378129721
|
| 78 |
},
|
| 79 |
"stages": {
|
| 80 |
+
"t0->plan_done": 4.274795697070658,
|
| 81 |
+
"plan_done->semantic_done": 9.236809186637402,
|
| 82 |
+
"semantic_done->nar_done": 2.7416386613622308,
|
| 83 |
+
"nar_done->vae_done": 2.2930279998108745,
|
| 84 |
+
"plan_outside_timer": 0.0024732016026973724,
|
| 85 |
+
"semantic_outside_timer": 0.0014043841511011124,
|
| 86 |
+
"semantic_close_seconds": 0.00040849484503269196,
|
| 87 |
"vae": {
|
| 88 |
+
"load_s": 0.2927389871329069,
|
| 89 |
+
"decode_s": 1.9990807436406612,
|
| 90 |
+
"post_s": 1.1868774890899658e-05,
|
| 91 |
"tiles": [
|
| 92 |
+
1.7568,
|
| 93 |
+
0.0142,
|
| 94 |
+
0.0142,
|
| 95 |
+
0.0141,
|
| 96 |
+
0.0141,
|
| 97 |
+
0.014,
|
| 98 |
+
0.014,
|
| 99 |
+
0.014,
|
| 100 |
+
0.0136,
|
| 101 |
+
0.0135,
|
| 102 |
+
0.0136,
|
| 103 |
+
0.0135,
|
| 104 |
+
0.0139,
|
| 105 |
+
0.0138,
|
| 106 |
+
0.0137,
|
| 107 |
+
0.0137,
|
| 108 |
+
0.0342
|
| 109 |
],
|
| 110 |
"cudnn_benchmark": false,
|
| 111 |
"cudnn_deterministic": true,
|
| 112 |
+
"core_frames": 256,
|
| 113 |
"input_device": "cpu",
|
| 114 |
"input_contiguous": false
|
| 115 |
},
|
| 116 |
"nar_prefill_source": "prefill",
|
| 117 |
+
"abc_close_seconds": 0.00031330157071352005,
|
| 118 |
+
"memory": {
|
| 119 |
+
"t0": {
|
| 120 |
+
"peak_gib": 2.831,
|
| 121 |
+
"alloc_gib": 2.831,
|
| 122 |
+
"reserved_gib": 3.068,
|
| 123 |
+
"nvml_used_gib": 3.928
|
| 124 |
+
},
|
| 125 |
+
"plan_done": {
|
| 126 |
+
"peak_gib": 3.108,
|
| 127 |
+
"alloc_gib": 2.855,
|
| 128 |
+
"reserved_gib": 3.371,
|
| 129 |
+
"nvml_used_gib": 4.328
|
| 130 |
+
},
|
| 131 |
+
"semantic_done": {
|
| 132 |
+
"peak_gib": 3.566,
|
| 133 |
+
"alloc_gib": 2.863,
|
| 134 |
+
"reserved_gib": 3.719,
|
| 135 |
+
"nvml_used_gib": 4.678
|
| 136 |
+
},
|
| 137 |
+
"nar_done": {
|
| 138 |
+
"peak_gib": 4.048,
|
| 139 |
+
"alloc_gib": 2.864,
|
| 140 |
+
"reserved_gib": 4.357,
|
| 141 |
+
"nvml_used_gib": 5.317
|
| 142 |
+
},
|
| 143 |
+
"vae_done": {
|
| 144 |
+
"peak_gib": 3.512,
|
| 145 |
+
"alloc_gib": 3.111,
|
| 146 |
+
"reserved_gib": 4.357,
|
| 147 |
+
"nvml_used_gib": 5.326
|
| 148 |
+
}
|
| 149 |
+
}
|
| 150 |
},
|
| 151 |
+
"e2e_seconds": 18.54658360313624
|
| 152 |
},
|
| 153 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 154 |
"cfg": 1.0,
|
|
|
|
| 158 |
"kv_cache_dtype": "int8",
|
| 159 |
"captured_sampling": true,
|
| 160 |
"fuse_rms_quant": true,
|
| 161 |
+
"orbitquant_version": "0.9.8",
|
| 162 |
"profile": "fast",
|
| 163 |
"allocator_backend": "native",
|
| 164 |
"modules": 224,
|
evaluation/profiles/exact/run-00/plan.json
CHANGED
|
@@ -9,12 +9,12 @@
|
|
| 9 |
"id": "song"
|
| 10 |
},
|
| 11 |
"timing": {
|
| 12 |
-
"seconds":
|
| 13 |
-
"prefill_seconds":
|
| 14 |
-
"ttft_seconds":
|
| 15 |
"output_tokens": 1424,
|
| 16 |
"content_tokens": 1423,
|
| 17 |
-
"output_tps":
|
| 18 |
"prefix_tokens": 423,
|
| 19 |
"cfg_branches": 1,
|
| 20 |
"sampling": "combined_cuda_graph",
|
|
@@ -25,8 +25,8 @@
|
|
| 25 |
"lookahead": true,
|
| 26 |
"unroll2": false,
|
| 27 |
"first_step_seconds": null,
|
| 28 |
-
"capture_seconds": 0.
|
| 29 |
-
"close_seconds": 0.
|
| 30 |
},
|
| 31 |
"truncated": false,
|
| 32 |
"prefix": [
|
|
|
|
| 9 |
"id": "song"
|
| 10 |
},
|
| 11 |
"timing": {
|
| 12 |
+
"seconds": 4.272322495467961,
|
| 13 |
+
"prefill_seconds": 0.6354847745969892,
|
| 14 |
+
"ttft_seconds": 0.6366398222744465,
|
| 15 |
"output_tokens": 1424,
|
| 16 |
"content_tokens": 1423,
|
| 17 |
+
"output_tps": 333.3081717287414,
|
| 18 |
"prefix_tokens": 423,
|
| 19 |
"cfg_branches": 1,
|
| 20 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 25 |
"lookahead": true,
|
| 26 |
"unroll2": false,
|
| 27 |
"first_step_seconds": null,
|
| 28 |
+
"capture_seconds": 0.0413797153159976,
|
| 29 |
+
"close_seconds": 0.00031330157071352005
|
| 30 |
},
|
| 31 |
"truncated": false,
|
| 32 |
"prefix": [
|
evaluation/profiles/exact/run-00/plan_manifest.json
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
{
|
| 2 |
-
"plan.json": "
|
| 3 |
"abc_tokens.npy": "509147d79fcf8fbe4db07e170c5adcd99986df1ef0ae60607e034e5d43d66992",
|
| 4 |
"prefix.npy": "fe7c7aed0a2cd948b22ce2df55442ca1012152a1c577745afcff296d51b6d594",
|
| 5 |
"score.abc": "027b9b0b33d4c6f837675c864eacac2954c420a18244eb55c6707fbf32b7748d"
|
|
|
|
| 1 |
{
|
| 2 |
+
"plan.json": "6e00a465ef2f20ebd78c1f88580e02c481bc1f4e028e3877299acfbd48adcffe",
|
| 3 |
"abc_tokens.npy": "509147d79fcf8fbe4db07e170c5adcd99986df1ef0ae60607e034e5d43d66992",
|
| 4 |
"prefix.npy": "fe7c7aed0a2cd948b22ce2df55442ca1012152a1c577745afcff296d51b6d594",
|
| 5 |
"score.abc": "027b9b0b33d4c6f837675c864eacac2954c420a18244eb55c6707fbf32b7748d"
|
evaluation/profiles/exact/run-00/result.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"status": "complete",
|
| 3 |
-
"identity": "
|
| 4 |
"truncated": {
|
| 5 |
"abc": false,
|
| 6 |
"semantic": false
|
|
@@ -29,12 +29,12 @@
|
|
| 29 |
},
|
| 30 |
"timing": {
|
| 31 |
"abc": {
|
| 32 |
-
"seconds":
|
| 33 |
-
"prefill_seconds":
|
| 34 |
-
"ttft_seconds":
|
| 35 |
"output_tokens": 1424,
|
| 36 |
"content_tokens": 1423,
|
| 37 |
-
"output_tps":
|
| 38 |
"prefix_tokens": 423,
|
| 39 |
"cfg_branches": 1,
|
| 40 |
"sampling": "combined_cuda_graph",
|
|
@@ -45,16 +45,16 @@
|
|
| 45 |
"lookahead": true,
|
| 46 |
"unroll2": false,
|
| 47 |
"first_step_seconds": null,
|
| 48 |
-
"capture_seconds": 0.
|
| 49 |
-
"close_seconds": 0.
|
| 50 |
},
|
| 51 |
"semantic": {
|
| 52 |
-
"seconds": 9.
|
| 53 |
-
"prefill_seconds": 0.
|
| 54 |
-
"ttft_seconds": 0.
|
| 55 |
"output_tokens": 4224,
|
| 56 |
"content_tokens": 4223,
|
| 57 |
-
"output_tps":
|
| 58 |
"prefix_tokens": 1848,
|
| 59 |
"cfg_branches": 1,
|
| 60 |
"sampling": "combined_cuda_graph",
|
|
@@ -65,48 +65,88 @@
|
|
| 65 |
"lookahead": true,
|
| 66 |
"unroll2": false,
|
| 67 |
"first_step_seconds": null,
|
| 68 |
-
"capture_seconds": 0.
|
| 69 |
-
"close_seconds": 0.
|
| 70 |
},
|
| 71 |
-
"nar_seconds":
|
| 72 |
-
"vae_seconds":
|
| 73 |
"load": {
|
| 74 |
-
"resolve_and_integrity_seconds":
|
| 75 |
-
"mot_load_seconds":
|
| 76 |
},
|
| 77 |
"stages": {
|
| 78 |
-
"t0->plan_done":
|
| 79 |
-
"plan_done->semantic_done": 9.
|
| 80 |
-
"semantic_done->nar_done":
|
| 81 |
-
"nar_done->vae_done":
|
| 82 |
-
"plan_outside_timer": 0.
|
| 83 |
-
"semantic_outside_timer": 0.
|
| 84 |
-
"semantic_close_seconds": 0.
|
| 85 |
"vae": {
|
| 86 |
-
"load_s":
|
| 87 |
-
"decode_s":
|
| 88 |
-
"post_s":
|
| 89 |
"tiles": [
|
| 90 |
-
|
| 91 |
-
0.
|
| 92 |
-
0.
|
| 93 |
-
0.
|
| 94 |
-
0.
|
| 95 |
-
0.
|
| 96 |
-
0.
|
| 97 |
-
0.
|
| 98 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 99 |
],
|
| 100 |
"cudnn_benchmark": false,
|
| 101 |
"cudnn_deterministic": true,
|
| 102 |
-
"core_frames":
|
| 103 |
"input_device": "cpu",
|
| 104 |
"input_contiguous": false
|
| 105 |
},
|
| 106 |
"nar_prefill_source": "prefill",
|
| 107 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 108 |
},
|
| 109 |
-
"e2e_seconds":
|
| 110 |
},
|
| 111 |
"artifacts": {
|
| 112 |
"abc_tokens.npy": {
|
|
@@ -114,11 +154,11 @@
|
|
| 114 |
"bytes": 5820
|
| 115 |
},
|
| 116 |
"audio.flac": {
|
| 117 |
-
"sha256": "
|
| 118 |
-
"bytes":
|
| 119 |
},
|
| 120 |
"config.json": {
|
| 121 |
-
"sha256": "
|
| 122 |
"bytes": 2476
|
| 123 |
},
|
| 124 |
"latent.npy": {
|
|
@@ -126,11 +166,11 @@
|
|
| 126 |
"bytes": 1081216
|
| 127 |
},
|
| 128 |
"plan.json": {
|
| 129 |
-
"sha256": "
|
| 130 |
-
"bytes":
|
| 131 |
},
|
| 132 |
"plan_manifest.json": {
|
| 133 |
-
"sha256": "
|
| 134 |
"bytes": 341
|
| 135 |
},
|
| 136 |
"prefix.npy": {
|
|
|
|
| 1 |
{
|
| 2 |
"status": "complete",
|
| 3 |
+
"identity": "660a195b70043d34681193d19d9addcc3039a7f82df474b415b607ea08266e52",
|
| 4 |
"truncated": {
|
| 5 |
"abc": false,
|
| 6 |
"semantic": false
|
|
|
|
| 29 |
},
|
| 30 |
"timing": {
|
| 31 |
"abc": {
|
| 32 |
+
"seconds": 4.272322495467961,
|
| 33 |
+
"prefill_seconds": 0.6354847745969892,
|
| 34 |
+
"ttft_seconds": 0.6366398222744465,
|
| 35 |
"output_tokens": 1424,
|
| 36 |
"content_tokens": 1423,
|
| 37 |
+
"output_tps": 333.3081717287414,
|
| 38 |
"prefix_tokens": 423,
|
| 39 |
"cfg_branches": 1,
|
| 40 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 45 |
"lookahead": true,
|
| 46 |
"unroll2": false,
|
| 47 |
"first_step_seconds": null,
|
| 48 |
+
"capture_seconds": 0.0413797153159976,
|
| 49 |
+
"close_seconds": 0.00031330157071352005
|
| 50 |
},
|
| 51 |
"semantic": {
|
| 52 |
+
"seconds": 9.2354048024863,
|
| 53 |
+
"prefill_seconds": 0.05777517054229975,
|
| 54 |
+
"ttft_seconds": 0.05779500026255846,
|
| 55 |
"output_tokens": 4224,
|
| 56 |
"content_tokens": 4223,
|
| 57 |
+
"output_tps": 457.37031460308486,
|
| 58 |
"prefix_tokens": 1848,
|
| 59 |
"cfg_branches": 1,
|
| 60 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 65 |
"lookahead": true,
|
| 66 |
"unroll2": false,
|
| 67 |
"first_step_seconds": null,
|
| 68 |
+
"capture_seconds": 0.032724469900131226,
|
| 69 |
+
"close_seconds": 0.00040849484503269196
|
| 70 |
},
|
| 71 |
+
"nar_seconds": 2.741346231661737,
|
| 72 |
+
"vae_seconds": 2.292730061337352,
|
| 73 |
"load": {
|
| 74 |
+
"resolve_and_integrity_seconds": 0.20317224133759737,
|
| 75 |
+
"mot_load_seconds": 0.6788896378129721
|
| 76 |
},
|
| 77 |
"stages": {
|
| 78 |
+
"t0->plan_done": 4.274795697070658,
|
| 79 |
+
"plan_done->semantic_done": 9.236809186637402,
|
| 80 |
+
"semantic_done->nar_done": 2.7416386613622308,
|
| 81 |
+
"nar_done->vae_done": 2.2930279998108745,
|
| 82 |
+
"plan_outside_timer": 0.0024732016026973724,
|
| 83 |
+
"semantic_outside_timer": 0.0014043841511011124,
|
| 84 |
+
"semantic_close_seconds": 0.00040849484503269196,
|
| 85 |
"vae": {
|
| 86 |
+
"load_s": 0.2927389871329069,
|
| 87 |
+
"decode_s": 1.9990807436406612,
|
| 88 |
+
"post_s": 1.1868774890899658e-05,
|
| 89 |
"tiles": [
|
| 90 |
+
1.7568,
|
| 91 |
+
0.0142,
|
| 92 |
+
0.0142,
|
| 93 |
+
0.0141,
|
| 94 |
+
0.0141,
|
| 95 |
+
0.014,
|
| 96 |
+
0.014,
|
| 97 |
+
0.014,
|
| 98 |
+
0.0136,
|
| 99 |
+
0.0135,
|
| 100 |
+
0.0136,
|
| 101 |
+
0.0135,
|
| 102 |
+
0.0139,
|
| 103 |
+
0.0138,
|
| 104 |
+
0.0137,
|
| 105 |
+
0.0137,
|
| 106 |
+
0.0342
|
| 107 |
],
|
| 108 |
"cudnn_benchmark": false,
|
| 109 |
"cudnn_deterministic": true,
|
| 110 |
+
"core_frames": 256,
|
| 111 |
"input_device": "cpu",
|
| 112 |
"input_contiguous": false
|
| 113 |
},
|
| 114 |
"nar_prefill_source": "prefill",
|
| 115 |
+
"abc_close_seconds": 0.00031330157071352005,
|
| 116 |
+
"memory": {
|
| 117 |
+
"t0": {
|
| 118 |
+
"peak_gib": 2.831,
|
| 119 |
+
"alloc_gib": 2.831,
|
| 120 |
+
"reserved_gib": 3.068,
|
| 121 |
+
"nvml_used_gib": 3.928
|
| 122 |
+
},
|
| 123 |
+
"plan_done": {
|
| 124 |
+
"peak_gib": 3.108,
|
| 125 |
+
"alloc_gib": 2.855,
|
| 126 |
+
"reserved_gib": 3.371,
|
| 127 |
+
"nvml_used_gib": 4.328
|
| 128 |
+
},
|
| 129 |
+
"semantic_done": {
|
| 130 |
+
"peak_gib": 3.566,
|
| 131 |
+
"alloc_gib": 2.863,
|
| 132 |
+
"reserved_gib": 3.719,
|
| 133 |
+
"nvml_used_gib": 4.678
|
| 134 |
+
},
|
| 135 |
+
"nar_done": {
|
| 136 |
+
"peak_gib": 4.048,
|
| 137 |
+
"alloc_gib": 2.864,
|
| 138 |
+
"reserved_gib": 4.357,
|
| 139 |
+
"nvml_used_gib": 5.317
|
| 140 |
+
},
|
| 141 |
+
"vae_done": {
|
| 142 |
+
"peak_gib": 3.512,
|
| 143 |
+
"alloc_gib": 3.111,
|
| 144 |
+
"reserved_gib": 4.357,
|
| 145 |
+
"nvml_used_gib": 5.326
|
| 146 |
+
}
|
| 147 |
+
}
|
| 148 |
},
|
| 149 |
+
"e2e_seconds": 18.54658360313624
|
| 150 |
},
|
| 151 |
"artifacts": {
|
| 152 |
"abc_tokens.npy": {
|
|
|
|
| 154 |
"bytes": 5820
|
| 155 |
},
|
| 156 |
"audio.flac": {
|
| 157 |
+
"sha256": "2c8ab1306a961caeecaa97f22ff855ea5c41810e87037454cd0b45f91f74d7fb",
|
| 158 |
+
"bytes": 34180094
|
| 159 |
},
|
| 160 |
"config.json": {
|
| 161 |
+
"sha256": "9759afacf903510fff557ab5bb8b459b1dca120b1d7ae1e6aaf8f3a7995a6008",
|
| 162 |
"bytes": 2476
|
| 163 |
},
|
| 164 |
"latent.npy": {
|
|
|
|
| 166 |
"bytes": 1081216
|
| 167 |
},
|
| 168 |
"plan.json": {
|
| 169 |
+
"sha256": "6e00a465ef2f20ebd78c1f88580e02c481bc1f4e028e3877299acfbd48adcffe",
|
| 170 |
+
"bytes": 33016
|
| 171 |
},
|
| 172 |
"plan_manifest.json": {
|
| 173 |
+
"sha256": "a49450598666b62d5c3fe0e84b23786fb4de7299a70d21fdeaebef74f48480ef",
|
| 174 |
"bytes": 341
|
| 175 |
},
|
| 176 |
"prefix.npy": {
|
evaluation/profiles/exact/run-01/config.json
CHANGED
|
@@ -34,12 +34,12 @@
|
|
| 34 |
"model_dtype": "bfloat16",
|
| 35 |
"vae_dtype": "float32",
|
| 36 |
"vae_decode": "halo_crop",
|
| 37 |
-
"vae_core_frames":
|
| 38 |
"vae_halo_frames": 16,
|
| 39 |
"device": "cuda:0",
|
| 40 |
"memory_budget_gib": 18.0,
|
| 41 |
"offload_ar": false,
|
| 42 |
-
"runtime_sha256": "
|
| 43 |
"decoder_release": "standard",
|
| 44 |
"validation_status": "unvalidated",
|
| 45 |
"runtime_profile": "fast",
|
|
@@ -47,7 +47,7 @@
|
|
| 47 |
"captured_sampling": true,
|
| 48 |
"kv_cache_dtype": "int8",
|
| 49 |
"int8_attention": "ordered",
|
| 50 |
-
"runtime_bundle_sha256": "
|
| 51 |
"vae_compute_dtype": "torch.float16",
|
| 52 |
"expert_staging": false,
|
| 53 |
"orbitquant": {
|
|
|
|
| 34 |
"model_dtype": "bfloat16",
|
| 35 |
"vae_dtype": "float32",
|
| 36 |
"vae_decode": "halo_crop",
|
| 37 |
+
"vae_core_frames": 256,
|
| 38 |
"vae_halo_frames": 16,
|
| 39 |
"device": "cuda:0",
|
| 40 |
"memory_budget_gib": 18.0,
|
| 41 |
"offload_ar": false,
|
| 42 |
+
"runtime_sha256": "c354c01c9b04ef1dad51891857040d20790473fac7cb787042b70f6f9f451643",
|
| 43 |
"decoder_release": "standard",
|
| 44 |
"validation_status": "unvalidated",
|
| 45 |
"runtime_profile": "fast",
|
|
|
|
| 47 |
"captured_sampling": true,
|
| 48 |
"kv_cache_dtype": "int8",
|
| 49 |
"int8_attention": "ordered",
|
| 50 |
+
"runtime_bundle_sha256": "51c8dd94465183f85c9759831fba4318bd89780458923f2cd3a3a5daa1e37952",
|
| 51 |
"vae_compute_dtype": "torch.float16",
|
| 52 |
"expert_staging": false,
|
| 53 |
"orbitquant": {
|
evaluation/profiles/exact/run-01/metrics.json
CHANGED
|
@@ -1,41 +1,42 @@
|
|
| 1 |
{
|
| 2 |
"bundle": "v11",
|
| 3 |
-
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default",
|
| 4 |
"flags": [
|
| 5 |
"abc_crop",
|
| 6 |
"lookahead",
|
| 7 |
"resident",
|
| 8 |
"vae_compile",
|
|
|
|
| 9 |
"vae_mode=default"
|
| 10 |
],
|
| 11 |
-
"native_sha256": "
|
| 12 |
"repeat": 1,
|
| 13 |
-
"wall_seconds":
|
| 14 |
-
"load_seconds":
|
| 15 |
"audio_seconds": 168.91866666666667,
|
| 16 |
-
"audio_per_wall":
|
| 17 |
-
"abc_s":
|
| 18 |
"abc_tokens": 1424,
|
| 19 |
-
"sem_s": 9.
|
| 20 |
"sem_tokens": 4224,
|
| 21 |
-
"abc_ms_per_tok": 2.
|
| 22 |
-
"sem_ms_per_tok": 2.
|
| 23 |
-
"nar_s":
|
| 24 |
-
"vae_s": 0.
|
| 25 |
-
"nvml_peak_gib": 5.
|
| 26 |
-
"torch_peak_gib":
|
| 27 |
"torch_reserved_peak_gib": 4.630859375,
|
| 28 |
"semantic_sha256": "36a231d034d554c12388e82b4cd4be2598079ae23f0e7fdfb14187e339adcb0c",
|
| 29 |
-
"waveform_sha256": "
|
| 30 |
"latent_sha256": "5c566f70f058de65c5f3e8daf3dab9846cb354ad72b26df9680d472f193b6c7d",
|
| 31 |
"timing": {
|
| 32 |
"abc": {
|
| 33 |
-
"seconds":
|
| 34 |
-
"prefill_seconds": 0.
|
| 35 |
-
"ttft_seconds": 0.
|
| 36 |
"output_tokens": 1424,
|
| 37 |
"content_tokens": 1423,
|
| 38 |
-
"output_tps":
|
| 39 |
"prefix_tokens": 423,
|
| 40 |
"cfg_branches": 1,
|
| 41 |
"sampling": "combined_cuda_graph",
|
|
@@ -46,16 +47,16 @@
|
|
| 46 |
"lookahead": true,
|
| 47 |
"unroll2": false,
|
| 48 |
"first_step_seconds": null,
|
| 49 |
-
"capture_seconds": 0.
|
| 50 |
-
"close_seconds": 0.
|
| 51 |
},
|
| 52 |
"semantic": {
|
| 53 |
-
"seconds": 9.
|
| 54 |
-
"prefill_seconds": 0.
|
| 55 |
-
"ttft_seconds": 0.
|
| 56 |
"output_tokens": 4224,
|
| 57 |
"content_tokens": 4223,
|
| 58 |
-
"output_tps":
|
| 59 |
"prefix_tokens": 1848,
|
| 60 |
"cfg_branches": 1,
|
| 61 |
"sampling": "combined_cuda_graph",
|
|
@@ -66,48 +67,88 @@
|
|
| 66 |
"lookahead": true,
|
| 67 |
"unroll2": false,
|
| 68 |
"first_step_seconds": null,
|
| 69 |
-
"capture_seconds": 0.
|
| 70 |
-
"close_seconds": 0.
|
| 71 |
},
|
| 72 |
-
"nar_seconds":
|
| 73 |
-
"vae_seconds": 0.
|
| 74 |
"load": {
|
| 75 |
-
"resolve_and_integrity_seconds":
|
| 76 |
-
"mot_load_seconds":
|
| 77 |
},
|
| 78 |
"stages": {
|
| 79 |
-
"t0->plan_done":
|
| 80 |
-
"plan_done->semantic_done": 9.
|
| 81 |
-
"semantic_done->nar_done":
|
| 82 |
-
"nar_done->vae_done": 0.
|
| 83 |
-
"plan_outside_timer": 0.
|
| 84 |
-
"semantic_outside_timer": 0.
|
| 85 |
-
"semantic_close_seconds": 0.
|
| 86 |
"vae": {
|
| 87 |
-
"load_s": 0.
|
| 88 |
-
"decode_s": 0.
|
| 89 |
-
"post_s":
|
| 90 |
"tiles": [
|
| 91 |
-
0.
|
| 92 |
-
0.
|
| 93 |
-
0.
|
| 94 |
-
0.
|
| 95 |
-
0.
|
| 96 |
-
0.
|
| 97 |
-
0.
|
| 98 |
-
0.
|
| 99 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 100 |
],
|
| 101 |
"cudnn_benchmark": false,
|
| 102 |
"cudnn_deterministic": true,
|
| 103 |
-
"core_frames":
|
| 104 |
"input_device": "cpu",
|
| 105 |
"input_contiguous": false
|
| 106 |
},
|
| 107 |
"nar_prefill_source": "prefill",
|
| 108 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 109 |
},
|
| 110 |
-
"e2e_seconds":
|
| 111 |
},
|
| 112 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 113 |
"cfg": 1.0,
|
|
@@ -117,7 +158,7 @@
|
|
| 117 |
"kv_cache_dtype": "int8",
|
| 118 |
"captured_sampling": true,
|
| 119 |
"fuse_rms_quant": true,
|
| 120 |
-
"orbitquant_version": "0.9.
|
| 121 |
"profile": "fast",
|
| 122 |
"allocator_backend": "native",
|
| 123 |
"modules": 224,
|
|
|
|
| 1 |
{
|
| 2 |
"bundle": "v11",
|
| 3 |
+
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,vae_core=256",
|
| 4 |
"flags": [
|
| 5 |
"abc_crop",
|
| 6 |
"lookahead",
|
| 7 |
"resident",
|
| 8 |
"vae_compile",
|
| 9 |
+
"vae_core=256",
|
| 10 |
"vae_mode=default"
|
| 11 |
],
|
| 12 |
+
"native_sha256": "118d64db97c3200ef785e83b8d703ebc3b480a799bb17a6affb104d1b4faa28d",
|
| 13 |
"repeat": 1,
|
| 14 |
+
"wall_seconds": 15.850000811740756,
|
| 15 |
+
"load_seconds": 2.2395440358668566,
|
| 16 |
"audio_seconds": 168.91866666666667,
|
| 17 |
+
"audio_per_wall": 10.657328581430834,
|
| 18 |
+
"abc_s": 3.6763406014069915,
|
| 19 |
"abc_tokens": 1424,
|
| 20 |
+
"sem_s": 9.223421677947044,
|
| 21 |
"sem_tokens": 4224,
|
| 22 |
+
"abc_ms_per_tok": 2.5816998605386177,
|
| 23 |
+
"sem_ms_per_tok": 2.1835752078473116,
|
| 24 |
+
"nar_s": 2.7084463676437736,
|
| 25 |
+
"vae_s": 0.23706442769616842,
|
| 26 |
+
"nvml_peak_gib": 5.599853515625,
|
| 27 |
+
"torch_peak_gib": 3.12686824798584,
|
| 28 |
"torch_reserved_peak_gib": 4.630859375,
|
| 29 |
"semantic_sha256": "36a231d034d554c12388e82b4cd4be2598079ae23f0e7fdfb14187e339adcb0c",
|
| 30 |
+
"waveform_sha256": "e0aeec8e4144ed157e18c54b2e8141a8ba6051343a3ca178cd85bfbc133749c9",
|
| 31 |
"latent_sha256": "5c566f70f058de65c5f3e8daf3dab9846cb354ad72b26df9680d472f193b6c7d",
|
| 32 |
"timing": {
|
| 33 |
"abc": {
|
| 34 |
+
"seconds": 3.6763406014069915,
|
| 35 |
+
"prefill_seconds": 0.04775886610150337,
|
| 36 |
+
"ttft_seconds": 0.04778050817549229,
|
| 37 |
"output_tokens": 1424,
|
| 38 |
"content_tokens": 1423,
|
| 39 |
+
"output_tps": 387.341695014606,
|
| 40 |
"prefix_tokens": 423,
|
| 41 |
"cfg_branches": 1,
|
| 42 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 47 |
"lookahead": true,
|
| 48 |
"unroll2": false,
|
| 49 |
"first_step_seconds": null,
|
| 50 |
+
"capture_seconds": 0.03480528108775616,
|
| 51 |
+
"close_seconds": 0.00036335643380880356
|
| 52 |
},
|
| 53 |
"semantic": {
|
| 54 |
+
"seconds": 9.223421677947044,
|
| 55 |
+
"prefill_seconds": 0.05041100736707449,
|
| 56 |
+
"ttft_seconds": 0.050430997274816036,
|
| 57 |
"output_tokens": 4224,
|
| 58 |
"content_tokens": 4223,
|
| 59 |
+
"output_tps": 457.964532847877,
|
| 60 |
"prefix_tokens": 1848,
|
| 61 |
"cfg_branches": 1,
|
| 62 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 67 |
"lookahead": true,
|
| 68 |
"unroll2": false,
|
| 69 |
"first_step_seconds": null,
|
| 70 |
+
"capture_seconds": 0.03198968339711428,
|
| 71 |
+
"close_seconds": 0.0004133721813559532
|
| 72 |
},
|
| 73 |
+
"nar_seconds": 2.7084463676437736,
|
| 74 |
+
"vae_seconds": 0.23706442769616842,
|
| 75 |
"load": {
|
| 76 |
+
"resolve_and_integrity_seconds": 0.20317224133759737,
|
| 77 |
+
"mot_load_seconds": 0.6788896378129721
|
| 78 |
},
|
| 79 |
"stages": {
|
| 80 |
+
"t0->plan_done": 3.6779328491538763,
|
| 81 |
+
"plan_done->semantic_done": 9.22484851628542,
|
| 82 |
+
"semantic_done->nar_done": 2.708730294369161,
|
| 83 |
+
"nar_done->vae_done": 0.2373927216976881,
|
| 84 |
+
"plan_outside_timer": 0.0015922477468848228,
|
| 85 |
+
"semantic_outside_timer": 0.0014268383383750916,
|
| 86 |
+
"semantic_close_seconds": 0.0004133721813559532,
|
| 87 |
"vae": {
|
| 88 |
+
"load_s": 0.0010568005964159966,
|
| 89 |
+
"decode_s": 0.2359218681231141,
|
| 90 |
+
"post_s": 1.5283003449440002e-05,
|
| 91 |
"tiles": [
|
| 92 |
+
0.0142,
|
| 93 |
+
0.0136,
|
| 94 |
+
0.0135,
|
| 95 |
+
0.0135,
|
| 96 |
+
0.0136,
|
| 97 |
+
0.0136,
|
| 98 |
+
0.0137,
|
| 99 |
+
0.0138,
|
| 100 |
+
0.0136,
|
| 101 |
+
0.0137,
|
| 102 |
+
0.0136,
|
| 103 |
+
0.0137,
|
| 104 |
+
0.0136,
|
| 105 |
+
0.0136,
|
| 106 |
+
0.0137,
|
| 107 |
+
0.0136,
|
| 108 |
+
0.0174
|
| 109 |
],
|
| 110 |
"cudnn_benchmark": false,
|
| 111 |
"cudnn_deterministic": true,
|
| 112 |
+
"core_frames": 256,
|
| 113 |
"input_device": "cpu",
|
| 114 |
"input_contiguous": false
|
| 115 |
},
|
| 116 |
"nar_prefill_source": "prefill",
|
| 117 |
+
"abc_close_seconds": 0.00036335643380880356,
|
| 118 |
+
"memory": {
|
| 119 |
+
"t0": {
|
| 120 |
+
"peak_gib": 3.111,
|
| 121 |
+
"alloc_gib": 3.111,
|
| 122 |
+
"reserved_gib": 4.357,
|
| 123 |
+
"nvml_used_gib": 5.326
|
| 124 |
+
},
|
| 125 |
+
"plan_done": {
|
| 126 |
+
"peak_gib": 3.386,
|
| 127 |
+
"alloc_gib": 3.119,
|
| 128 |
+
"reserved_gib": 3.645,
|
| 129 |
+
"nvml_used_gib": 4.614
|
| 130 |
+
},
|
| 131 |
+
"semantic_done": {
|
| 132 |
+
"peak_gib": 3.83,
|
| 133 |
+
"alloc_gib": 3.127,
|
| 134 |
+
"reserved_gib": 3.992,
|
| 135 |
+
"nvml_used_gib": 4.961
|
| 136 |
+
},
|
| 137 |
+
"nar_done": {
|
| 138 |
+
"peak_gib": 4.312,
|
| 139 |
+
"alloc_gib": 3.127,
|
| 140 |
+
"reserved_gib": 4.631,
|
| 141 |
+
"nvml_used_gib": 5.6
|
| 142 |
+
},
|
| 143 |
+
"vae_done": {
|
| 144 |
+
"peak_gib": 3.528,
|
| 145 |
+
"alloc_gib": 3.127,
|
| 146 |
+
"reserved_gib": 4.631,
|
| 147 |
+
"nvml_used_gib": 5.6
|
| 148 |
+
}
|
| 149 |
+
}
|
| 150 |
},
|
| 151 |
+
"e2e_seconds": 15.849312796257436
|
| 152 |
},
|
| 153 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 154 |
"cfg": 1.0,
|
|
|
|
| 158 |
"kv_cache_dtype": "int8",
|
| 159 |
"captured_sampling": true,
|
| 160 |
"fuse_rms_quant": true,
|
| 161 |
+
"orbitquant_version": "0.9.8",
|
| 162 |
"profile": "fast",
|
| 163 |
"allocator_backend": "native",
|
| 164 |
"modules": 224,
|
evaluation/profiles/exact/run-01/plan.json
CHANGED
|
@@ -9,12 +9,12 @@
|
|
| 9 |
"id": "song"
|
| 10 |
},
|
| 11 |
"timing": {
|
| 12 |
-
"seconds":
|
| 13 |
-
"prefill_seconds": 0.
|
| 14 |
-
"ttft_seconds": 0.
|
| 15 |
"output_tokens": 1424,
|
| 16 |
"content_tokens": 1423,
|
| 17 |
-
"output_tps":
|
| 18 |
"prefix_tokens": 423,
|
| 19 |
"cfg_branches": 1,
|
| 20 |
"sampling": "combined_cuda_graph",
|
|
@@ -25,8 +25,8 @@
|
|
| 25 |
"lookahead": true,
|
| 26 |
"unroll2": false,
|
| 27 |
"first_step_seconds": null,
|
| 28 |
-
"capture_seconds": 0.
|
| 29 |
-
"close_seconds": 0.
|
| 30 |
},
|
| 31 |
"truncated": false,
|
| 32 |
"prefix": [
|
|
|
|
| 9 |
"id": "song"
|
| 10 |
},
|
| 11 |
"timing": {
|
| 12 |
+
"seconds": 3.6763406014069915,
|
| 13 |
+
"prefill_seconds": 0.04775886610150337,
|
| 14 |
+
"ttft_seconds": 0.04778050817549229,
|
| 15 |
"output_tokens": 1424,
|
| 16 |
"content_tokens": 1423,
|
| 17 |
+
"output_tps": 387.341695014606,
|
| 18 |
"prefix_tokens": 423,
|
| 19 |
"cfg_branches": 1,
|
| 20 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 25 |
"lookahead": true,
|
| 26 |
"unroll2": false,
|
| 27 |
"first_step_seconds": null,
|
| 28 |
+
"capture_seconds": 0.03480528108775616,
|
| 29 |
+
"close_seconds": 0.00036335643380880356
|
| 30 |
},
|
| 31 |
"truncated": false,
|
| 32 |
"prefix": [
|
evaluation/profiles/exact/run-01/plan_manifest.json
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
{
|
| 2 |
-
"plan.json": "
|
| 3 |
"abc_tokens.npy": "509147d79fcf8fbe4db07e170c5adcd99986df1ef0ae60607e034e5d43d66992",
|
| 4 |
"prefix.npy": "fe7c7aed0a2cd948b22ce2df55442ca1012152a1c577745afcff296d51b6d594",
|
| 5 |
"score.abc": "027b9b0b33d4c6f837675c864eacac2954c420a18244eb55c6707fbf32b7748d"
|
|
|
|
| 1 |
{
|
| 2 |
+
"plan.json": "8b926548d742c8ceac8733ef5e1d56de6bdcd76bc0df62989869cf431a274af8",
|
| 3 |
"abc_tokens.npy": "509147d79fcf8fbe4db07e170c5adcd99986df1ef0ae60607e034e5d43d66992",
|
| 4 |
"prefix.npy": "fe7c7aed0a2cd948b22ce2df55442ca1012152a1c577745afcff296d51b6d594",
|
| 5 |
"score.abc": "027b9b0b33d4c6f837675c864eacac2954c420a18244eb55c6707fbf32b7748d"
|
evaluation/profiles/exact/run-01/result.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"status": "complete",
|
| 3 |
-
"identity": "
|
| 4 |
"truncated": {
|
| 5 |
"abc": false,
|
| 6 |
"semantic": false
|
|
@@ -29,12 +29,12 @@
|
|
| 29 |
},
|
| 30 |
"timing": {
|
| 31 |
"abc": {
|
| 32 |
-
"seconds":
|
| 33 |
-
"prefill_seconds": 0.
|
| 34 |
-
"ttft_seconds": 0.
|
| 35 |
"output_tokens": 1424,
|
| 36 |
"content_tokens": 1423,
|
| 37 |
-
"output_tps":
|
| 38 |
"prefix_tokens": 423,
|
| 39 |
"cfg_branches": 1,
|
| 40 |
"sampling": "combined_cuda_graph",
|
|
@@ -45,16 +45,16 @@
|
|
| 45 |
"lookahead": true,
|
| 46 |
"unroll2": false,
|
| 47 |
"first_step_seconds": null,
|
| 48 |
-
"capture_seconds": 0.
|
| 49 |
-
"close_seconds": 0.
|
| 50 |
},
|
| 51 |
"semantic": {
|
| 52 |
-
"seconds": 9.
|
| 53 |
-
"prefill_seconds": 0.
|
| 54 |
-
"ttft_seconds": 0.
|
| 55 |
"output_tokens": 4224,
|
| 56 |
"content_tokens": 4223,
|
| 57 |
-
"output_tps":
|
| 58 |
"prefix_tokens": 1848,
|
| 59 |
"cfg_branches": 1,
|
| 60 |
"sampling": "combined_cuda_graph",
|
|
@@ -65,48 +65,88 @@
|
|
| 65 |
"lookahead": true,
|
| 66 |
"unroll2": false,
|
| 67 |
"first_step_seconds": null,
|
| 68 |
-
"capture_seconds": 0.
|
| 69 |
-
"close_seconds": 0.
|
| 70 |
},
|
| 71 |
-
"nar_seconds":
|
| 72 |
-
"vae_seconds": 0.
|
| 73 |
"load": {
|
| 74 |
-
"resolve_and_integrity_seconds":
|
| 75 |
-
"mot_load_seconds":
|
| 76 |
},
|
| 77 |
"stages": {
|
| 78 |
-
"t0->plan_done":
|
| 79 |
-
"plan_done->semantic_done": 9.
|
| 80 |
-
"semantic_done->nar_done":
|
| 81 |
-
"nar_done->vae_done": 0.
|
| 82 |
-
"plan_outside_timer": 0.
|
| 83 |
-
"semantic_outside_timer": 0.
|
| 84 |
-
"semantic_close_seconds": 0.
|
| 85 |
"vae": {
|
| 86 |
-
"load_s": 0.
|
| 87 |
-
"decode_s": 0.
|
| 88 |
-
"post_s":
|
| 89 |
"tiles": [
|
| 90 |
-
0.
|
| 91 |
-
0.
|
| 92 |
-
0.
|
| 93 |
-
0.
|
| 94 |
-
0.
|
| 95 |
-
0.
|
| 96 |
-
0.
|
| 97 |
-
0.
|
| 98 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 99 |
],
|
| 100 |
"cudnn_benchmark": false,
|
| 101 |
"cudnn_deterministic": true,
|
| 102 |
-
"core_frames":
|
| 103 |
"input_device": "cpu",
|
| 104 |
"input_contiguous": false
|
| 105 |
},
|
| 106 |
"nar_prefill_source": "prefill",
|
| 107 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 108 |
},
|
| 109 |
-
"e2e_seconds":
|
| 110 |
},
|
| 111 |
"artifacts": {
|
| 112 |
"abc_tokens.npy": {
|
|
@@ -114,11 +154,11 @@
|
|
| 114 |
"bytes": 5820
|
| 115 |
},
|
| 116 |
"audio.flac": {
|
| 117 |
-
"sha256": "
|
| 118 |
-
"bytes":
|
| 119 |
},
|
| 120 |
"config.json": {
|
| 121 |
-
"sha256": "
|
| 122 |
"bytes": 2476
|
| 123 |
},
|
| 124 |
"latent.npy": {
|
|
@@ -126,11 +166,11 @@
|
|
| 126 |
"bytes": 1081216
|
| 127 |
},
|
| 128 |
"plan.json": {
|
| 129 |
-
"sha256": "
|
| 130 |
-
"bytes":
|
| 131 |
},
|
| 132 |
"plan_manifest.json": {
|
| 133 |
-
"sha256": "
|
| 134 |
"bytes": 341
|
| 135 |
},
|
| 136 |
"prefix.npy": {
|
|
|
|
| 1 |
{
|
| 2 |
"status": "complete",
|
| 3 |
+
"identity": "660a195b70043d34681193d19d9addcc3039a7f82df474b415b607ea08266e52",
|
| 4 |
"truncated": {
|
| 5 |
"abc": false,
|
| 6 |
"semantic": false
|
|
|
|
| 29 |
},
|
| 30 |
"timing": {
|
| 31 |
"abc": {
|
| 32 |
+
"seconds": 3.6763406014069915,
|
| 33 |
+
"prefill_seconds": 0.04775886610150337,
|
| 34 |
+
"ttft_seconds": 0.04778050817549229,
|
| 35 |
"output_tokens": 1424,
|
| 36 |
"content_tokens": 1423,
|
| 37 |
+
"output_tps": 387.341695014606,
|
| 38 |
"prefix_tokens": 423,
|
| 39 |
"cfg_branches": 1,
|
| 40 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 45 |
"lookahead": true,
|
| 46 |
"unroll2": false,
|
| 47 |
"first_step_seconds": null,
|
| 48 |
+
"capture_seconds": 0.03480528108775616,
|
| 49 |
+
"close_seconds": 0.00036335643380880356
|
| 50 |
},
|
| 51 |
"semantic": {
|
| 52 |
+
"seconds": 9.223421677947044,
|
| 53 |
+
"prefill_seconds": 0.05041100736707449,
|
| 54 |
+
"ttft_seconds": 0.050430997274816036,
|
| 55 |
"output_tokens": 4224,
|
| 56 |
"content_tokens": 4223,
|
| 57 |
+
"output_tps": 457.964532847877,
|
| 58 |
"prefix_tokens": 1848,
|
| 59 |
"cfg_branches": 1,
|
| 60 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 65 |
"lookahead": true,
|
| 66 |
"unroll2": false,
|
| 67 |
"first_step_seconds": null,
|
| 68 |
+
"capture_seconds": 0.03198968339711428,
|
| 69 |
+
"close_seconds": 0.0004133721813559532
|
| 70 |
},
|
| 71 |
+
"nar_seconds": 2.7084463676437736,
|
| 72 |
+
"vae_seconds": 0.23706442769616842,
|
| 73 |
"load": {
|
| 74 |
+
"resolve_and_integrity_seconds": 0.20317224133759737,
|
| 75 |
+
"mot_load_seconds": 0.6788896378129721
|
| 76 |
},
|
| 77 |
"stages": {
|
| 78 |
+
"t0->plan_done": 3.6779328491538763,
|
| 79 |
+
"plan_done->semantic_done": 9.22484851628542,
|
| 80 |
+
"semantic_done->nar_done": 2.708730294369161,
|
| 81 |
+
"nar_done->vae_done": 0.2373927216976881,
|
| 82 |
+
"plan_outside_timer": 0.0015922477468848228,
|
| 83 |
+
"semantic_outside_timer": 0.0014268383383750916,
|
| 84 |
+
"semantic_close_seconds": 0.0004133721813559532,
|
| 85 |
"vae": {
|
| 86 |
+
"load_s": 0.0010568005964159966,
|
| 87 |
+
"decode_s": 0.2359218681231141,
|
| 88 |
+
"post_s": 1.5283003449440002e-05,
|
| 89 |
"tiles": [
|
| 90 |
+
0.0142,
|
| 91 |
+
0.0136,
|
| 92 |
+
0.0135,
|
| 93 |
+
0.0135,
|
| 94 |
+
0.0136,
|
| 95 |
+
0.0136,
|
| 96 |
+
0.0137,
|
| 97 |
+
0.0138,
|
| 98 |
+
0.0136,
|
| 99 |
+
0.0137,
|
| 100 |
+
0.0136,
|
| 101 |
+
0.0137,
|
| 102 |
+
0.0136,
|
| 103 |
+
0.0136,
|
| 104 |
+
0.0137,
|
| 105 |
+
0.0136,
|
| 106 |
+
0.0174
|
| 107 |
],
|
| 108 |
"cudnn_benchmark": false,
|
| 109 |
"cudnn_deterministic": true,
|
| 110 |
+
"core_frames": 256,
|
| 111 |
"input_device": "cpu",
|
| 112 |
"input_contiguous": false
|
| 113 |
},
|
| 114 |
"nar_prefill_source": "prefill",
|
| 115 |
+
"abc_close_seconds": 0.00036335643380880356,
|
| 116 |
+
"memory": {
|
| 117 |
+
"t0": {
|
| 118 |
+
"peak_gib": 3.111,
|
| 119 |
+
"alloc_gib": 3.111,
|
| 120 |
+
"reserved_gib": 4.357,
|
| 121 |
+
"nvml_used_gib": 5.326
|
| 122 |
+
},
|
| 123 |
+
"plan_done": {
|
| 124 |
+
"peak_gib": 3.386,
|
| 125 |
+
"alloc_gib": 3.119,
|
| 126 |
+
"reserved_gib": 3.645,
|
| 127 |
+
"nvml_used_gib": 4.614
|
| 128 |
+
},
|
| 129 |
+
"semantic_done": {
|
| 130 |
+
"peak_gib": 3.83,
|
| 131 |
+
"alloc_gib": 3.127,
|
| 132 |
+
"reserved_gib": 3.992,
|
| 133 |
+
"nvml_used_gib": 4.961
|
| 134 |
+
},
|
| 135 |
+
"nar_done": {
|
| 136 |
+
"peak_gib": 4.312,
|
| 137 |
+
"alloc_gib": 3.127,
|
| 138 |
+
"reserved_gib": 4.631,
|
| 139 |
+
"nvml_used_gib": 5.6
|
| 140 |
+
},
|
| 141 |
+
"vae_done": {
|
| 142 |
+
"peak_gib": 3.528,
|
| 143 |
+
"alloc_gib": 3.127,
|
| 144 |
+
"reserved_gib": 4.631,
|
| 145 |
+
"nvml_used_gib": 5.6
|
| 146 |
+
}
|
| 147 |
+
}
|
| 148 |
},
|
| 149 |
+
"e2e_seconds": 15.849312796257436
|
| 150 |
},
|
| 151 |
"artifacts": {
|
| 152 |
"abc_tokens.npy": {
|
|
|
|
| 154 |
"bytes": 5820
|
| 155 |
},
|
| 156 |
"audio.flac": {
|
| 157 |
+
"sha256": "2c8ab1306a961caeecaa97f22ff855ea5c41810e87037454cd0b45f91f74d7fb",
|
| 158 |
+
"bytes": 34180094
|
| 159 |
},
|
| 160 |
"config.json": {
|
| 161 |
+
"sha256": "9759afacf903510fff557ab5bb8b459b1dca120b1d7ae1e6aaf8f3a7995a6008",
|
| 162 |
"bytes": 2476
|
| 163 |
},
|
| 164 |
"latent.npy": {
|
|
|
|
| 166 |
"bytes": 1081216
|
| 167 |
},
|
| 168 |
"plan.json": {
|
| 169 |
+
"sha256": "8b926548d742c8ceac8733ef5e1d56de6bdcd76bc0df62989869cf431a274af8",
|
| 170 |
+
"bytes": 33019
|
| 171 |
},
|
| 172 |
"plan_manifest.json": {
|
| 173 |
+
"sha256": "a738569a1d7393ea4d9c30fda391c7271d0c085d41b01b4910182397a725d355",
|
| 174 |
"bytes": 341
|
| 175 |
},
|
| 176 |
"prefix.npy": {
|
evaluation/profiles/exact/summary.json
CHANGED
|
@@ -1,42 +1,43 @@
|
|
| 1 |
[
|
| 2 |
{
|
| 3 |
"bundle": "v11",
|
| 4 |
-
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default",
|
| 5 |
"flags": [
|
| 6 |
"abc_crop",
|
| 7 |
"lookahead",
|
| 8 |
"resident",
|
| 9 |
"vae_compile",
|
|
|
|
| 10 |
"vae_mode=default"
|
| 11 |
],
|
| 12 |
-
"native_sha256": "
|
| 13 |
"repeat": 0,
|
| 14 |
-
"wall_seconds":
|
| 15 |
-
"load_seconds":
|
| 16 |
"audio_seconds": 168.91866666666667,
|
| 17 |
-
"audio_per_wall":
|
| 18 |
-
"abc_s":
|
| 19 |
"abc_tokens": 1424,
|
| 20 |
-
"sem_s": 9.
|
| 21 |
"sem_tokens": 4224,
|
| 22 |
-
"abc_ms_per_tok":
|
| 23 |
-
"sem_ms_per_tok": 2.
|
| 24 |
-
"nar_s":
|
| 25 |
-
"vae_s":
|
| 26 |
-
"nvml_peak_gib": 5.
|
| 27 |
-
"torch_peak_gib":
|
| 28 |
"torch_reserved_peak_gib": 4.357421875,
|
| 29 |
"semantic_sha256": "36a231d034d554c12388e82b4cd4be2598079ae23f0e7fdfb14187e339adcb0c",
|
| 30 |
-
"waveform_sha256": "
|
| 31 |
"latent_sha256": "5c566f70f058de65c5f3e8daf3dab9846cb354ad72b26df9680d472f193b6c7d",
|
| 32 |
"timing": {
|
| 33 |
"abc": {
|
| 34 |
-
"seconds":
|
| 35 |
-
"prefill_seconds":
|
| 36 |
-
"ttft_seconds":
|
| 37 |
"output_tokens": 1424,
|
| 38 |
"content_tokens": 1423,
|
| 39 |
-
"output_tps":
|
| 40 |
"prefix_tokens": 423,
|
| 41 |
"cfg_branches": 1,
|
| 42 |
"sampling": "combined_cuda_graph",
|
|
@@ -47,16 +48,16 @@
|
|
| 47 |
"lookahead": true,
|
| 48 |
"unroll2": false,
|
| 49 |
"first_step_seconds": null,
|
| 50 |
-
"capture_seconds": 0.
|
| 51 |
-
"close_seconds": 0.
|
| 52 |
},
|
| 53 |
"semantic": {
|
| 54 |
-
"seconds": 9.
|
| 55 |
-
"prefill_seconds": 0.
|
| 56 |
-
"ttft_seconds": 0.
|
| 57 |
"output_tokens": 4224,
|
| 58 |
"content_tokens": 4223,
|
| 59 |
-
"output_tps":
|
| 60 |
"prefix_tokens": 1848,
|
| 61 |
"cfg_branches": 1,
|
| 62 |
"sampling": "combined_cuda_graph",
|
|
@@ -67,48 +68,88 @@
|
|
| 67 |
"lookahead": true,
|
| 68 |
"unroll2": false,
|
| 69 |
"first_step_seconds": null,
|
| 70 |
-
"capture_seconds": 0.
|
| 71 |
-
"close_seconds": 0.
|
| 72 |
},
|
| 73 |
-
"nar_seconds":
|
| 74 |
-
"vae_seconds":
|
| 75 |
"load": {
|
| 76 |
-
"resolve_and_integrity_seconds":
|
| 77 |
-
"mot_load_seconds":
|
| 78 |
},
|
| 79 |
"stages": {
|
| 80 |
-
"t0->plan_done":
|
| 81 |
-
"plan_done->semantic_done": 9.
|
| 82 |
-
"semantic_done->nar_done":
|
| 83 |
-
"nar_done->vae_done":
|
| 84 |
-
"plan_outside_timer": 0.
|
| 85 |
-
"semantic_outside_timer": 0.
|
| 86 |
-
"semantic_close_seconds": 0.
|
| 87 |
"vae": {
|
| 88 |
-
"load_s":
|
| 89 |
-
"decode_s":
|
| 90 |
-
"post_s":
|
| 91 |
"tiles": [
|
| 92 |
-
|
| 93 |
-
0.
|
| 94 |
-
0.
|
| 95 |
-
0.
|
| 96 |
-
0.
|
| 97 |
-
0.
|
| 98 |
-
0.
|
| 99 |
-
0.
|
| 100 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 101 |
],
|
| 102 |
"cudnn_benchmark": false,
|
| 103 |
"cudnn_deterministic": true,
|
| 104 |
-
"core_frames":
|
| 105 |
"input_device": "cpu",
|
| 106 |
"input_contiguous": false
|
| 107 |
},
|
| 108 |
"nar_prefill_source": "prefill",
|
| 109 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 110 |
},
|
| 111 |
-
"e2e_seconds":
|
| 112 |
},
|
| 113 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 114 |
"cfg": 1.0,
|
|
@@ -118,7 +159,7 @@
|
|
| 118 |
"kv_cache_dtype": "int8",
|
| 119 |
"captured_sampling": true,
|
| 120 |
"fuse_rms_quant": true,
|
| 121 |
-
"orbitquant_version": "0.9.
|
| 122 |
"profile": "fast",
|
| 123 |
"allocator_backend": "native",
|
| 124 |
"modules": 224,
|
|
@@ -140,42 +181,43 @@
|
|
| 140 |
},
|
| 141 |
{
|
| 142 |
"bundle": "v11",
|
| 143 |
-
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default",
|
| 144 |
"flags": [
|
| 145 |
"abc_crop",
|
| 146 |
"lookahead",
|
| 147 |
"resident",
|
| 148 |
"vae_compile",
|
|
|
|
| 149 |
"vae_mode=default"
|
| 150 |
],
|
| 151 |
-
"native_sha256": "
|
| 152 |
"repeat": 1,
|
| 153 |
-
"wall_seconds":
|
| 154 |
-
"load_seconds":
|
| 155 |
"audio_seconds": 168.91866666666667,
|
| 156 |
-
"audio_per_wall":
|
| 157 |
-
"abc_s":
|
| 158 |
"abc_tokens": 1424,
|
| 159 |
-
"sem_s": 9.
|
| 160 |
"sem_tokens": 4224,
|
| 161 |
-
"abc_ms_per_tok": 2.
|
| 162 |
-
"sem_ms_per_tok": 2.
|
| 163 |
-
"nar_s":
|
| 164 |
-
"vae_s": 0.
|
| 165 |
-
"nvml_peak_gib": 5.
|
| 166 |
-
"torch_peak_gib":
|
| 167 |
"torch_reserved_peak_gib": 4.630859375,
|
| 168 |
"semantic_sha256": "36a231d034d554c12388e82b4cd4be2598079ae23f0e7fdfb14187e339adcb0c",
|
| 169 |
-
"waveform_sha256": "
|
| 170 |
"latent_sha256": "5c566f70f058de65c5f3e8daf3dab9846cb354ad72b26df9680d472f193b6c7d",
|
| 171 |
"timing": {
|
| 172 |
"abc": {
|
| 173 |
-
"seconds":
|
| 174 |
-
"prefill_seconds": 0.
|
| 175 |
-
"ttft_seconds": 0.
|
| 176 |
"output_tokens": 1424,
|
| 177 |
"content_tokens": 1423,
|
| 178 |
-
"output_tps":
|
| 179 |
"prefix_tokens": 423,
|
| 180 |
"cfg_branches": 1,
|
| 181 |
"sampling": "combined_cuda_graph",
|
|
@@ -186,16 +228,16 @@
|
|
| 186 |
"lookahead": true,
|
| 187 |
"unroll2": false,
|
| 188 |
"first_step_seconds": null,
|
| 189 |
-
"capture_seconds": 0.
|
| 190 |
-
"close_seconds": 0.
|
| 191 |
},
|
| 192 |
"semantic": {
|
| 193 |
-
"seconds": 9.
|
| 194 |
-
"prefill_seconds": 0.
|
| 195 |
-
"ttft_seconds": 0.
|
| 196 |
"output_tokens": 4224,
|
| 197 |
"content_tokens": 4223,
|
| 198 |
-
"output_tps":
|
| 199 |
"prefix_tokens": 1848,
|
| 200 |
"cfg_branches": 1,
|
| 201 |
"sampling": "combined_cuda_graph",
|
|
@@ -206,48 +248,88 @@
|
|
| 206 |
"lookahead": true,
|
| 207 |
"unroll2": false,
|
| 208 |
"first_step_seconds": null,
|
| 209 |
-
"capture_seconds": 0.
|
| 210 |
-
"close_seconds": 0.
|
| 211 |
},
|
| 212 |
-
"nar_seconds":
|
| 213 |
-
"vae_seconds": 0.
|
| 214 |
"load": {
|
| 215 |
-
"resolve_and_integrity_seconds":
|
| 216 |
-
"mot_load_seconds":
|
| 217 |
},
|
| 218 |
"stages": {
|
| 219 |
-
"t0->plan_done":
|
| 220 |
-
"plan_done->semantic_done": 9.
|
| 221 |
-
"semantic_done->nar_done":
|
| 222 |
-
"nar_done->vae_done": 0.
|
| 223 |
-
"plan_outside_timer": 0.
|
| 224 |
-
"semantic_outside_timer": 0.
|
| 225 |
-
"semantic_close_seconds": 0.
|
| 226 |
"vae": {
|
| 227 |
-
"load_s": 0.
|
| 228 |
-
"decode_s": 0.
|
| 229 |
-
"post_s":
|
| 230 |
"tiles": [
|
| 231 |
-
0.
|
| 232 |
-
0.
|
| 233 |
-
0.
|
| 234 |
-
0.
|
| 235 |
-
0.
|
| 236 |
-
0.
|
| 237 |
-
0.
|
| 238 |
-
0.
|
| 239 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 240 |
],
|
| 241 |
"cudnn_benchmark": false,
|
| 242 |
"cudnn_deterministic": true,
|
| 243 |
-
"core_frames":
|
| 244 |
"input_device": "cpu",
|
| 245 |
"input_contiguous": false
|
| 246 |
},
|
| 247 |
"nar_prefill_source": "prefill",
|
| 248 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 249 |
},
|
| 250 |
-
"e2e_seconds":
|
| 251 |
},
|
| 252 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 253 |
"cfg": 1.0,
|
|
@@ -257,7 +339,7 @@
|
|
| 257 |
"kv_cache_dtype": "int8",
|
| 258 |
"captured_sampling": true,
|
| 259 |
"fuse_rms_quant": true,
|
| 260 |
-
"orbitquant_version": "0.9.
|
| 261 |
"profile": "fast",
|
| 262 |
"allocator_backend": "native",
|
| 263 |
"modules": 224,
|
|
|
|
| 1 |
[
|
| 2 |
{
|
| 3 |
"bundle": "v11",
|
| 4 |
+
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,vae_core=256",
|
| 5 |
"flags": [
|
| 6 |
"abc_crop",
|
| 7 |
"lookahead",
|
| 8 |
"resident",
|
| 9 |
"vae_compile",
|
| 10 |
+
"vae_core=256",
|
| 11 |
"vae_mode=default"
|
| 12 |
],
|
| 13 |
+
"native_sha256": "118d64db97c3200ef785e83b8d703ebc3b480a799bb17a6affb104d1b4faa28d",
|
| 14 |
"repeat": 0,
|
| 15 |
+
"wall_seconds": 18.547393172048032,
|
| 16 |
+
"load_seconds": 2.2395440358668566,
|
| 17 |
"audio_seconds": 168.91866666666667,
|
| 18 |
+
"audio_per_wall": 9.107407445335047,
|
| 19 |
+
"abc_s": 4.272322495467961,
|
| 20 |
"abc_tokens": 1424,
|
| 21 |
+
"sem_s": 9.2354048024863,
|
| 22 |
"sem_tokens": 4224,
|
| 23 |
+
"abc_ms_per_tok": 3.0002264715364895,
|
| 24 |
+
"sem_ms_per_tok": 2.186412121800734,
|
| 25 |
+
"nar_s": 2.741346231661737,
|
| 26 |
+
"vae_s": 2.292730061337352,
|
| 27 |
+
"nvml_peak_gib": 5.326416015625,
|
| 28 |
+
"torch_peak_gib": 3.11099910736084,
|
| 29 |
"torch_reserved_peak_gib": 4.357421875,
|
| 30 |
"semantic_sha256": "36a231d034d554c12388e82b4cd4be2598079ae23f0e7fdfb14187e339adcb0c",
|
| 31 |
+
"waveform_sha256": "e0aeec8e4144ed157e18c54b2e8141a8ba6051343a3ca178cd85bfbc133749c9",
|
| 32 |
"latent_sha256": "5c566f70f058de65c5f3e8daf3dab9846cb354ad72b26df9680d472f193b6c7d",
|
| 33 |
"timing": {
|
| 34 |
"abc": {
|
| 35 |
+
"seconds": 4.272322495467961,
|
| 36 |
+
"prefill_seconds": 0.6354847745969892,
|
| 37 |
+
"ttft_seconds": 0.6366398222744465,
|
| 38 |
"output_tokens": 1424,
|
| 39 |
"content_tokens": 1423,
|
| 40 |
+
"output_tps": 333.3081717287414,
|
| 41 |
"prefix_tokens": 423,
|
| 42 |
"cfg_branches": 1,
|
| 43 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 48 |
"lookahead": true,
|
| 49 |
"unroll2": false,
|
| 50 |
"first_step_seconds": null,
|
| 51 |
+
"capture_seconds": 0.0413797153159976,
|
| 52 |
+
"close_seconds": 0.00031330157071352005
|
| 53 |
},
|
| 54 |
"semantic": {
|
| 55 |
+
"seconds": 9.2354048024863,
|
| 56 |
+
"prefill_seconds": 0.05777517054229975,
|
| 57 |
+
"ttft_seconds": 0.05779500026255846,
|
| 58 |
"output_tokens": 4224,
|
| 59 |
"content_tokens": 4223,
|
| 60 |
+
"output_tps": 457.37031460308486,
|
| 61 |
"prefix_tokens": 1848,
|
| 62 |
"cfg_branches": 1,
|
| 63 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 68 |
"lookahead": true,
|
| 69 |
"unroll2": false,
|
| 70 |
"first_step_seconds": null,
|
| 71 |
+
"capture_seconds": 0.032724469900131226,
|
| 72 |
+
"close_seconds": 0.00040849484503269196
|
| 73 |
},
|
| 74 |
+
"nar_seconds": 2.741346231661737,
|
| 75 |
+
"vae_seconds": 2.292730061337352,
|
| 76 |
"load": {
|
| 77 |
+
"resolve_and_integrity_seconds": 0.20317224133759737,
|
| 78 |
+
"mot_load_seconds": 0.6788896378129721
|
| 79 |
},
|
| 80 |
"stages": {
|
| 81 |
+
"t0->plan_done": 4.274795697070658,
|
| 82 |
+
"plan_done->semantic_done": 9.236809186637402,
|
| 83 |
+
"semantic_done->nar_done": 2.7416386613622308,
|
| 84 |
+
"nar_done->vae_done": 2.2930279998108745,
|
| 85 |
+
"plan_outside_timer": 0.0024732016026973724,
|
| 86 |
+
"semantic_outside_timer": 0.0014043841511011124,
|
| 87 |
+
"semantic_close_seconds": 0.00040849484503269196,
|
| 88 |
"vae": {
|
| 89 |
+
"load_s": 0.2927389871329069,
|
| 90 |
+
"decode_s": 1.9990807436406612,
|
| 91 |
+
"post_s": 1.1868774890899658e-05,
|
| 92 |
"tiles": [
|
| 93 |
+
1.7568,
|
| 94 |
+
0.0142,
|
| 95 |
+
0.0142,
|
| 96 |
+
0.0141,
|
| 97 |
+
0.0141,
|
| 98 |
+
0.014,
|
| 99 |
+
0.014,
|
| 100 |
+
0.014,
|
| 101 |
+
0.0136,
|
| 102 |
+
0.0135,
|
| 103 |
+
0.0136,
|
| 104 |
+
0.0135,
|
| 105 |
+
0.0139,
|
| 106 |
+
0.0138,
|
| 107 |
+
0.0137,
|
| 108 |
+
0.0137,
|
| 109 |
+
0.0342
|
| 110 |
],
|
| 111 |
"cudnn_benchmark": false,
|
| 112 |
"cudnn_deterministic": true,
|
| 113 |
+
"core_frames": 256,
|
| 114 |
"input_device": "cpu",
|
| 115 |
"input_contiguous": false
|
| 116 |
},
|
| 117 |
"nar_prefill_source": "prefill",
|
| 118 |
+
"abc_close_seconds": 0.00031330157071352005,
|
| 119 |
+
"memory": {
|
| 120 |
+
"t0": {
|
| 121 |
+
"peak_gib": 2.831,
|
| 122 |
+
"alloc_gib": 2.831,
|
| 123 |
+
"reserved_gib": 3.068,
|
| 124 |
+
"nvml_used_gib": 3.928
|
| 125 |
+
},
|
| 126 |
+
"plan_done": {
|
| 127 |
+
"peak_gib": 3.108,
|
| 128 |
+
"alloc_gib": 2.855,
|
| 129 |
+
"reserved_gib": 3.371,
|
| 130 |
+
"nvml_used_gib": 4.328
|
| 131 |
+
},
|
| 132 |
+
"semantic_done": {
|
| 133 |
+
"peak_gib": 3.566,
|
| 134 |
+
"alloc_gib": 2.863,
|
| 135 |
+
"reserved_gib": 3.719,
|
| 136 |
+
"nvml_used_gib": 4.678
|
| 137 |
+
},
|
| 138 |
+
"nar_done": {
|
| 139 |
+
"peak_gib": 4.048,
|
| 140 |
+
"alloc_gib": 2.864,
|
| 141 |
+
"reserved_gib": 4.357,
|
| 142 |
+
"nvml_used_gib": 5.317
|
| 143 |
+
},
|
| 144 |
+
"vae_done": {
|
| 145 |
+
"peak_gib": 3.512,
|
| 146 |
+
"alloc_gib": 3.111,
|
| 147 |
+
"reserved_gib": 4.357,
|
| 148 |
+
"nvml_used_gib": 5.326
|
| 149 |
+
}
|
| 150 |
+
}
|
| 151 |
},
|
| 152 |
+
"e2e_seconds": 18.54658360313624
|
| 153 |
},
|
| 154 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 155 |
"cfg": 1.0,
|
|
|
|
| 159 |
"kv_cache_dtype": "int8",
|
| 160 |
"captured_sampling": true,
|
| 161 |
"fuse_rms_quant": true,
|
| 162 |
+
"orbitquant_version": "0.9.8",
|
| 163 |
"profile": "fast",
|
| 164 |
"allocator_backend": "native",
|
| 165 |
"modules": 224,
|
|
|
|
| 181 |
},
|
| 182 |
{
|
| 183 |
"bundle": "v11",
|
| 184 |
+
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,vae_core=256",
|
| 185 |
"flags": [
|
| 186 |
"abc_crop",
|
| 187 |
"lookahead",
|
| 188 |
"resident",
|
| 189 |
"vae_compile",
|
| 190 |
+
"vae_core=256",
|
| 191 |
"vae_mode=default"
|
| 192 |
],
|
| 193 |
+
"native_sha256": "118d64db97c3200ef785e83b8d703ebc3b480a799bb17a6affb104d1b4faa28d",
|
| 194 |
"repeat": 1,
|
| 195 |
+
"wall_seconds": 15.850000811740756,
|
| 196 |
+
"load_seconds": 2.2395440358668566,
|
| 197 |
"audio_seconds": 168.91866666666667,
|
| 198 |
+
"audio_per_wall": 10.657328581430834,
|
| 199 |
+
"abc_s": 3.6763406014069915,
|
| 200 |
"abc_tokens": 1424,
|
| 201 |
+
"sem_s": 9.223421677947044,
|
| 202 |
"sem_tokens": 4224,
|
| 203 |
+
"abc_ms_per_tok": 2.5816998605386177,
|
| 204 |
+
"sem_ms_per_tok": 2.1835752078473116,
|
| 205 |
+
"nar_s": 2.7084463676437736,
|
| 206 |
+
"vae_s": 0.23706442769616842,
|
| 207 |
+
"nvml_peak_gib": 5.599853515625,
|
| 208 |
+
"torch_peak_gib": 3.12686824798584,
|
| 209 |
"torch_reserved_peak_gib": 4.630859375,
|
| 210 |
"semantic_sha256": "36a231d034d554c12388e82b4cd4be2598079ae23f0e7fdfb14187e339adcb0c",
|
| 211 |
+
"waveform_sha256": "e0aeec8e4144ed157e18c54b2e8141a8ba6051343a3ca178cd85bfbc133749c9",
|
| 212 |
"latent_sha256": "5c566f70f058de65c5f3e8daf3dab9846cb354ad72b26df9680d472f193b6c7d",
|
| 213 |
"timing": {
|
| 214 |
"abc": {
|
| 215 |
+
"seconds": 3.6763406014069915,
|
| 216 |
+
"prefill_seconds": 0.04775886610150337,
|
| 217 |
+
"ttft_seconds": 0.04778050817549229,
|
| 218 |
"output_tokens": 1424,
|
| 219 |
"content_tokens": 1423,
|
| 220 |
+
"output_tps": 387.341695014606,
|
| 221 |
"prefix_tokens": 423,
|
| 222 |
"cfg_branches": 1,
|
| 223 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 228 |
"lookahead": true,
|
| 229 |
"unroll2": false,
|
| 230 |
"first_step_seconds": null,
|
| 231 |
+
"capture_seconds": 0.03480528108775616,
|
| 232 |
+
"close_seconds": 0.00036335643380880356
|
| 233 |
},
|
| 234 |
"semantic": {
|
| 235 |
+
"seconds": 9.223421677947044,
|
| 236 |
+
"prefill_seconds": 0.05041100736707449,
|
| 237 |
+
"ttft_seconds": 0.050430997274816036,
|
| 238 |
"output_tokens": 4224,
|
| 239 |
"content_tokens": 4223,
|
| 240 |
+
"output_tps": 457.964532847877,
|
| 241 |
"prefix_tokens": 1848,
|
| 242 |
"cfg_branches": 1,
|
| 243 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 248 |
"lookahead": true,
|
| 249 |
"unroll2": false,
|
| 250 |
"first_step_seconds": null,
|
| 251 |
+
"capture_seconds": 0.03198968339711428,
|
| 252 |
+
"close_seconds": 0.0004133721813559532
|
| 253 |
},
|
| 254 |
+
"nar_seconds": 2.7084463676437736,
|
| 255 |
+
"vae_seconds": 0.23706442769616842,
|
| 256 |
"load": {
|
| 257 |
+
"resolve_and_integrity_seconds": 0.20317224133759737,
|
| 258 |
+
"mot_load_seconds": 0.6788896378129721
|
| 259 |
},
|
| 260 |
"stages": {
|
| 261 |
+
"t0->plan_done": 3.6779328491538763,
|
| 262 |
+
"plan_done->semantic_done": 9.22484851628542,
|
| 263 |
+
"semantic_done->nar_done": 2.708730294369161,
|
| 264 |
+
"nar_done->vae_done": 0.2373927216976881,
|
| 265 |
+
"plan_outside_timer": 0.0015922477468848228,
|
| 266 |
+
"semantic_outside_timer": 0.0014268383383750916,
|
| 267 |
+
"semantic_close_seconds": 0.0004133721813559532,
|
| 268 |
"vae": {
|
| 269 |
+
"load_s": 0.0010568005964159966,
|
| 270 |
+
"decode_s": 0.2359218681231141,
|
| 271 |
+
"post_s": 1.5283003449440002e-05,
|
| 272 |
"tiles": [
|
| 273 |
+
0.0142,
|
| 274 |
+
0.0136,
|
| 275 |
+
0.0135,
|
| 276 |
+
0.0135,
|
| 277 |
+
0.0136,
|
| 278 |
+
0.0136,
|
| 279 |
+
0.0137,
|
| 280 |
+
0.0138,
|
| 281 |
+
0.0136,
|
| 282 |
+
0.0137,
|
| 283 |
+
0.0136,
|
| 284 |
+
0.0137,
|
| 285 |
+
0.0136,
|
| 286 |
+
0.0136,
|
| 287 |
+
0.0137,
|
| 288 |
+
0.0136,
|
| 289 |
+
0.0174
|
| 290 |
],
|
| 291 |
"cudnn_benchmark": false,
|
| 292 |
"cudnn_deterministic": true,
|
| 293 |
+
"core_frames": 256,
|
| 294 |
"input_device": "cpu",
|
| 295 |
"input_contiguous": false
|
| 296 |
},
|
| 297 |
"nar_prefill_source": "prefill",
|
| 298 |
+
"abc_close_seconds": 0.00036335643380880356,
|
| 299 |
+
"memory": {
|
| 300 |
+
"t0": {
|
| 301 |
+
"peak_gib": 3.111,
|
| 302 |
+
"alloc_gib": 3.111,
|
| 303 |
+
"reserved_gib": 4.357,
|
| 304 |
+
"nvml_used_gib": 5.326
|
| 305 |
+
},
|
| 306 |
+
"plan_done": {
|
| 307 |
+
"peak_gib": 3.386,
|
| 308 |
+
"alloc_gib": 3.119,
|
| 309 |
+
"reserved_gib": 3.645,
|
| 310 |
+
"nvml_used_gib": 4.614
|
| 311 |
+
},
|
| 312 |
+
"semantic_done": {
|
| 313 |
+
"peak_gib": 3.83,
|
| 314 |
+
"alloc_gib": 3.127,
|
| 315 |
+
"reserved_gib": 3.992,
|
| 316 |
+
"nvml_used_gib": 4.961
|
| 317 |
+
},
|
| 318 |
+
"nar_done": {
|
| 319 |
+
"peak_gib": 4.312,
|
| 320 |
+
"alloc_gib": 3.127,
|
| 321 |
+
"reserved_gib": 4.631,
|
| 322 |
+
"nvml_used_gib": 5.6
|
| 323 |
+
},
|
| 324 |
+
"vae_done": {
|
| 325 |
+
"peak_gib": 3.528,
|
| 326 |
+
"alloc_gib": 3.127,
|
| 327 |
+
"reserved_gib": 4.631,
|
| 328 |
+
"nvml_used_gib": 5.6
|
| 329 |
+
}
|
| 330 |
+
}
|
| 331 |
},
|
| 332 |
+
"e2e_seconds": 15.849312796257436
|
| 333 |
},
|
| 334 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 335 |
"cfg": 1.0,
|
|
|
|
| 339 |
"kv_cache_dtype": "int8",
|
| 340 |
"captured_sampling": true,
|
| 341 |
"fuse_rms_quant": true,
|
| 342 |
+
"orbitquant_version": "0.9.8",
|
| 343 |
"profile": "fast",
|
| 344 |
"allocator_backend": "native",
|
| 345 |
"modules": 224,
|
evaluation/profiles/fast-cfg1.5/run-00/config.json
CHANGED
|
@@ -35,12 +35,12 @@
|
|
| 35 |
"model_dtype": "bfloat16",
|
| 36 |
"vae_dtype": "float32",
|
| 37 |
"vae_decode": "halo_crop",
|
| 38 |
-
"vae_core_frames":
|
| 39 |
"vae_halo_frames": 16,
|
| 40 |
"device": "cuda:0",
|
| 41 |
"memory_budget_gib": 18.0,
|
| 42 |
"offload_ar": false,
|
| 43 |
-
"runtime_sha256": "
|
| 44 |
"decoder_release": "standard",
|
| 45 |
"validation_status": "unvalidated",
|
| 46 |
"runtime_profile": "fast",
|
|
@@ -48,7 +48,7 @@
|
|
| 48 |
"captured_sampling": true,
|
| 49 |
"kv_cache_dtype": "int8",
|
| 50 |
"int8_attention": "ordered",
|
| 51 |
-
"runtime_bundle_sha256": "
|
| 52 |
"vae_compute_dtype": "torch.float16",
|
| 53 |
"expert_staging": false,
|
| 54 |
"orbitquant": {
|
|
|
|
| 35 |
"model_dtype": "bfloat16",
|
| 36 |
"vae_dtype": "float32",
|
| 37 |
"vae_decode": "halo_crop",
|
| 38 |
+
"vae_core_frames": 256,
|
| 39 |
"vae_halo_frames": 16,
|
| 40 |
"device": "cuda:0",
|
| 41 |
"memory_budget_gib": 18.0,
|
| 42 |
"offload_ar": false,
|
| 43 |
+
"runtime_sha256": "bb189091c79b42dd3eca4f24f6239294df3c517fb2b149fc3b30ada3be287119",
|
| 44 |
"decoder_release": "standard",
|
| 45 |
"validation_status": "unvalidated",
|
| 46 |
"runtime_profile": "fast",
|
|
|
|
| 48 |
"captured_sampling": true,
|
| 49 |
"kv_cache_dtype": "int8",
|
| 50 |
"int8_attention": "ordered",
|
| 51 |
+
"runtime_bundle_sha256": "269214c6459739a0f08889532d5fc76f3e6f48f3908807466e14e5da4ecb1c98",
|
| 52 |
"vae_compute_dtype": "torch.float16",
|
| 53 |
"expert_staging": false,
|
| 54 |
"orbitquant": {
|
evaluation/profiles/fast-cfg1.5/run-00/metrics.json
CHANGED
|
@@ -1,45 +1,46 @@
|
|
| 1 |
{
|
| 2 |
"bundle": "v11",
|
| 3 |
-
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,kv_handoff,
|
| 4 |
"flags": [
|
| 5 |
"abc_crop",
|
| 6 |
"int8_heads",
|
| 7 |
"kv_handoff",
|
|
|
|
| 8 |
"lookahead",
|
| 9 |
-
"nar_compile",
|
| 10 |
"resident",
|
| 11 |
"vae_compile",
|
|
|
|
| 12 |
"vae_mode=default",
|
| 13 |
"w8a8"
|
| 14 |
],
|
| 15 |
-
"native_sha256": "
|
| 16 |
"repeat": 0,
|
| 17 |
-
"wall_seconds":
|
| 18 |
-
"load_seconds":
|
| 19 |
"audio_seconds": 140.63866666666667,
|
| 20 |
-
"audio_per_wall":
|
| 21 |
-
"abc_s":
|
| 22 |
"abc_tokens": 1249,
|
| 23 |
-
"sem_s":
|
| 24 |
"sem_tokens": 3517,
|
| 25 |
-
"abc_ms_per_tok":
|
| 26 |
-
"sem_ms_per_tok": 2.
|
| 27 |
-
"nar_s":
|
| 28 |
-
"vae_s":
|
| 29 |
-
"nvml_peak_gib":
|
| 30 |
-
"torch_peak_gib":
|
| 31 |
-
"torch_reserved_peak_gib":
|
| 32 |
"semantic_sha256": "4c3a54b7d2d1a1be8bd05d35f4b773e1dc375533dcfe49694b1791126c3797a4",
|
| 33 |
-
"waveform_sha256": "
|
| 34 |
-
"latent_sha256": "
|
| 35 |
"timing": {
|
| 36 |
"abc": {
|
| 37 |
-
"seconds":
|
| 38 |
-
"prefill_seconds":
|
| 39 |
-
"ttft_seconds":
|
| 40 |
"output_tokens": 1249,
|
| 41 |
"content_tokens": 1248,
|
| 42 |
-
"output_tps":
|
| 43 |
"prefix_tokens": 423,
|
| 44 |
"cfg_branches": 1,
|
| 45 |
"sampling": "combined_cuda_graph",
|
|
@@ -50,16 +51,16 @@
|
|
| 50 |
"lookahead": true,
|
| 51 |
"unroll2": false,
|
| 52 |
"first_step_seconds": null,
|
| 53 |
-
"capture_seconds": 0.
|
| 54 |
-
"close_seconds": 0.
|
| 55 |
},
|
| 56 |
"semantic": {
|
| 57 |
-
"seconds":
|
| 58 |
-
"prefill_seconds":
|
| 59 |
-
"ttft_seconds":
|
| 60 |
"output_tokens": 3517,
|
| 61 |
"content_tokens": 3516,
|
| 62 |
-
"output_tps":
|
| 63 |
"prefix_tokens": 1673,
|
| 64 |
"cfg_branches": 2,
|
| 65 |
"sampling": "combined_cuda_graph",
|
|
@@ -70,46 +71,85 @@
|
|
| 70 |
"lookahead": true,
|
| 71 |
"unroll2": false,
|
| 72 |
"first_step_seconds": null,
|
| 73 |
-
"capture_seconds": 0.
|
| 74 |
-
"close_seconds": 0.
|
| 75 |
},
|
| 76 |
-
"nar_seconds":
|
| 77 |
-
"vae_seconds":
|
| 78 |
"load": {
|
| 79 |
-
"resolve_and_integrity_seconds":
|
| 80 |
-
"mot_load_seconds":
|
| 81 |
},
|
| 82 |
"stages": {
|
| 83 |
-
"t0->plan_done":
|
| 84 |
-
"plan_done->semantic_done":
|
| 85 |
-
"semantic_done->nar_done":
|
| 86 |
-
"nar_done->vae_done":
|
| 87 |
-
"plan_outside_timer": 0.
|
| 88 |
-
"semantic_outside_timer": 0.
|
| 89 |
-
"semantic_close_seconds": 0.
|
| 90 |
"vae": {
|
| 91 |
-
"load_s":
|
| 92 |
-
"decode_s":
|
| 93 |
-
"post_s":
|
| 94 |
"tiles": [
|
| 95 |
-
|
| 96 |
-
0.
|
| 97 |
-
0.
|
| 98 |
-
0.
|
| 99 |
-
0.
|
| 100 |
-
0.
|
| 101 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 102 |
],
|
| 103 |
"cudnn_benchmark": false,
|
| 104 |
"cudnn_deterministic": true,
|
| 105 |
-
"core_frames":
|
| 106 |
"input_device": "cpu",
|
| 107 |
"input_contiguous": false
|
| 108 |
},
|
| 109 |
"nar_prefill_source": "handoff",
|
| 110 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 111 |
},
|
| 112 |
-
"e2e_seconds":
|
| 113 |
},
|
| 114 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 115 |
"cfg": 1.5,
|
|
@@ -119,7 +159,7 @@
|
|
| 119 |
"kv_cache_dtype": "int8",
|
| 120 |
"captured_sampling": true,
|
| 121 |
"fuse_rms_quant": true,
|
| 122 |
-
"orbitquant_version": "0.9.
|
| 123 |
"profile": "fast",
|
| 124 |
"allocator_backend": "native",
|
| 125 |
"modules": 224,
|
|
|
|
| 1 |
{
|
| 2 |
"bundle": "v11",
|
| 3 |
+
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,vae_core=256,kv_handoff,lazy_handoff,int8_heads,w8a8",
|
| 4 |
"flags": [
|
| 5 |
"abc_crop",
|
| 6 |
"int8_heads",
|
| 7 |
"kv_handoff",
|
| 8 |
+
"lazy_handoff",
|
| 9 |
"lookahead",
|
|
|
|
| 10 |
"resident",
|
| 11 |
"vae_compile",
|
| 12 |
+
"vae_core=256",
|
| 13 |
"vae_mode=default",
|
| 14 |
"w8a8"
|
| 15 |
],
|
| 16 |
+
"native_sha256": "118d64db97c3200ef785e83b8d703ebc3b480a799bb17a6affb104d1b4faa28d",
|
| 17 |
"repeat": 0,
|
| 18 |
+
"wall_seconds": 17.9887717962265,
|
| 19 |
+
"load_seconds": 2.2472441755235195,
|
| 20 |
"audio_seconds": 140.63866666666667,
|
| 21 |
+
"audio_per_wall": 7.818136127346303,
|
| 22 |
+
"abc_s": 3.3777515897527337,
|
| 23 |
"abc_tokens": 1249,
|
| 24 |
+
"sem_s": 10.058794901706278,
|
| 25 |
"sem_tokens": 3517,
|
| 26 |
+
"abc_ms_per_tok": 2.7043647636130776,
|
| 27 |
+
"sem_ms_per_tok": 2.860049730368575,
|
| 28 |
+
"nar_s": 2.5382185569033027,
|
| 29 |
+
"vae_s": 2.0050907684490085,
|
| 30 |
+
"nvml_peak_gib": 4.939697265625,
|
| 31 |
+
"torch_peak_gib": 2.735057830810547,
|
| 32 |
+
"torch_reserved_peak_gib": 3.908203125,
|
| 33 |
"semantic_sha256": "4c3a54b7d2d1a1be8bd05d35f4b773e1dc375533dcfe49694b1791126c3797a4",
|
| 34 |
+
"waveform_sha256": "71e4c0580fdd3ed99a0bfd5738f2482a0582ee0e66577625dc7d7c982a11182f",
|
| 35 |
+
"latent_sha256": "411c80bab4d16cac636006e231760df8697a8cad7457a8cdd14bb12df46a292d",
|
| 36 |
"timing": {
|
| 37 |
"abc": {
|
| 38 |
+
"seconds": 3.3777515897527337,
|
| 39 |
+
"prefill_seconds": 0.6063994308933616,
|
| 40 |
+
"ttft_seconds": 0.6074786949902773,
|
| 41 |
"output_tokens": 1249,
|
| 42 |
"content_tokens": 1248,
|
| 43 |
+
"output_tps": 369.77260370157427,
|
| 44 |
"prefix_tokens": 423,
|
| 45 |
"cfg_branches": 1,
|
| 46 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 51 |
"lookahead": true,
|
| 52 |
"unroll2": false,
|
| 53 |
"first_step_seconds": null,
|
| 54 |
+
"capture_seconds": 0.0399644561111927,
|
| 55 |
+
"close_seconds": 0.0002987785264849663
|
| 56 |
},
|
| 57 |
"semantic": {
|
| 58 |
+
"seconds": 10.058794901706278,
|
| 59 |
+
"prefill_seconds": 1.4141903733834624,
|
| 60 |
+
"ttft_seconds": 1.4142127567902207,
|
| 61 |
"output_tokens": 3517,
|
| 62 |
"content_tokens": 3516,
|
| 63 |
+
"output_tps": 349.6442699516032,
|
| 64 |
"prefix_tokens": 1673,
|
| 65 |
"cfg_branches": 2,
|
| 66 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 71 |
"lookahead": true,
|
| 72 |
"unroll2": false,
|
| 73 |
"first_step_seconds": null,
|
| 74 |
+
"capture_seconds": 0.2498519727960229,
|
| 75 |
+
"close_seconds": 0.00044487975537776947
|
| 76 |
},
|
| 77 |
+
"nar_seconds": 2.5382185569033027,
|
| 78 |
+
"vae_seconds": 2.0050907684490085,
|
| 79 |
"load": {
|
| 80 |
+
"resolve_and_integrity_seconds": 0.1945256581529975,
|
| 81 |
+
"mot_load_seconds": 0.7143985284492373
|
| 82 |
},
|
| 83 |
"stages": {
|
| 84 |
+
"t0->plan_done": 3.3831612318754196,
|
| 85 |
+
"plan_done->semantic_done": 10.060313027352095,
|
| 86 |
+
"semantic_done->nar_done": 2.538495653308928,
|
| 87 |
+
"nar_done->vae_done": 2.0053876154124737,
|
| 88 |
+
"plan_outside_timer": 0.005409642122685909,
|
| 89 |
+
"semantic_outside_timer": 0.0015181256458163261,
|
| 90 |
+
"semantic_close_seconds": 0.00044487975537776947,
|
| 91 |
"vae": {
|
| 92 |
+
"load_s": 0.027230621315538883,
|
| 93 |
+
"decode_s": 1.9777783742174506,
|
| 94 |
+
"post_s": 1.3129785656929016e-05,
|
| 95 |
"tiles": [
|
| 96 |
+
1.7687,
|
| 97 |
+
0.0138,
|
| 98 |
+
0.0136,
|
| 99 |
+
0.0135,
|
| 100 |
+
0.0137,
|
| 101 |
+
0.0139,
|
| 102 |
+
0.0139,
|
| 103 |
+
0.014,
|
| 104 |
+
0.0144,
|
| 105 |
+
0.0142,
|
| 106 |
+
0.0143,
|
| 107 |
+
0.0141,
|
| 108 |
+
0.0141,
|
| 109 |
+
0.0412
|
| 110 |
],
|
| 111 |
"cudnn_benchmark": false,
|
| 112 |
"cudnn_deterministic": true,
|
| 113 |
+
"core_frames": 256,
|
| 114 |
"input_device": "cpu",
|
| 115 |
"input_contiguous": false
|
| 116 |
},
|
| 117 |
"nar_prefill_source": "handoff",
|
| 118 |
+
"abc_close_seconds": 0.0002987785264849663,
|
| 119 |
+
"memory": {
|
| 120 |
+
"t0": {
|
| 121 |
+
"peak_gib": 2.479,
|
| 122 |
+
"alloc_gib": 2.479,
|
| 123 |
+
"reserved_gib": 2.756,
|
| 124 |
+
"nvml_used_gib": 3.674
|
| 125 |
+
},
|
| 126 |
+
"plan_done": {
|
| 127 |
+
"peak_gib": 2.754,
|
| 128 |
+
"alloc_gib": 2.479,
|
| 129 |
+
"reserved_gib": 2.98,
|
| 130 |
+
"nvml_used_gib": 3.922
|
| 131 |
+
},
|
| 132 |
+
"semantic_done": {
|
| 133 |
+
"peak_gib": 3.757,
|
| 134 |
+
"alloc_gib": 3.654,
|
| 135 |
+
"reserved_gib": 3.895,
|
| 136 |
+
"nvml_used_gib": 4.838
|
| 137 |
+
},
|
| 138 |
+
"nar_done": {
|
| 139 |
+
"peak_gib": 3.68,
|
| 140 |
+
"alloc_gib": 2.488,
|
| 141 |
+
"reserved_gib": 3.908,
|
| 142 |
+
"nvml_used_gib": 4.867
|
| 143 |
+
},
|
| 144 |
+
"vae_done": {
|
| 145 |
+
"peak_gib": 3.215,
|
| 146 |
+
"alloc_gib": 2.735,
|
| 147 |
+
"reserved_gib": 3.908,
|
| 148 |
+
"nvml_used_gib": 4.877
|
| 149 |
+
}
|
| 150 |
+
}
|
| 151 |
},
|
| 152 |
+
"e2e_seconds": 17.98769250139594
|
| 153 |
},
|
| 154 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 155 |
"cfg": 1.5,
|
|
|
|
| 159 |
"kv_cache_dtype": "int8",
|
| 160 |
"captured_sampling": true,
|
| 161 |
"fuse_rms_quant": true,
|
| 162 |
+
"orbitquant_version": "0.9.8",
|
| 163 |
"profile": "fast",
|
| 164 |
"allocator_backend": "native",
|
| 165 |
"modules": 224,
|
evaluation/profiles/fast-cfg1.5/run-00/plan.json
CHANGED
|
@@ -9,12 +9,12 @@
|
|
| 9 |
"id": "song"
|
| 10 |
},
|
| 11 |
"timing": {
|
| 12 |
-
"seconds":
|
| 13 |
-
"prefill_seconds":
|
| 14 |
-
"ttft_seconds":
|
| 15 |
"output_tokens": 1249,
|
| 16 |
"content_tokens": 1248,
|
| 17 |
-
"output_tps":
|
| 18 |
"prefix_tokens": 423,
|
| 19 |
"cfg_branches": 1,
|
| 20 |
"sampling": "combined_cuda_graph",
|
|
@@ -25,8 +25,8 @@
|
|
| 25 |
"lookahead": true,
|
| 26 |
"unroll2": false,
|
| 27 |
"first_step_seconds": null,
|
| 28 |
-
"capture_seconds": 0.
|
| 29 |
-
"close_seconds": 0.
|
| 30 |
},
|
| 31 |
"truncated": false,
|
| 32 |
"prefix": [
|
|
|
|
| 9 |
"id": "song"
|
| 10 |
},
|
| 11 |
"timing": {
|
| 12 |
+
"seconds": 3.3777515897527337,
|
| 13 |
+
"prefill_seconds": 0.6063994308933616,
|
| 14 |
+
"ttft_seconds": 0.6074786949902773,
|
| 15 |
"output_tokens": 1249,
|
| 16 |
"content_tokens": 1248,
|
| 17 |
+
"output_tps": 369.77260370157427,
|
| 18 |
"prefix_tokens": 423,
|
| 19 |
"cfg_branches": 1,
|
| 20 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 25 |
"lookahead": true,
|
| 26 |
"unroll2": false,
|
| 27 |
"first_step_seconds": null,
|
| 28 |
+
"capture_seconds": 0.0399644561111927,
|
| 29 |
+
"close_seconds": 0.0002987785264849663
|
| 30 |
},
|
| 31 |
"truncated": false,
|
| 32 |
"prefix": [
|
evaluation/profiles/fast-cfg1.5/run-00/plan_manifest.json
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
{
|
| 2 |
-
"plan.json": "
|
| 3 |
"abc_tokens.npy": "2a3f7d12e43955d2b4d6ec2c08f926be3a76a2fdf22a2952fa37148f6ca9c196",
|
| 4 |
"prefix.npy": "b542a03e835df9d2f05a4c2ffdaaf9eca5e2c49b510e51b6560f0d985952bf85",
|
| 5 |
"score.abc": "8d7ed2e1687e35072e41ff1b64b852068ff81d496bb8142d83019b53b66498ab"
|
|
|
|
| 1 |
{
|
| 2 |
+
"plan.json": "f22a94c8a3249590e6bb573c0f05fe1dbd6e6aa2eab6ee2faddb57c5180c86f1",
|
| 3 |
"abc_tokens.npy": "2a3f7d12e43955d2b4d6ec2c08f926be3a76a2fdf22a2952fa37148f6ca9c196",
|
| 4 |
"prefix.npy": "b542a03e835df9d2f05a4c2ffdaaf9eca5e2c49b510e51b6560f0d985952bf85",
|
| 5 |
"score.abc": "8d7ed2e1687e35072e41ff1b64b852068ff81d496bb8142d83019b53b66498ab"
|
evaluation/profiles/fast-cfg1.5/run-00/result.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"status": "complete",
|
| 3 |
-
"identity": "
|
| 4 |
"truncated": {
|
| 5 |
"abc": false,
|
| 6 |
"semantic": false
|
|
@@ -29,12 +29,12 @@
|
|
| 29 |
},
|
| 30 |
"timing": {
|
| 31 |
"abc": {
|
| 32 |
-
"seconds":
|
| 33 |
-
"prefill_seconds":
|
| 34 |
-
"ttft_seconds":
|
| 35 |
"output_tokens": 1249,
|
| 36 |
"content_tokens": 1248,
|
| 37 |
-
"output_tps":
|
| 38 |
"prefix_tokens": 423,
|
| 39 |
"cfg_branches": 1,
|
| 40 |
"sampling": "combined_cuda_graph",
|
|
@@ -45,16 +45,16 @@
|
|
| 45 |
"lookahead": true,
|
| 46 |
"unroll2": false,
|
| 47 |
"first_step_seconds": null,
|
| 48 |
-
"capture_seconds": 0.
|
| 49 |
-
"close_seconds": 0.
|
| 50 |
},
|
| 51 |
"semantic": {
|
| 52 |
-
"seconds":
|
| 53 |
-
"prefill_seconds":
|
| 54 |
-
"ttft_seconds":
|
| 55 |
"output_tokens": 3517,
|
| 56 |
"content_tokens": 3516,
|
| 57 |
-
"output_tps":
|
| 58 |
"prefix_tokens": 1673,
|
| 59 |
"cfg_branches": 2,
|
| 60 |
"sampling": "combined_cuda_graph",
|
|
@@ -65,46 +65,85 @@
|
|
| 65 |
"lookahead": true,
|
| 66 |
"unroll2": false,
|
| 67 |
"first_step_seconds": null,
|
| 68 |
-
"capture_seconds": 0.
|
| 69 |
-
"close_seconds": 0.
|
| 70 |
},
|
| 71 |
-
"nar_seconds":
|
| 72 |
-
"vae_seconds":
|
| 73 |
"load": {
|
| 74 |
-
"resolve_and_integrity_seconds":
|
| 75 |
-
"mot_load_seconds":
|
| 76 |
},
|
| 77 |
"stages": {
|
| 78 |
-
"t0->plan_done":
|
| 79 |
-
"plan_done->semantic_done":
|
| 80 |
-
"semantic_done->nar_done":
|
| 81 |
-
"nar_done->vae_done":
|
| 82 |
-
"plan_outside_timer": 0.
|
| 83 |
-
"semantic_outside_timer": 0.
|
| 84 |
-
"semantic_close_seconds": 0.
|
| 85 |
"vae": {
|
| 86 |
-
"load_s":
|
| 87 |
-
"decode_s":
|
| 88 |
-
"post_s":
|
| 89 |
"tiles": [
|
| 90 |
-
|
| 91 |
-
0.
|
| 92 |
-
0.
|
| 93 |
-
0.
|
| 94 |
-
0.
|
| 95 |
-
0.
|
| 96 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 97 |
],
|
| 98 |
"cudnn_benchmark": false,
|
| 99 |
"cudnn_deterministic": true,
|
| 100 |
-
"core_frames":
|
| 101 |
"input_device": "cpu",
|
| 102 |
"input_contiguous": false
|
| 103 |
},
|
| 104 |
"nar_prefill_source": "handoff",
|
| 105 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 106 |
},
|
| 107 |
-
"e2e_seconds":
|
| 108 |
},
|
| 109 |
"artifacts": {
|
| 110 |
"abc_tokens.npy": {
|
|
@@ -112,23 +151,23 @@
|
|
| 112 |
"bytes": 5120
|
| 113 |
},
|
| 114 |
"audio.flac": {
|
| 115 |
-
"sha256": "
|
| 116 |
-
"bytes":
|
| 117 |
},
|
| 118 |
"config.json": {
|
| 119 |
-
"sha256": "
|
| 120 |
"bytes": 2498
|
| 121 |
},
|
| 122 |
"latent.npy": {
|
| 123 |
-
"sha256": "
|
| 124 |
"bytes": 900224
|
| 125 |
},
|
| 126 |
"plan.json": {
|
| 127 |
-
"sha256": "
|
| 128 |
-
"bytes":
|
| 129 |
},
|
| 130 |
"plan_manifest.json": {
|
| 131 |
-
"sha256": "
|
| 132 |
"bytes": 341
|
| 133 |
},
|
| 134 |
"prefix.npy": {
|
|
|
|
| 1 |
{
|
| 2 |
"status": "complete",
|
| 3 |
+
"identity": "6577a328645c5543b5880aa7a6a718bf86ac6d552aa3179782f57a3673eab0b1",
|
| 4 |
"truncated": {
|
| 5 |
"abc": false,
|
| 6 |
"semantic": false
|
|
|
|
| 29 |
},
|
| 30 |
"timing": {
|
| 31 |
"abc": {
|
| 32 |
+
"seconds": 3.3777515897527337,
|
| 33 |
+
"prefill_seconds": 0.6063994308933616,
|
| 34 |
+
"ttft_seconds": 0.6074786949902773,
|
| 35 |
"output_tokens": 1249,
|
| 36 |
"content_tokens": 1248,
|
| 37 |
+
"output_tps": 369.77260370157427,
|
| 38 |
"prefix_tokens": 423,
|
| 39 |
"cfg_branches": 1,
|
| 40 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 45 |
"lookahead": true,
|
| 46 |
"unroll2": false,
|
| 47 |
"first_step_seconds": null,
|
| 48 |
+
"capture_seconds": 0.0399644561111927,
|
| 49 |
+
"close_seconds": 0.0002987785264849663
|
| 50 |
},
|
| 51 |
"semantic": {
|
| 52 |
+
"seconds": 10.058794901706278,
|
| 53 |
+
"prefill_seconds": 1.4141903733834624,
|
| 54 |
+
"ttft_seconds": 1.4142127567902207,
|
| 55 |
"output_tokens": 3517,
|
| 56 |
"content_tokens": 3516,
|
| 57 |
+
"output_tps": 349.6442699516032,
|
| 58 |
"prefix_tokens": 1673,
|
| 59 |
"cfg_branches": 2,
|
| 60 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 65 |
"lookahead": true,
|
| 66 |
"unroll2": false,
|
| 67 |
"first_step_seconds": null,
|
| 68 |
+
"capture_seconds": 0.2498519727960229,
|
| 69 |
+
"close_seconds": 0.00044487975537776947
|
| 70 |
},
|
| 71 |
+
"nar_seconds": 2.5382185569033027,
|
| 72 |
+
"vae_seconds": 2.0050907684490085,
|
| 73 |
"load": {
|
| 74 |
+
"resolve_and_integrity_seconds": 0.1945256581529975,
|
| 75 |
+
"mot_load_seconds": 0.7143985284492373
|
| 76 |
},
|
| 77 |
"stages": {
|
| 78 |
+
"t0->plan_done": 3.3831612318754196,
|
| 79 |
+
"plan_done->semantic_done": 10.060313027352095,
|
| 80 |
+
"semantic_done->nar_done": 2.538495653308928,
|
| 81 |
+
"nar_done->vae_done": 2.0053876154124737,
|
| 82 |
+
"plan_outside_timer": 0.005409642122685909,
|
| 83 |
+
"semantic_outside_timer": 0.0015181256458163261,
|
| 84 |
+
"semantic_close_seconds": 0.00044487975537776947,
|
| 85 |
"vae": {
|
| 86 |
+
"load_s": 0.027230621315538883,
|
| 87 |
+
"decode_s": 1.9777783742174506,
|
| 88 |
+
"post_s": 1.3129785656929016e-05,
|
| 89 |
"tiles": [
|
| 90 |
+
1.7687,
|
| 91 |
+
0.0138,
|
| 92 |
+
0.0136,
|
| 93 |
+
0.0135,
|
| 94 |
+
0.0137,
|
| 95 |
+
0.0139,
|
| 96 |
+
0.0139,
|
| 97 |
+
0.014,
|
| 98 |
+
0.0144,
|
| 99 |
+
0.0142,
|
| 100 |
+
0.0143,
|
| 101 |
+
0.0141,
|
| 102 |
+
0.0141,
|
| 103 |
+
0.0412
|
| 104 |
],
|
| 105 |
"cudnn_benchmark": false,
|
| 106 |
"cudnn_deterministic": true,
|
| 107 |
+
"core_frames": 256,
|
| 108 |
"input_device": "cpu",
|
| 109 |
"input_contiguous": false
|
| 110 |
},
|
| 111 |
"nar_prefill_source": "handoff",
|
| 112 |
+
"abc_close_seconds": 0.0002987785264849663,
|
| 113 |
+
"memory": {
|
| 114 |
+
"t0": {
|
| 115 |
+
"peak_gib": 2.479,
|
| 116 |
+
"alloc_gib": 2.479,
|
| 117 |
+
"reserved_gib": 2.756,
|
| 118 |
+
"nvml_used_gib": 3.674
|
| 119 |
+
},
|
| 120 |
+
"plan_done": {
|
| 121 |
+
"peak_gib": 2.754,
|
| 122 |
+
"alloc_gib": 2.479,
|
| 123 |
+
"reserved_gib": 2.98,
|
| 124 |
+
"nvml_used_gib": 3.922
|
| 125 |
+
},
|
| 126 |
+
"semantic_done": {
|
| 127 |
+
"peak_gib": 3.757,
|
| 128 |
+
"alloc_gib": 3.654,
|
| 129 |
+
"reserved_gib": 3.895,
|
| 130 |
+
"nvml_used_gib": 4.838
|
| 131 |
+
},
|
| 132 |
+
"nar_done": {
|
| 133 |
+
"peak_gib": 3.68,
|
| 134 |
+
"alloc_gib": 2.488,
|
| 135 |
+
"reserved_gib": 3.908,
|
| 136 |
+
"nvml_used_gib": 4.867
|
| 137 |
+
},
|
| 138 |
+
"vae_done": {
|
| 139 |
+
"peak_gib": 3.215,
|
| 140 |
+
"alloc_gib": 2.735,
|
| 141 |
+
"reserved_gib": 3.908,
|
| 142 |
+
"nvml_used_gib": 4.877
|
| 143 |
+
}
|
| 144 |
+
}
|
| 145 |
},
|
| 146 |
+
"e2e_seconds": 17.98769250139594
|
| 147 |
},
|
| 148 |
"artifacts": {
|
| 149 |
"abc_tokens.npy": {
|
|
|
|
| 151 |
"bytes": 5120
|
| 152 |
},
|
| 153 |
"audio.flac": {
|
| 154 |
+
"sha256": "2fb3d6deec41df9a2a691949c25242451fc53fcfc997b0fb38065d23809c79df",
|
| 155 |
+
"bytes": 29089984
|
| 156 |
},
|
| 157 |
"config.json": {
|
| 158 |
+
"sha256": "4584bf7a27383453ed3fec41d52c3ca5410e0c4ce7893eb0c18be0a6413f6af1",
|
| 159 |
"bytes": 2498
|
| 160 |
},
|
| 161 |
"latent.npy": {
|
| 162 |
+
"sha256": "dc67936c7daf305a24d3bd0d0e4fe63d109701484ad2d09bd6504df258b38f47",
|
| 163 |
"bytes": 900224
|
| 164 |
},
|
| 165 |
"plan.json": {
|
| 166 |
+
"sha256": "f22a94c8a3249590e6bb573c0f05fe1dbd6e6aa2eab6ee2faddb57c5180c86f1",
|
| 167 |
+
"bytes": 30337
|
| 168 |
},
|
| 169 |
"plan_manifest.json": {
|
| 170 |
+
"sha256": "ac1fab9a8b5e0377938251e4f97231c7b4c092fd3de0abf176b75272252c398d",
|
| 171 |
"bytes": 341
|
| 172 |
},
|
| 173 |
"prefix.npy": {
|
evaluation/profiles/fast-cfg1.5/run-01/config.json
CHANGED
|
@@ -35,12 +35,12 @@
|
|
| 35 |
"model_dtype": "bfloat16",
|
| 36 |
"vae_dtype": "float32",
|
| 37 |
"vae_decode": "halo_crop",
|
| 38 |
-
"vae_core_frames":
|
| 39 |
"vae_halo_frames": 16,
|
| 40 |
"device": "cuda:0",
|
| 41 |
"memory_budget_gib": 18.0,
|
| 42 |
"offload_ar": false,
|
| 43 |
-
"runtime_sha256": "
|
| 44 |
"decoder_release": "standard",
|
| 45 |
"validation_status": "unvalidated",
|
| 46 |
"runtime_profile": "fast",
|
|
@@ -48,7 +48,7 @@
|
|
| 48 |
"captured_sampling": true,
|
| 49 |
"kv_cache_dtype": "int8",
|
| 50 |
"int8_attention": "ordered",
|
| 51 |
-
"runtime_bundle_sha256": "
|
| 52 |
"vae_compute_dtype": "torch.float16",
|
| 53 |
"expert_staging": false,
|
| 54 |
"orbitquant": {
|
|
|
|
| 35 |
"model_dtype": "bfloat16",
|
| 36 |
"vae_dtype": "float32",
|
| 37 |
"vae_decode": "halo_crop",
|
| 38 |
+
"vae_core_frames": 256,
|
| 39 |
"vae_halo_frames": 16,
|
| 40 |
"device": "cuda:0",
|
| 41 |
"memory_budget_gib": 18.0,
|
| 42 |
"offload_ar": false,
|
| 43 |
+
"runtime_sha256": "bb189091c79b42dd3eca4f24f6239294df3c517fb2b149fc3b30ada3be287119",
|
| 44 |
"decoder_release": "standard",
|
| 45 |
"validation_status": "unvalidated",
|
| 46 |
"runtime_profile": "fast",
|
|
|
|
| 48 |
"captured_sampling": true,
|
| 49 |
"kv_cache_dtype": "int8",
|
| 50 |
"int8_attention": "ordered",
|
| 51 |
+
"runtime_bundle_sha256": "269214c6459739a0f08889532d5fc76f3e6f48f3908807466e14e5da4ecb1c98",
|
| 52 |
"vae_compute_dtype": "torch.float16",
|
| 53 |
"expert_staging": false,
|
| 54 |
"orbitquant": {
|
evaluation/profiles/fast-cfg1.5/run-01/metrics.json
CHANGED
|
@@ -1,45 +1,46 @@
|
|
| 1 |
{
|
| 2 |
"bundle": "v11",
|
| 3 |
-
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,kv_handoff,
|
| 4 |
"flags": [
|
| 5 |
"abc_crop",
|
| 6 |
"int8_heads",
|
| 7 |
"kv_handoff",
|
|
|
|
| 8 |
"lookahead",
|
| 9 |
-
"nar_compile",
|
| 10 |
"resident",
|
| 11 |
"vae_compile",
|
|
|
|
| 12 |
"vae_mode=default",
|
| 13 |
"w8a8"
|
| 14 |
],
|
| 15 |
-
"native_sha256": "
|
| 16 |
"repeat": 1,
|
| 17 |
-
"wall_seconds":
|
| 18 |
-
"load_seconds":
|
| 19 |
"audio_seconds": 140.63866666666667,
|
| 20 |
-
"audio_per_wall":
|
| 21 |
-
"abc_s":
|
| 22 |
"abc_tokens": 1249,
|
| 23 |
-
"sem_s":
|
| 24 |
"sem_tokens": 3517,
|
| 25 |
-
"abc_ms_per_tok": 2.
|
| 26 |
-
"sem_ms_per_tok": 2.
|
| 27 |
-
"nar_s":
|
| 28 |
-
"vae_s": 0.
|
| 29 |
-
"nvml_peak_gib": 5.
|
| 30 |
-
"torch_peak_gib":
|
| 31 |
-
"torch_reserved_peak_gib": 4.
|
| 32 |
"semantic_sha256": "4c3a54b7d2d1a1be8bd05d35f4b773e1dc375533dcfe49694b1791126c3797a4",
|
| 33 |
-
"waveform_sha256": "
|
| 34 |
-
"latent_sha256": "
|
| 35 |
"timing": {
|
| 36 |
"abc": {
|
| 37 |
-
"seconds":
|
| 38 |
-
"prefill_seconds": 0.
|
| 39 |
-
"ttft_seconds": 0.
|
| 40 |
"output_tokens": 1249,
|
| 41 |
"content_tokens": 1248,
|
| 42 |
-
"output_tps":
|
| 43 |
"prefix_tokens": 423,
|
| 44 |
"cfg_branches": 1,
|
| 45 |
"sampling": "combined_cuda_graph",
|
|
@@ -50,16 +51,16 @@
|
|
| 50 |
"lookahead": true,
|
| 51 |
"unroll2": false,
|
| 52 |
"first_step_seconds": null,
|
| 53 |
-
"capture_seconds": 0.
|
| 54 |
-
"close_seconds": 0.
|
| 55 |
},
|
| 56 |
"semantic": {
|
| 57 |
-
"seconds":
|
| 58 |
-
"prefill_seconds": 0.
|
| 59 |
-
"ttft_seconds": 0.
|
| 60 |
"output_tokens": 3517,
|
| 61 |
"content_tokens": 3516,
|
| 62 |
-
"output_tps":
|
| 63 |
"prefix_tokens": 1673,
|
| 64 |
"cfg_branches": 2,
|
| 65 |
"sampling": "combined_cuda_graph",
|
|
@@ -70,46 +71,85 @@
|
|
| 70 |
"lookahead": true,
|
| 71 |
"unroll2": false,
|
| 72 |
"first_step_seconds": null,
|
| 73 |
-
"capture_seconds": 0.
|
| 74 |
-
"close_seconds": 0.
|
| 75 |
},
|
| 76 |
-
"nar_seconds":
|
| 77 |
-
"vae_seconds": 0.
|
| 78 |
"load": {
|
| 79 |
-
"resolve_and_integrity_seconds":
|
| 80 |
-
"mot_load_seconds":
|
| 81 |
},
|
| 82 |
"stages": {
|
| 83 |
-
"t0->plan_done":
|
| 84 |
-
"plan_done->semantic_done":
|
| 85 |
-
"semantic_done->nar_done":
|
| 86 |
-
"nar_done->vae_done": 0.
|
| 87 |
-
"plan_outside_timer": 0.
|
| 88 |
-
"semantic_outside_timer": 0.
|
| 89 |
-
"semantic_close_seconds": 0.
|
| 90 |
"vae": {
|
| 91 |
-
"load_s": 0.
|
| 92 |
-
"decode_s": 0.
|
| 93 |
-
"post_s":
|
| 94 |
"tiles": [
|
| 95 |
-
0.
|
| 96 |
-
0.
|
| 97 |
-
0.
|
| 98 |
-
0.
|
| 99 |
-
0.
|
| 100 |
-
0.
|
| 101 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 102 |
],
|
| 103 |
"cudnn_benchmark": false,
|
| 104 |
"cudnn_deterministic": true,
|
| 105 |
-
"core_frames":
|
| 106 |
"input_device": "cpu",
|
| 107 |
"input_contiguous": false
|
| 108 |
},
|
| 109 |
"nar_prefill_source": "handoff",
|
| 110 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 111 |
},
|
| 112 |
-
"e2e_seconds":
|
| 113 |
},
|
| 114 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 115 |
"cfg": 1.5,
|
|
@@ -119,7 +159,7 @@
|
|
| 119 |
"kv_cache_dtype": "int8",
|
| 120 |
"captured_sampling": true,
|
| 121 |
"fuse_rms_quant": true,
|
| 122 |
-
"orbitquant_version": "0.9.
|
| 123 |
"profile": "fast",
|
| 124 |
"allocator_backend": "native",
|
| 125 |
"modules": 224,
|
|
|
|
| 1 |
{
|
| 2 |
"bundle": "v11",
|
| 3 |
+
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,vae_core=256,kv_handoff,lazy_handoff,int8_heads,w8a8",
|
| 4 |
"flags": [
|
| 5 |
"abc_crop",
|
| 6 |
"int8_heads",
|
| 7 |
"kv_handoff",
|
| 8 |
+
"lazy_handoff",
|
| 9 |
"lookahead",
|
|
|
|
| 10 |
"resident",
|
| 11 |
"vae_compile",
|
| 12 |
+
"vae_core=256",
|
| 13 |
"vae_mode=default",
|
| 14 |
"w8a8"
|
| 15 |
],
|
| 16 |
+
"native_sha256": "118d64db97c3200ef785e83b8d703ebc3b480a799bb17a6affb104d1b4faa28d",
|
| 17 |
"repeat": 1,
|
| 18 |
+
"wall_seconds": 13.461460947990417,
|
| 19 |
+
"load_seconds": 2.2472441755235195,
|
| 20 |
"audio_seconds": 140.63866666666667,
|
| 21 |
+
"audio_per_wall": 10.447503967811294,
|
| 22 |
+
"abc_s": 2.8130570147186518,
|
| 23 |
"abc_tokens": 1249,
|
| 24 |
+
"sem_s": 8.530152230523527,
|
| 25 |
"sem_tokens": 3517,
|
| 26 |
+
"abc_ms_per_tok": 2.2522474097026834,
|
| 27 |
+
"sem_ms_per_tok": 2.4254058090769197,
|
| 28 |
+
"nar_s": 1.906809726729989,
|
| 29 |
+
"vae_s": 0.2063934374600649,
|
| 30 |
+
"nvml_peak_gib": 5.199462890625,
|
| 31 |
+
"torch_peak_gib": 2.735057830810547,
|
| 32 |
+
"torch_reserved_peak_gib": 4.123046875,
|
| 33 |
"semantic_sha256": "4c3a54b7d2d1a1be8bd05d35f4b773e1dc375533dcfe49694b1791126c3797a4",
|
| 34 |
+
"waveform_sha256": "71e4c0580fdd3ed99a0bfd5738f2482a0582ee0e66577625dc7d7c982a11182f",
|
| 35 |
+
"latent_sha256": "411c80bab4d16cac636006e231760df8697a8cad7457a8cdd14bb12df46a292d",
|
| 36 |
"timing": {
|
| 37 |
"abc": {
|
| 38 |
+
"seconds": 2.8130570147186518,
|
| 39 |
+
"prefill_seconds": 0.04717915412038565,
|
| 40 |
+
"ttft_seconds": 0.047200084663927555,
|
| 41 |
"output_tokens": 1249,
|
| 42 |
"content_tokens": 1248,
|
| 43 |
+
"output_tps": 444.00095464290433,
|
| 44 |
"prefix_tokens": 423,
|
| 45 |
"cfg_branches": 1,
|
| 46 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 51 |
"lookahead": true,
|
| 52 |
"unroll2": false,
|
| 53 |
"first_step_seconds": null,
|
| 54 |
+
"capture_seconds": 0.033918716944754124,
|
| 55 |
+
"close_seconds": 0.00039780884981155396
|
| 56 |
},
|
| 57 |
"semantic": {
|
| 58 |
+
"seconds": 8.530152230523527,
|
| 59 |
+
"prefill_seconds": 0.08340269047766924,
|
| 60 |
+
"ttft_seconds": 0.08342811837792397,
|
| 61 |
"output_tokens": 3517,
|
| 62 |
"content_tokens": 3516,
|
| 63 |
+
"output_tps": 412.3021377526048,
|
| 64 |
"prefix_tokens": 1673,
|
| 65 |
"cfg_branches": 2,
|
| 66 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 71 |
"lookahead": true,
|
| 72 |
"unroll2": false,
|
| 73 |
"first_step_seconds": null,
|
| 74 |
+
"capture_seconds": 0.03369541093707085,
|
| 75 |
+
"close_seconds": 0.00037660542875528336
|
| 76 |
},
|
| 77 |
+
"nar_seconds": 1.906809726729989,
|
| 78 |
+
"vae_seconds": 0.2063934374600649,
|
| 79 |
"load": {
|
| 80 |
+
"resolve_and_integrity_seconds": 0.1945256581529975,
|
| 81 |
+
"mot_load_seconds": 0.7143985284492373
|
| 82 |
},
|
| 83 |
"stages": {
|
| 84 |
+
"t0->plan_done": 2.814796471968293,
|
| 85 |
+
"plan_done->semantic_done": 8.531699027866125,
|
| 86 |
+
"semantic_done->nar_done": 1.9072276363149285,
|
| 87 |
+
"nar_done->vae_done": 0.2066958835348487,
|
| 88 |
+
"plan_outside_timer": 0.0017394572496414185,
|
| 89 |
+
"semantic_outside_timer": 0.0015467973425984383,
|
| 90 |
+
"semantic_close_seconds": 0.00037660542875528336,
|
| 91 |
"vae": {
|
| 92 |
+
"load_s": 0.0009437203407287598,
|
| 93 |
+
"decode_s": 0.20538921747356653,
|
| 94 |
+
"post_s": 1.1157244443893433e-05,
|
| 95 |
"tiles": [
|
| 96 |
+
0.0144,
|
| 97 |
+
0.0138,
|
| 98 |
+
0.0139,
|
| 99 |
+
0.0139,
|
| 100 |
+
0.0137,
|
| 101 |
+
0.0138,
|
| 102 |
+
0.0138,
|
| 103 |
+
0.0137,
|
| 104 |
+
0.0137,
|
| 105 |
+
0.0137,
|
| 106 |
+
0.0136,
|
| 107 |
+
0.0136,
|
| 108 |
+
0.0137,
|
| 109 |
+
0.026
|
| 110 |
],
|
| 111 |
"cudnn_benchmark": false,
|
| 112 |
"cudnn_deterministic": true,
|
| 113 |
+
"core_frames": 256,
|
| 114 |
"input_device": "cpu",
|
| 115 |
"input_contiguous": false
|
| 116 |
},
|
| 117 |
"nar_prefill_source": "handoff",
|
| 118 |
+
"abc_close_seconds": 0.00039780884981155396,
|
| 119 |
+
"memory": {
|
| 120 |
+
"t0": {
|
| 121 |
+
"peak_gib": 2.735,
|
| 122 |
+
"alloc_gib": 2.735,
|
| 123 |
+
"reserved_gib": 3.908,
|
| 124 |
+
"nvml_used_gib": 4.877
|
| 125 |
+
},
|
| 126 |
+
"plan_done": {
|
| 127 |
+
"peak_gib": 3.01,
|
| 128 |
+
"alloc_gib": 2.735,
|
| 129 |
+
"reserved_gib": 3.215,
|
| 130 |
+
"nvml_used_gib": 4.184
|
| 131 |
+
},
|
| 132 |
+
"semantic_done": {
|
| 133 |
+
"peak_gib": 4.014,
|
| 134 |
+
"alloc_gib": 3.911,
|
| 135 |
+
"reserved_gib": 4.109,
|
| 136 |
+
"nvml_used_gib": 5.078
|
| 137 |
+
},
|
| 138 |
+
"nar_done": {
|
| 139 |
+
"peak_gib": 3.936,
|
| 140 |
+
"alloc_gib": 2.735,
|
| 141 |
+
"reserved_gib": 4.123,
|
| 142 |
+
"nvml_used_gib": 5.092
|
| 143 |
+
},
|
| 144 |
+
"vae_done": {
|
| 145 |
+
"peak_gib": 3.215,
|
| 146 |
+
"alloc_gib": 2.735,
|
| 147 |
+
"reserved_gib": 4.123,
|
| 148 |
+
"nvml_used_gib": 5.092
|
| 149 |
+
}
|
| 150 |
+
}
|
| 151 |
},
|
| 152 |
+
"e2e_seconds": 13.460732371546328
|
| 153 |
},
|
| 154 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 155 |
"cfg": 1.5,
|
|
|
|
| 159 |
"kv_cache_dtype": "int8",
|
| 160 |
"captured_sampling": true,
|
| 161 |
"fuse_rms_quant": true,
|
| 162 |
+
"orbitquant_version": "0.9.8",
|
| 163 |
"profile": "fast",
|
| 164 |
"allocator_backend": "native",
|
| 165 |
"modules": 224,
|
evaluation/profiles/fast-cfg1.5/run-01/plan.json
CHANGED
|
@@ -9,12 +9,12 @@
|
|
| 9 |
"id": "song"
|
| 10 |
},
|
| 11 |
"timing": {
|
| 12 |
-
"seconds":
|
| 13 |
-
"prefill_seconds": 0.
|
| 14 |
-
"ttft_seconds": 0.
|
| 15 |
"output_tokens": 1249,
|
| 16 |
"content_tokens": 1248,
|
| 17 |
-
"output_tps":
|
| 18 |
"prefix_tokens": 423,
|
| 19 |
"cfg_branches": 1,
|
| 20 |
"sampling": "combined_cuda_graph",
|
|
@@ -25,8 +25,8 @@
|
|
| 25 |
"lookahead": true,
|
| 26 |
"unroll2": false,
|
| 27 |
"first_step_seconds": null,
|
| 28 |
-
"capture_seconds": 0.
|
| 29 |
-
"close_seconds": 0.
|
| 30 |
},
|
| 31 |
"truncated": false,
|
| 32 |
"prefix": [
|
|
|
|
| 9 |
"id": "song"
|
| 10 |
},
|
| 11 |
"timing": {
|
| 12 |
+
"seconds": 2.8130570147186518,
|
| 13 |
+
"prefill_seconds": 0.04717915412038565,
|
| 14 |
+
"ttft_seconds": 0.047200084663927555,
|
| 15 |
"output_tokens": 1249,
|
| 16 |
"content_tokens": 1248,
|
| 17 |
+
"output_tps": 444.00095464290433,
|
| 18 |
"prefix_tokens": 423,
|
| 19 |
"cfg_branches": 1,
|
| 20 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 25 |
"lookahead": true,
|
| 26 |
"unroll2": false,
|
| 27 |
"first_step_seconds": null,
|
| 28 |
+
"capture_seconds": 0.033918716944754124,
|
| 29 |
+
"close_seconds": 0.00039780884981155396
|
| 30 |
},
|
| 31 |
"truncated": false,
|
| 32 |
"prefix": [
|
evaluation/profiles/fast-cfg1.5/run-01/plan_manifest.json
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
{
|
| 2 |
-
"plan.json": "
|
| 3 |
"abc_tokens.npy": "2a3f7d12e43955d2b4d6ec2c08f926be3a76a2fdf22a2952fa37148f6ca9c196",
|
| 4 |
"prefix.npy": "b542a03e835df9d2f05a4c2ffdaaf9eca5e2c49b510e51b6560f0d985952bf85",
|
| 5 |
"score.abc": "8d7ed2e1687e35072e41ff1b64b852068ff81d496bb8142d83019b53b66498ab"
|
|
|
|
| 1 |
{
|
| 2 |
+
"plan.json": "500905e85275ec425ab6f4e717b6a74e56e9584ca12fea060d9f9136e7b9f474",
|
| 3 |
"abc_tokens.npy": "2a3f7d12e43955d2b4d6ec2c08f926be3a76a2fdf22a2952fa37148f6ca9c196",
|
| 4 |
"prefix.npy": "b542a03e835df9d2f05a4c2ffdaaf9eca5e2c49b510e51b6560f0d985952bf85",
|
| 5 |
"score.abc": "8d7ed2e1687e35072e41ff1b64b852068ff81d496bb8142d83019b53b66498ab"
|
evaluation/profiles/fast-cfg1.5/run-01/result.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"status": "complete",
|
| 3 |
-
"identity": "
|
| 4 |
"truncated": {
|
| 5 |
"abc": false,
|
| 6 |
"semantic": false
|
|
@@ -29,12 +29,12 @@
|
|
| 29 |
},
|
| 30 |
"timing": {
|
| 31 |
"abc": {
|
| 32 |
-
"seconds":
|
| 33 |
-
"prefill_seconds": 0.
|
| 34 |
-
"ttft_seconds": 0.
|
| 35 |
"output_tokens": 1249,
|
| 36 |
"content_tokens": 1248,
|
| 37 |
-
"output_tps":
|
| 38 |
"prefix_tokens": 423,
|
| 39 |
"cfg_branches": 1,
|
| 40 |
"sampling": "combined_cuda_graph",
|
|
@@ -45,16 +45,16 @@
|
|
| 45 |
"lookahead": true,
|
| 46 |
"unroll2": false,
|
| 47 |
"first_step_seconds": null,
|
| 48 |
-
"capture_seconds": 0.
|
| 49 |
-
"close_seconds": 0.
|
| 50 |
},
|
| 51 |
"semantic": {
|
| 52 |
-
"seconds":
|
| 53 |
-
"prefill_seconds": 0.
|
| 54 |
-
"ttft_seconds": 0.
|
| 55 |
"output_tokens": 3517,
|
| 56 |
"content_tokens": 3516,
|
| 57 |
-
"output_tps":
|
| 58 |
"prefix_tokens": 1673,
|
| 59 |
"cfg_branches": 2,
|
| 60 |
"sampling": "combined_cuda_graph",
|
|
@@ -65,46 +65,85 @@
|
|
| 65 |
"lookahead": true,
|
| 66 |
"unroll2": false,
|
| 67 |
"first_step_seconds": null,
|
| 68 |
-
"capture_seconds": 0.
|
| 69 |
-
"close_seconds": 0.
|
| 70 |
},
|
| 71 |
-
"nar_seconds":
|
| 72 |
-
"vae_seconds": 0.
|
| 73 |
"load": {
|
| 74 |
-
"resolve_and_integrity_seconds":
|
| 75 |
-
"mot_load_seconds":
|
| 76 |
},
|
| 77 |
"stages": {
|
| 78 |
-
"t0->plan_done":
|
| 79 |
-
"plan_done->semantic_done":
|
| 80 |
-
"semantic_done->nar_done":
|
| 81 |
-
"nar_done->vae_done": 0.
|
| 82 |
-
"plan_outside_timer": 0.
|
| 83 |
-
"semantic_outside_timer": 0.
|
| 84 |
-
"semantic_close_seconds": 0.
|
| 85 |
"vae": {
|
| 86 |
-
"load_s": 0.
|
| 87 |
-
"decode_s": 0.
|
| 88 |
-
"post_s":
|
| 89 |
"tiles": [
|
| 90 |
-
0.
|
| 91 |
-
0.
|
| 92 |
-
0.
|
| 93 |
-
0.
|
| 94 |
-
0.
|
| 95 |
-
0.
|
| 96 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 97 |
],
|
| 98 |
"cudnn_benchmark": false,
|
| 99 |
"cudnn_deterministic": true,
|
| 100 |
-
"core_frames":
|
| 101 |
"input_device": "cpu",
|
| 102 |
"input_contiguous": false
|
| 103 |
},
|
| 104 |
"nar_prefill_source": "handoff",
|
| 105 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 106 |
},
|
| 107 |
-
"e2e_seconds":
|
| 108 |
},
|
| 109 |
"artifacts": {
|
| 110 |
"abc_tokens.npy": {
|
|
@@ -112,23 +151,23 @@
|
|
| 112 |
"bytes": 5120
|
| 113 |
},
|
| 114 |
"audio.flac": {
|
| 115 |
-
"sha256": "
|
| 116 |
-
"bytes":
|
| 117 |
},
|
| 118 |
"config.json": {
|
| 119 |
-
"sha256": "
|
| 120 |
"bytes": 2498
|
| 121 |
},
|
| 122 |
"latent.npy": {
|
| 123 |
-
"sha256": "
|
| 124 |
"bytes": 900224
|
| 125 |
},
|
| 126 |
"plan.json": {
|
| 127 |
-
"sha256": "
|
| 128 |
-
"bytes":
|
| 129 |
},
|
| 130 |
"plan_manifest.json": {
|
| 131 |
-
"sha256": "
|
| 132 |
"bytes": 341
|
| 133 |
},
|
| 134 |
"prefix.npy": {
|
|
|
|
| 1 |
{
|
| 2 |
"status": "complete",
|
| 3 |
+
"identity": "6577a328645c5543b5880aa7a6a718bf86ac6d552aa3179782f57a3673eab0b1",
|
| 4 |
"truncated": {
|
| 5 |
"abc": false,
|
| 6 |
"semantic": false
|
|
|
|
| 29 |
},
|
| 30 |
"timing": {
|
| 31 |
"abc": {
|
| 32 |
+
"seconds": 2.8130570147186518,
|
| 33 |
+
"prefill_seconds": 0.04717915412038565,
|
| 34 |
+
"ttft_seconds": 0.047200084663927555,
|
| 35 |
"output_tokens": 1249,
|
| 36 |
"content_tokens": 1248,
|
| 37 |
+
"output_tps": 444.00095464290433,
|
| 38 |
"prefix_tokens": 423,
|
| 39 |
"cfg_branches": 1,
|
| 40 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 45 |
"lookahead": true,
|
| 46 |
"unroll2": false,
|
| 47 |
"first_step_seconds": null,
|
| 48 |
+
"capture_seconds": 0.033918716944754124,
|
| 49 |
+
"close_seconds": 0.00039780884981155396
|
| 50 |
},
|
| 51 |
"semantic": {
|
| 52 |
+
"seconds": 8.530152230523527,
|
| 53 |
+
"prefill_seconds": 0.08340269047766924,
|
| 54 |
+
"ttft_seconds": 0.08342811837792397,
|
| 55 |
"output_tokens": 3517,
|
| 56 |
"content_tokens": 3516,
|
| 57 |
+
"output_tps": 412.3021377526048,
|
| 58 |
"prefix_tokens": 1673,
|
| 59 |
"cfg_branches": 2,
|
| 60 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 65 |
"lookahead": true,
|
| 66 |
"unroll2": false,
|
| 67 |
"first_step_seconds": null,
|
| 68 |
+
"capture_seconds": 0.03369541093707085,
|
| 69 |
+
"close_seconds": 0.00037660542875528336
|
| 70 |
},
|
| 71 |
+
"nar_seconds": 1.906809726729989,
|
| 72 |
+
"vae_seconds": 0.2063934374600649,
|
| 73 |
"load": {
|
| 74 |
+
"resolve_and_integrity_seconds": 0.1945256581529975,
|
| 75 |
+
"mot_load_seconds": 0.7143985284492373
|
| 76 |
},
|
| 77 |
"stages": {
|
| 78 |
+
"t0->plan_done": 2.814796471968293,
|
| 79 |
+
"plan_done->semantic_done": 8.531699027866125,
|
| 80 |
+
"semantic_done->nar_done": 1.9072276363149285,
|
| 81 |
+
"nar_done->vae_done": 0.2066958835348487,
|
| 82 |
+
"plan_outside_timer": 0.0017394572496414185,
|
| 83 |
+
"semantic_outside_timer": 0.0015467973425984383,
|
| 84 |
+
"semantic_close_seconds": 0.00037660542875528336,
|
| 85 |
"vae": {
|
| 86 |
+
"load_s": 0.0009437203407287598,
|
| 87 |
+
"decode_s": 0.20538921747356653,
|
| 88 |
+
"post_s": 1.1157244443893433e-05,
|
| 89 |
"tiles": [
|
| 90 |
+
0.0144,
|
| 91 |
+
0.0138,
|
| 92 |
+
0.0139,
|
| 93 |
+
0.0139,
|
| 94 |
+
0.0137,
|
| 95 |
+
0.0138,
|
| 96 |
+
0.0138,
|
| 97 |
+
0.0137,
|
| 98 |
+
0.0137,
|
| 99 |
+
0.0137,
|
| 100 |
+
0.0136,
|
| 101 |
+
0.0136,
|
| 102 |
+
0.0137,
|
| 103 |
+
0.026
|
| 104 |
],
|
| 105 |
"cudnn_benchmark": false,
|
| 106 |
"cudnn_deterministic": true,
|
| 107 |
+
"core_frames": 256,
|
| 108 |
"input_device": "cpu",
|
| 109 |
"input_contiguous": false
|
| 110 |
},
|
| 111 |
"nar_prefill_source": "handoff",
|
| 112 |
+
"abc_close_seconds": 0.00039780884981155396,
|
| 113 |
+
"memory": {
|
| 114 |
+
"t0": {
|
| 115 |
+
"peak_gib": 2.735,
|
| 116 |
+
"alloc_gib": 2.735,
|
| 117 |
+
"reserved_gib": 3.908,
|
| 118 |
+
"nvml_used_gib": 4.877
|
| 119 |
+
},
|
| 120 |
+
"plan_done": {
|
| 121 |
+
"peak_gib": 3.01,
|
| 122 |
+
"alloc_gib": 2.735,
|
| 123 |
+
"reserved_gib": 3.215,
|
| 124 |
+
"nvml_used_gib": 4.184
|
| 125 |
+
},
|
| 126 |
+
"semantic_done": {
|
| 127 |
+
"peak_gib": 4.014,
|
| 128 |
+
"alloc_gib": 3.911,
|
| 129 |
+
"reserved_gib": 4.109,
|
| 130 |
+
"nvml_used_gib": 5.078
|
| 131 |
+
},
|
| 132 |
+
"nar_done": {
|
| 133 |
+
"peak_gib": 3.936,
|
| 134 |
+
"alloc_gib": 2.735,
|
| 135 |
+
"reserved_gib": 4.123,
|
| 136 |
+
"nvml_used_gib": 5.092
|
| 137 |
+
},
|
| 138 |
+
"vae_done": {
|
| 139 |
+
"peak_gib": 3.215,
|
| 140 |
+
"alloc_gib": 2.735,
|
| 141 |
+
"reserved_gib": 4.123,
|
| 142 |
+
"nvml_used_gib": 5.092
|
| 143 |
+
}
|
| 144 |
+
}
|
| 145 |
},
|
| 146 |
+
"e2e_seconds": 13.460732371546328
|
| 147 |
},
|
| 148 |
"artifacts": {
|
| 149 |
"abc_tokens.npy": {
|
|
|
|
| 151 |
"bytes": 5120
|
| 152 |
},
|
| 153 |
"audio.flac": {
|
| 154 |
+
"sha256": "2fb3d6deec41df9a2a691949c25242451fc53fcfc997b0fb38065d23809c79df",
|
| 155 |
+
"bytes": 29089984
|
| 156 |
},
|
| 157 |
"config.json": {
|
| 158 |
+
"sha256": "4584bf7a27383453ed3fec41d52c3ca5410e0c4ce7893eb0c18be0a6413f6af1",
|
| 159 |
"bytes": 2498
|
| 160 |
},
|
| 161 |
"latent.npy": {
|
| 162 |
+
"sha256": "dc67936c7daf305a24d3bd0d0e4fe63d109701484ad2d09bd6504df258b38f47",
|
| 163 |
"bytes": 900224
|
| 164 |
},
|
| 165 |
"plan.json": {
|
| 166 |
+
"sha256": "500905e85275ec425ab6f4e717b6a74e56e9584ca12fea060d9f9136e7b9f474",
|
| 167 |
+
"bytes": 30343
|
| 168 |
},
|
| 169 |
"plan_manifest.json": {
|
| 170 |
+
"sha256": "c48c7a04343d2998767c80f020bf5cf0d1ca767d0586d8a0dc5ebfc1d226681d",
|
| 171 |
"bytes": 341
|
| 172 |
},
|
| 173 |
"prefix.npy": {
|
evaluation/profiles/fast-cfg1.5/summary.json
CHANGED
|
@@ -1,46 +1,47 @@
|
|
| 1 |
[
|
| 2 |
{
|
| 3 |
"bundle": "v11",
|
| 4 |
-
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,kv_handoff,
|
| 5 |
"flags": [
|
| 6 |
"abc_crop",
|
| 7 |
"int8_heads",
|
| 8 |
"kv_handoff",
|
|
|
|
| 9 |
"lookahead",
|
| 10 |
-
"nar_compile",
|
| 11 |
"resident",
|
| 12 |
"vae_compile",
|
|
|
|
| 13 |
"vae_mode=default",
|
| 14 |
"w8a8"
|
| 15 |
],
|
| 16 |
-
"native_sha256": "
|
| 17 |
"repeat": 0,
|
| 18 |
-
"wall_seconds":
|
| 19 |
-
"load_seconds":
|
| 20 |
"audio_seconds": 140.63866666666667,
|
| 21 |
-
"audio_per_wall":
|
| 22 |
-
"abc_s":
|
| 23 |
"abc_tokens": 1249,
|
| 24 |
-
"sem_s":
|
| 25 |
"sem_tokens": 3517,
|
| 26 |
-
"abc_ms_per_tok":
|
| 27 |
-
"sem_ms_per_tok": 2.
|
| 28 |
-
"nar_s":
|
| 29 |
-
"vae_s":
|
| 30 |
-
"nvml_peak_gib":
|
| 31 |
-
"torch_peak_gib":
|
| 32 |
-
"torch_reserved_peak_gib":
|
| 33 |
"semantic_sha256": "4c3a54b7d2d1a1be8bd05d35f4b773e1dc375533dcfe49694b1791126c3797a4",
|
| 34 |
-
"waveform_sha256": "
|
| 35 |
-
"latent_sha256": "
|
| 36 |
"timing": {
|
| 37 |
"abc": {
|
| 38 |
-
"seconds":
|
| 39 |
-
"prefill_seconds":
|
| 40 |
-
"ttft_seconds":
|
| 41 |
"output_tokens": 1249,
|
| 42 |
"content_tokens": 1248,
|
| 43 |
-
"output_tps":
|
| 44 |
"prefix_tokens": 423,
|
| 45 |
"cfg_branches": 1,
|
| 46 |
"sampling": "combined_cuda_graph",
|
|
@@ -51,16 +52,16 @@
|
|
| 51 |
"lookahead": true,
|
| 52 |
"unroll2": false,
|
| 53 |
"first_step_seconds": null,
|
| 54 |
-
"capture_seconds": 0.
|
| 55 |
-
"close_seconds": 0.
|
| 56 |
},
|
| 57 |
"semantic": {
|
| 58 |
-
"seconds":
|
| 59 |
-
"prefill_seconds":
|
| 60 |
-
"ttft_seconds":
|
| 61 |
"output_tokens": 3517,
|
| 62 |
"content_tokens": 3516,
|
| 63 |
-
"output_tps":
|
| 64 |
"prefix_tokens": 1673,
|
| 65 |
"cfg_branches": 2,
|
| 66 |
"sampling": "combined_cuda_graph",
|
|
@@ -71,46 +72,85 @@
|
|
| 71 |
"lookahead": true,
|
| 72 |
"unroll2": false,
|
| 73 |
"first_step_seconds": null,
|
| 74 |
-
"capture_seconds": 0.
|
| 75 |
-
"close_seconds": 0.
|
| 76 |
},
|
| 77 |
-
"nar_seconds":
|
| 78 |
-
"vae_seconds":
|
| 79 |
"load": {
|
| 80 |
-
"resolve_and_integrity_seconds":
|
| 81 |
-
"mot_load_seconds":
|
| 82 |
},
|
| 83 |
"stages": {
|
| 84 |
-
"t0->plan_done":
|
| 85 |
-
"plan_done->semantic_done":
|
| 86 |
-
"semantic_done->nar_done":
|
| 87 |
-
"nar_done->vae_done":
|
| 88 |
-
"plan_outside_timer": 0.
|
| 89 |
-
"semantic_outside_timer": 0.
|
| 90 |
-
"semantic_close_seconds": 0.
|
| 91 |
"vae": {
|
| 92 |
-
"load_s":
|
| 93 |
-
"decode_s":
|
| 94 |
-
"post_s":
|
| 95 |
"tiles": [
|
| 96 |
-
|
| 97 |
-
0.
|
| 98 |
-
0.
|
| 99 |
-
0.
|
| 100 |
-
0.
|
| 101 |
-
0.
|
| 102 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 103 |
],
|
| 104 |
"cudnn_benchmark": false,
|
| 105 |
"cudnn_deterministic": true,
|
| 106 |
-
"core_frames":
|
| 107 |
"input_device": "cpu",
|
| 108 |
"input_contiguous": false
|
| 109 |
},
|
| 110 |
"nar_prefill_source": "handoff",
|
| 111 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 112 |
},
|
| 113 |
-
"e2e_seconds":
|
| 114 |
},
|
| 115 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 116 |
"cfg": 1.5,
|
|
@@ -120,7 +160,7 @@
|
|
| 120 |
"kv_cache_dtype": "int8",
|
| 121 |
"captured_sampling": true,
|
| 122 |
"fuse_rms_quant": true,
|
| 123 |
-
"orbitquant_version": "0.9.
|
| 124 |
"profile": "fast",
|
| 125 |
"allocator_backend": "native",
|
| 126 |
"modules": 224,
|
|
@@ -143,46 +183,47 @@
|
|
| 143 |
},
|
| 144 |
{
|
| 145 |
"bundle": "v11",
|
| 146 |
-
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,kv_handoff,
|
| 147 |
"flags": [
|
| 148 |
"abc_crop",
|
| 149 |
"int8_heads",
|
| 150 |
"kv_handoff",
|
|
|
|
| 151 |
"lookahead",
|
| 152 |
-
"nar_compile",
|
| 153 |
"resident",
|
| 154 |
"vae_compile",
|
|
|
|
| 155 |
"vae_mode=default",
|
| 156 |
"w8a8"
|
| 157 |
],
|
| 158 |
-
"native_sha256": "
|
| 159 |
"repeat": 1,
|
| 160 |
-
"wall_seconds":
|
| 161 |
-
"load_seconds":
|
| 162 |
"audio_seconds": 140.63866666666667,
|
| 163 |
-
"audio_per_wall":
|
| 164 |
-
"abc_s":
|
| 165 |
"abc_tokens": 1249,
|
| 166 |
-
"sem_s":
|
| 167 |
"sem_tokens": 3517,
|
| 168 |
-
"abc_ms_per_tok": 2.
|
| 169 |
-
"sem_ms_per_tok": 2.
|
| 170 |
-
"nar_s":
|
| 171 |
-
"vae_s": 0.
|
| 172 |
-
"nvml_peak_gib": 5.
|
| 173 |
-
"torch_peak_gib":
|
| 174 |
-
"torch_reserved_peak_gib": 4.
|
| 175 |
"semantic_sha256": "4c3a54b7d2d1a1be8bd05d35f4b773e1dc375533dcfe49694b1791126c3797a4",
|
| 176 |
-
"waveform_sha256": "
|
| 177 |
-
"latent_sha256": "
|
| 178 |
"timing": {
|
| 179 |
"abc": {
|
| 180 |
-
"seconds":
|
| 181 |
-
"prefill_seconds": 0.
|
| 182 |
-
"ttft_seconds": 0.
|
| 183 |
"output_tokens": 1249,
|
| 184 |
"content_tokens": 1248,
|
| 185 |
-
"output_tps":
|
| 186 |
"prefix_tokens": 423,
|
| 187 |
"cfg_branches": 1,
|
| 188 |
"sampling": "combined_cuda_graph",
|
|
@@ -193,16 +234,16 @@
|
|
| 193 |
"lookahead": true,
|
| 194 |
"unroll2": false,
|
| 195 |
"first_step_seconds": null,
|
| 196 |
-
"capture_seconds": 0.
|
| 197 |
-
"close_seconds": 0.
|
| 198 |
},
|
| 199 |
"semantic": {
|
| 200 |
-
"seconds":
|
| 201 |
-
"prefill_seconds": 0.
|
| 202 |
-
"ttft_seconds": 0.
|
| 203 |
"output_tokens": 3517,
|
| 204 |
"content_tokens": 3516,
|
| 205 |
-
"output_tps":
|
| 206 |
"prefix_tokens": 1673,
|
| 207 |
"cfg_branches": 2,
|
| 208 |
"sampling": "combined_cuda_graph",
|
|
@@ -213,46 +254,85 @@
|
|
| 213 |
"lookahead": true,
|
| 214 |
"unroll2": false,
|
| 215 |
"first_step_seconds": null,
|
| 216 |
-
"capture_seconds": 0.
|
| 217 |
-
"close_seconds": 0.
|
| 218 |
},
|
| 219 |
-
"nar_seconds":
|
| 220 |
-
"vae_seconds": 0.
|
| 221 |
"load": {
|
| 222 |
-
"resolve_and_integrity_seconds":
|
| 223 |
-
"mot_load_seconds":
|
| 224 |
},
|
| 225 |
"stages": {
|
| 226 |
-
"t0->plan_done":
|
| 227 |
-
"plan_done->semantic_done":
|
| 228 |
-
"semantic_done->nar_done":
|
| 229 |
-
"nar_done->vae_done": 0.
|
| 230 |
-
"plan_outside_timer": 0.
|
| 231 |
-
"semantic_outside_timer": 0.
|
| 232 |
-
"semantic_close_seconds": 0.
|
| 233 |
"vae": {
|
| 234 |
-
"load_s": 0.
|
| 235 |
-
"decode_s": 0.
|
| 236 |
-
"post_s":
|
| 237 |
"tiles": [
|
| 238 |
-
0.
|
| 239 |
-
0.
|
| 240 |
-
0.
|
| 241 |
-
0.
|
| 242 |
-
0.
|
| 243 |
-
0.
|
| 244 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 245 |
],
|
| 246 |
"cudnn_benchmark": false,
|
| 247 |
"cudnn_deterministic": true,
|
| 248 |
-
"core_frames":
|
| 249 |
"input_device": "cpu",
|
| 250 |
"input_contiguous": false
|
| 251 |
},
|
| 252 |
"nar_prefill_source": "handoff",
|
| 253 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 254 |
},
|
| 255 |
-
"e2e_seconds":
|
| 256 |
},
|
| 257 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 258 |
"cfg": 1.5,
|
|
@@ -262,7 +342,7 @@
|
|
| 262 |
"kv_cache_dtype": "int8",
|
| 263 |
"captured_sampling": true,
|
| 264 |
"fuse_rms_quant": true,
|
| 265 |
-
"orbitquant_version": "0.9.
|
| 266 |
"profile": "fast",
|
| 267 |
"allocator_backend": "native",
|
| 268 |
"modules": 224,
|
|
|
|
| 1 |
[
|
| 2 |
{
|
| 3 |
"bundle": "v11",
|
| 4 |
+
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,vae_core=256,kv_handoff,lazy_handoff,int8_heads,w8a8",
|
| 5 |
"flags": [
|
| 6 |
"abc_crop",
|
| 7 |
"int8_heads",
|
| 8 |
"kv_handoff",
|
| 9 |
+
"lazy_handoff",
|
| 10 |
"lookahead",
|
|
|
|
| 11 |
"resident",
|
| 12 |
"vae_compile",
|
| 13 |
+
"vae_core=256",
|
| 14 |
"vae_mode=default",
|
| 15 |
"w8a8"
|
| 16 |
],
|
| 17 |
+
"native_sha256": "118d64db97c3200ef785e83b8d703ebc3b480a799bb17a6affb104d1b4faa28d",
|
| 18 |
"repeat": 0,
|
| 19 |
+
"wall_seconds": 17.9887717962265,
|
| 20 |
+
"load_seconds": 2.2472441755235195,
|
| 21 |
"audio_seconds": 140.63866666666667,
|
| 22 |
+
"audio_per_wall": 7.818136127346303,
|
| 23 |
+
"abc_s": 3.3777515897527337,
|
| 24 |
"abc_tokens": 1249,
|
| 25 |
+
"sem_s": 10.058794901706278,
|
| 26 |
"sem_tokens": 3517,
|
| 27 |
+
"abc_ms_per_tok": 2.7043647636130776,
|
| 28 |
+
"sem_ms_per_tok": 2.860049730368575,
|
| 29 |
+
"nar_s": 2.5382185569033027,
|
| 30 |
+
"vae_s": 2.0050907684490085,
|
| 31 |
+
"nvml_peak_gib": 4.939697265625,
|
| 32 |
+
"torch_peak_gib": 2.735057830810547,
|
| 33 |
+
"torch_reserved_peak_gib": 3.908203125,
|
| 34 |
"semantic_sha256": "4c3a54b7d2d1a1be8bd05d35f4b773e1dc375533dcfe49694b1791126c3797a4",
|
| 35 |
+
"waveform_sha256": "71e4c0580fdd3ed99a0bfd5738f2482a0582ee0e66577625dc7d7c982a11182f",
|
| 36 |
+
"latent_sha256": "411c80bab4d16cac636006e231760df8697a8cad7457a8cdd14bb12df46a292d",
|
| 37 |
"timing": {
|
| 38 |
"abc": {
|
| 39 |
+
"seconds": 3.3777515897527337,
|
| 40 |
+
"prefill_seconds": 0.6063994308933616,
|
| 41 |
+
"ttft_seconds": 0.6074786949902773,
|
| 42 |
"output_tokens": 1249,
|
| 43 |
"content_tokens": 1248,
|
| 44 |
+
"output_tps": 369.77260370157427,
|
| 45 |
"prefix_tokens": 423,
|
| 46 |
"cfg_branches": 1,
|
| 47 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 52 |
"lookahead": true,
|
| 53 |
"unroll2": false,
|
| 54 |
"first_step_seconds": null,
|
| 55 |
+
"capture_seconds": 0.0399644561111927,
|
| 56 |
+
"close_seconds": 0.0002987785264849663
|
| 57 |
},
|
| 58 |
"semantic": {
|
| 59 |
+
"seconds": 10.058794901706278,
|
| 60 |
+
"prefill_seconds": 1.4141903733834624,
|
| 61 |
+
"ttft_seconds": 1.4142127567902207,
|
| 62 |
"output_tokens": 3517,
|
| 63 |
"content_tokens": 3516,
|
| 64 |
+
"output_tps": 349.6442699516032,
|
| 65 |
"prefix_tokens": 1673,
|
| 66 |
"cfg_branches": 2,
|
| 67 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 72 |
"lookahead": true,
|
| 73 |
"unroll2": false,
|
| 74 |
"first_step_seconds": null,
|
| 75 |
+
"capture_seconds": 0.2498519727960229,
|
| 76 |
+
"close_seconds": 0.00044487975537776947
|
| 77 |
},
|
| 78 |
+
"nar_seconds": 2.5382185569033027,
|
| 79 |
+
"vae_seconds": 2.0050907684490085,
|
| 80 |
"load": {
|
| 81 |
+
"resolve_and_integrity_seconds": 0.1945256581529975,
|
| 82 |
+
"mot_load_seconds": 0.7143985284492373
|
| 83 |
},
|
| 84 |
"stages": {
|
| 85 |
+
"t0->plan_done": 3.3831612318754196,
|
| 86 |
+
"plan_done->semantic_done": 10.060313027352095,
|
| 87 |
+
"semantic_done->nar_done": 2.538495653308928,
|
| 88 |
+
"nar_done->vae_done": 2.0053876154124737,
|
| 89 |
+
"plan_outside_timer": 0.005409642122685909,
|
| 90 |
+
"semantic_outside_timer": 0.0015181256458163261,
|
| 91 |
+
"semantic_close_seconds": 0.00044487975537776947,
|
| 92 |
"vae": {
|
| 93 |
+
"load_s": 0.027230621315538883,
|
| 94 |
+
"decode_s": 1.9777783742174506,
|
| 95 |
+
"post_s": 1.3129785656929016e-05,
|
| 96 |
"tiles": [
|
| 97 |
+
1.7687,
|
| 98 |
+
0.0138,
|
| 99 |
+
0.0136,
|
| 100 |
+
0.0135,
|
| 101 |
+
0.0137,
|
| 102 |
+
0.0139,
|
| 103 |
+
0.0139,
|
| 104 |
+
0.014,
|
| 105 |
+
0.0144,
|
| 106 |
+
0.0142,
|
| 107 |
+
0.0143,
|
| 108 |
+
0.0141,
|
| 109 |
+
0.0141,
|
| 110 |
+
0.0412
|
| 111 |
],
|
| 112 |
"cudnn_benchmark": false,
|
| 113 |
"cudnn_deterministic": true,
|
| 114 |
+
"core_frames": 256,
|
| 115 |
"input_device": "cpu",
|
| 116 |
"input_contiguous": false
|
| 117 |
},
|
| 118 |
"nar_prefill_source": "handoff",
|
| 119 |
+
"abc_close_seconds": 0.0002987785264849663,
|
| 120 |
+
"memory": {
|
| 121 |
+
"t0": {
|
| 122 |
+
"peak_gib": 2.479,
|
| 123 |
+
"alloc_gib": 2.479,
|
| 124 |
+
"reserved_gib": 2.756,
|
| 125 |
+
"nvml_used_gib": 3.674
|
| 126 |
+
},
|
| 127 |
+
"plan_done": {
|
| 128 |
+
"peak_gib": 2.754,
|
| 129 |
+
"alloc_gib": 2.479,
|
| 130 |
+
"reserved_gib": 2.98,
|
| 131 |
+
"nvml_used_gib": 3.922
|
| 132 |
+
},
|
| 133 |
+
"semantic_done": {
|
| 134 |
+
"peak_gib": 3.757,
|
| 135 |
+
"alloc_gib": 3.654,
|
| 136 |
+
"reserved_gib": 3.895,
|
| 137 |
+
"nvml_used_gib": 4.838
|
| 138 |
+
},
|
| 139 |
+
"nar_done": {
|
| 140 |
+
"peak_gib": 3.68,
|
| 141 |
+
"alloc_gib": 2.488,
|
| 142 |
+
"reserved_gib": 3.908,
|
| 143 |
+
"nvml_used_gib": 4.867
|
| 144 |
+
},
|
| 145 |
+
"vae_done": {
|
| 146 |
+
"peak_gib": 3.215,
|
| 147 |
+
"alloc_gib": 2.735,
|
| 148 |
+
"reserved_gib": 3.908,
|
| 149 |
+
"nvml_used_gib": 4.877
|
| 150 |
+
}
|
| 151 |
+
}
|
| 152 |
},
|
| 153 |
+
"e2e_seconds": 17.98769250139594
|
| 154 |
},
|
| 155 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 156 |
"cfg": 1.5,
|
|
|
|
| 160 |
"kv_cache_dtype": "int8",
|
| 161 |
"captured_sampling": true,
|
| 162 |
"fuse_rms_quant": true,
|
| 163 |
+
"orbitquant_version": "0.9.8",
|
| 164 |
"profile": "fast",
|
| 165 |
"allocator_backend": "native",
|
| 166 |
"modules": 224,
|
|
|
|
| 183 |
},
|
| 184 |
{
|
| 185 |
"bundle": "v11",
|
| 186 |
+
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,vae_core=256,kv_handoff,lazy_handoff,int8_heads,w8a8",
|
| 187 |
"flags": [
|
| 188 |
"abc_crop",
|
| 189 |
"int8_heads",
|
| 190 |
"kv_handoff",
|
| 191 |
+
"lazy_handoff",
|
| 192 |
"lookahead",
|
|
|
|
| 193 |
"resident",
|
| 194 |
"vae_compile",
|
| 195 |
+
"vae_core=256",
|
| 196 |
"vae_mode=default",
|
| 197 |
"w8a8"
|
| 198 |
],
|
| 199 |
+
"native_sha256": "118d64db97c3200ef785e83b8d703ebc3b480a799bb17a6affb104d1b4faa28d",
|
| 200 |
"repeat": 1,
|
| 201 |
+
"wall_seconds": 13.461460947990417,
|
| 202 |
+
"load_seconds": 2.2472441755235195,
|
| 203 |
"audio_seconds": 140.63866666666667,
|
| 204 |
+
"audio_per_wall": 10.447503967811294,
|
| 205 |
+
"abc_s": 2.8130570147186518,
|
| 206 |
"abc_tokens": 1249,
|
| 207 |
+
"sem_s": 8.530152230523527,
|
| 208 |
"sem_tokens": 3517,
|
| 209 |
+
"abc_ms_per_tok": 2.2522474097026834,
|
| 210 |
+
"sem_ms_per_tok": 2.4254058090769197,
|
| 211 |
+
"nar_s": 1.906809726729989,
|
| 212 |
+
"vae_s": 0.2063934374600649,
|
| 213 |
+
"nvml_peak_gib": 5.199462890625,
|
| 214 |
+
"torch_peak_gib": 2.735057830810547,
|
| 215 |
+
"torch_reserved_peak_gib": 4.123046875,
|
| 216 |
"semantic_sha256": "4c3a54b7d2d1a1be8bd05d35f4b773e1dc375533dcfe49694b1791126c3797a4",
|
| 217 |
+
"waveform_sha256": "71e4c0580fdd3ed99a0bfd5738f2482a0582ee0e66577625dc7d7c982a11182f",
|
| 218 |
+
"latent_sha256": "411c80bab4d16cac636006e231760df8697a8cad7457a8cdd14bb12df46a292d",
|
| 219 |
"timing": {
|
| 220 |
"abc": {
|
| 221 |
+
"seconds": 2.8130570147186518,
|
| 222 |
+
"prefill_seconds": 0.04717915412038565,
|
| 223 |
+
"ttft_seconds": 0.047200084663927555,
|
| 224 |
"output_tokens": 1249,
|
| 225 |
"content_tokens": 1248,
|
| 226 |
+
"output_tps": 444.00095464290433,
|
| 227 |
"prefix_tokens": 423,
|
| 228 |
"cfg_branches": 1,
|
| 229 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 234 |
"lookahead": true,
|
| 235 |
"unroll2": false,
|
| 236 |
"first_step_seconds": null,
|
| 237 |
+
"capture_seconds": 0.033918716944754124,
|
| 238 |
+
"close_seconds": 0.00039780884981155396
|
| 239 |
},
|
| 240 |
"semantic": {
|
| 241 |
+
"seconds": 8.530152230523527,
|
| 242 |
+
"prefill_seconds": 0.08340269047766924,
|
| 243 |
+
"ttft_seconds": 0.08342811837792397,
|
| 244 |
"output_tokens": 3517,
|
| 245 |
"content_tokens": 3516,
|
| 246 |
+
"output_tps": 412.3021377526048,
|
| 247 |
"prefix_tokens": 1673,
|
| 248 |
"cfg_branches": 2,
|
| 249 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 254 |
"lookahead": true,
|
| 255 |
"unroll2": false,
|
| 256 |
"first_step_seconds": null,
|
| 257 |
+
"capture_seconds": 0.03369541093707085,
|
| 258 |
+
"close_seconds": 0.00037660542875528336
|
| 259 |
},
|
| 260 |
+
"nar_seconds": 1.906809726729989,
|
| 261 |
+
"vae_seconds": 0.2063934374600649,
|
| 262 |
"load": {
|
| 263 |
+
"resolve_and_integrity_seconds": 0.1945256581529975,
|
| 264 |
+
"mot_load_seconds": 0.7143985284492373
|
| 265 |
},
|
| 266 |
"stages": {
|
| 267 |
+
"t0->plan_done": 2.814796471968293,
|
| 268 |
+
"plan_done->semantic_done": 8.531699027866125,
|
| 269 |
+
"semantic_done->nar_done": 1.9072276363149285,
|
| 270 |
+
"nar_done->vae_done": 0.2066958835348487,
|
| 271 |
+
"plan_outside_timer": 0.0017394572496414185,
|
| 272 |
+
"semantic_outside_timer": 0.0015467973425984383,
|
| 273 |
+
"semantic_close_seconds": 0.00037660542875528336,
|
| 274 |
"vae": {
|
| 275 |
+
"load_s": 0.0009437203407287598,
|
| 276 |
+
"decode_s": 0.20538921747356653,
|
| 277 |
+
"post_s": 1.1157244443893433e-05,
|
| 278 |
"tiles": [
|
| 279 |
+
0.0144,
|
| 280 |
+
0.0138,
|
| 281 |
+
0.0139,
|
| 282 |
+
0.0139,
|
| 283 |
+
0.0137,
|
| 284 |
+
0.0138,
|
| 285 |
+
0.0138,
|
| 286 |
+
0.0137,
|
| 287 |
+
0.0137,
|
| 288 |
+
0.0137,
|
| 289 |
+
0.0136,
|
| 290 |
+
0.0136,
|
| 291 |
+
0.0137,
|
| 292 |
+
0.026
|
| 293 |
],
|
| 294 |
"cudnn_benchmark": false,
|
| 295 |
"cudnn_deterministic": true,
|
| 296 |
+
"core_frames": 256,
|
| 297 |
"input_device": "cpu",
|
| 298 |
"input_contiguous": false
|
| 299 |
},
|
| 300 |
"nar_prefill_source": "handoff",
|
| 301 |
+
"abc_close_seconds": 0.00039780884981155396,
|
| 302 |
+
"memory": {
|
| 303 |
+
"t0": {
|
| 304 |
+
"peak_gib": 2.735,
|
| 305 |
+
"alloc_gib": 2.735,
|
| 306 |
+
"reserved_gib": 3.908,
|
| 307 |
+
"nvml_used_gib": 4.877
|
| 308 |
+
},
|
| 309 |
+
"plan_done": {
|
| 310 |
+
"peak_gib": 3.01,
|
| 311 |
+
"alloc_gib": 2.735,
|
| 312 |
+
"reserved_gib": 3.215,
|
| 313 |
+
"nvml_used_gib": 4.184
|
| 314 |
+
},
|
| 315 |
+
"semantic_done": {
|
| 316 |
+
"peak_gib": 4.014,
|
| 317 |
+
"alloc_gib": 3.911,
|
| 318 |
+
"reserved_gib": 4.109,
|
| 319 |
+
"nvml_used_gib": 5.078
|
| 320 |
+
},
|
| 321 |
+
"nar_done": {
|
| 322 |
+
"peak_gib": 3.936,
|
| 323 |
+
"alloc_gib": 2.735,
|
| 324 |
+
"reserved_gib": 4.123,
|
| 325 |
+
"nvml_used_gib": 5.092
|
| 326 |
+
},
|
| 327 |
+
"vae_done": {
|
| 328 |
+
"peak_gib": 3.215,
|
| 329 |
+
"alloc_gib": 2.735,
|
| 330 |
+
"reserved_gib": 4.123,
|
| 331 |
+
"nvml_used_gib": 5.092
|
| 332 |
+
}
|
| 333 |
+
}
|
| 334 |
},
|
| 335 |
+
"e2e_seconds": 13.460732371546328
|
| 336 |
},
|
| 337 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 338 |
"cfg": 1.5,
|
|
|
|
| 342 |
"kv_cache_dtype": "int8",
|
| 343 |
"captured_sampling": true,
|
| 344 |
"fuse_rms_quant": true,
|
| 345 |
+
"orbitquant_version": "0.9.8",
|
| 346 |
"profile": "fast",
|
| 347 |
"allocator_backend": "native",
|
| 348 |
"modules": 224,
|
evaluation/profiles/fast/run-00/config.json
CHANGED
|
@@ -34,12 +34,12 @@
|
|
| 34 |
"model_dtype": "bfloat16",
|
| 35 |
"vae_dtype": "float32",
|
| 36 |
"vae_decode": "halo_crop",
|
| 37 |
-
"vae_core_frames":
|
| 38 |
"vae_halo_frames": 16,
|
| 39 |
"device": "cuda:0",
|
| 40 |
"memory_budget_gib": 18.0,
|
| 41 |
"offload_ar": false,
|
| 42 |
-
"runtime_sha256": "
|
| 43 |
"decoder_release": "standard",
|
| 44 |
"validation_status": "unvalidated",
|
| 45 |
"runtime_profile": "fast",
|
|
@@ -47,7 +47,7 @@
|
|
| 47 |
"captured_sampling": true,
|
| 48 |
"kv_cache_dtype": "int8",
|
| 49 |
"int8_attention": "ordered",
|
| 50 |
-
"runtime_bundle_sha256": "
|
| 51 |
"vae_compute_dtype": "torch.float16",
|
| 52 |
"expert_staging": false,
|
| 53 |
"orbitquant": {
|
|
|
|
| 34 |
"model_dtype": "bfloat16",
|
| 35 |
"vae_dtype": "float32",
|
| 36 |
"vae_decode": "halo_crop",
|
| 37 |
+
"vae_core_frames": 256,
|
| 38 |
"vae_halo_frames": 16,
|
| 39 |
"device": "cuda:0",
|
| 40 |
"memory_budget_gib": 18.0,
|
| 41 |
"offload_ar": false,
|
| 42 |
+
"runtime_sha256": "bb189091c79b42dd3eca4f24f6239294df3c517fb2b149fc3b30ada3be287119",
|
| 43 |
"decoder_release": "standard",
|
| 44 |
"validation_status": "unvalidated",
|
| 45 |
"runtime_profile": "fast",
|
|
|
|
| 47 |
"captured_sampling": true,
|
| 48 |
"kv_cache_dtype": "int8",
|
| 49 |
"int8_attention": "ordered",
|
| 50 |
+
"runtime_bundle_sha256": "269214c6459739a0f08889532d5fc76f3e6f48f3908807466e14e5da4ecb1c98",
|
| 51 |
"vae_compute_dtype": "torch.float16",
|
| 52 |
"expert_staging": false,
|
| 53 |
"orbitquant": {
|
evaluation/profiles/fast/run-00/metrics.json
CHANGED
|
@@ -1,45 +1,46 @@
|
|
| 1 |
{
|
| 2 |
"bundle": "v11",
|
| 3 |
-
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,kv_handoff,
|
| 4 |
"flags": [
|
| 5 |
"abc_crop",
|
| 6 |
"int8_heads",
|
| 7 |
"kv_handoff",
|
|
|
|
| 8 |
"lookahead",
|
| 9 |
-
"nar_compile",
|
| 10 |
"resident",
|
| 11 |
"vae_compile",
|
|
|
|
| 12 |
"vae_mode=default",
|
| 13 |
"w8a8"
|
| 14 |
],
|
| 15 |
-
"native_sha256": "
|
| 16 |
"repeat": 0,
|
| 17 |
-
"wall_seconds":
|
| 18 |
-
"load_seconds":
|
| 19 |
"audio_seconds": 140.67866666666666,
|
| 20 |
-
"audio_per_wall":
|
| 21 |
-
"abc_s":
|
| 22 |
"abc_tokens": 1249,
|
| 23 |
-
"sem_s": 7.
|
| 24 |
"sem_tokens": 3518,
|
| 25 |
-
"abc_ms_per_tok":
|
| 26 |
-
"sem_ms_per_tok": 2.
|
| 27 |
-
"nar_s":
|
| 28 |
-
"vae_s":
|
| 29 |
-
"nvml_peak_gib": 4.
|
| 30 |
-
"torch_peak_gib":
|
| 31 |
-
"torch_reserved_peak_gib": 3.
|
| 32 |
"semantic_sha256": "bcb32247a0bd840b8b568746f93e93f823a87796b1f87fb55492da8eb741b473",
|
| 33 |
-
"waveform_sha256": "
|
| 34 |
-
"latent_sha256": "
|
| 35 |
"timing": {
|
| 36 |
"abc": {
|
| 37 |
-
"seconds":
|
| 38 |
-
"prefill_seconds":
|
| 39 |
-
"ttft_seconds":
|
| 40 |
"output_tokens": 1249,
|
| 41 |
"content_tokens": 1248,
|
| 42 |
-
"output_tps":
|
| 43 |
"prefix_tokens": 423,
|
| 44 |
"cfg_branches": 1,
|
| 45 |
"sampling": "combined_cuda_graph",
|
|
@@ -50,16 +51,16 @@
|
|
| 50 |
"lookahead": true,
|
| 51 |
"unroll2": false,
|
| 52 |
"first_step_seconds": null,
|
| 53 |
-
"capture_seconds": 0.
|
| 54 |
-
"close_seconds": 0.
|
| 55 |
},
|
| 56 |
"semantic": {
|
| 57 |
-
"seconds": 7.
|
| 58 |
-
"prefill_seconds": 0.
|
| 59 |
-
"ttft_seconds": 0.
|
| 60 |
"output_tokens": 3518,
|
| 61 |
"content_tokens": 3517,
|
| 62 |
-
"output_tps":
|
| 63 |
"prefix_tokens": 1673,
|
| 64 |
"cfg_branches": 1,
|
| 65 |
"sampling": "combined_cuda_graph",
|
|
@@ -70,46 +71,85 @@
|
|
| 70 |
"lookahead": true,
|
| 71 |
"unroll2": false,
|
| 72 |
"first_step_seconds": null,
|
| 73 |
-
"capture_seconds": 0.
|
| 74 |
-
"close_seconds": 0.
|
| 75 |
},
|
| 76 |
-
"nar_seconds":
|
| 77 |
-
"vae_seconds":
|
| 78 |
"load": {
|
| 79 |
-
"resolve_and_integrity_seconds":
|
| 80 |
-
"mot_load_seconds":
|
| 81 |
},
|
| 82 |
"stages": {
|
| 83 |
-
"t0->plan_done":
|
| 84 |
-
"plan_done->semantic_done": 7.
|
| 85 |
-
"semantic_done->nar_done":
|
| 86 |
-
"nar_done->vae_done":
|
| 87 |
-
"plan_outside_timer": 0.
|
| 88 |
-
"semantic_outside_timer": 0.
|
| 89 |
-
"semantic_close_seconds": 0.
|
| 90 |
"vae": {
|
| 91 |
-
"load_s":
|
| 92 |
-
"decode_s":
|
| 93 |
-
"post_s":
|
| 94 |
"tiles": [
|
| 95 |
-
|
| 96 |
-
0.
|
| 97 |
-
0.
|
| 98 |
-
0.
|
| 99 |
-
0.
|
| 100 |
-
0.
|
| 101 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 102 |
],
|
| 103 |
"cudnn_benchmark": false,
|
| 104 |
"cudnn_deterministic": true,
|
| 105 |
-
"core_frames":
|
| 106 |
"input_device": "cpu",
|
| 107 |
"input_contiguous": false
|
| 108 |
},
|
| 109 |
"nar_prefill_source": "handoff",
|
| 110 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 111 |
},
|
| 112 |
-
"e2e_seconds":
|
| 113 |
},
|
| 114 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 115 |
"cfg": 1.0,
|
|
@@ -119,7 +159,7 @@
|
|
| 119 |
"kv_cache_dtype": "int8",
|
| 120 |
"captured_sampling": true,
|
| 121 |
"fuse_rms_quant": true,
|
| 122 |
-
"orbitquant_version": "0.9.
|
| 123 |
"profile": "fast",
|
| 124 |
"allocator_backend": "native",
|
| 125 |
"modules": 224,
|
|
|
|
| 1 |
{
|
| 2 |
"bundle": "v11",
|
| 3 |
+
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,vae_core=256,kv_handoff,lazy_handoff,int8_heads,w8a8",
|
| 4 |
"flags": [
|
| 5 |
"abc_crop",
|
| 6 |
"int8_heads",
|
| 7 |
"kv_handoff",
|
| 8 |
+
"lazy_handoff",
|
| 9 |
"lookahead",
|
|
|
|
| 10 |
"resident",
|
| 11 |
"vae_compile",
|
| 12 |
+
"vae_core=256",
|
| 13 |
"vae_mode=default",
|
| 14 |
"w8a8"
|
| 15 |
],
|
| 16 |
+
"native_sha256": "118d64db97c3200ef785e83b8d703ebc3b480a799bb17a6affb104d1b4faa28d",
|
| 17 |
"repeat": 0,
|
| 18 |
+
"wall_seconds": 14.985645630396903,
|
| 19 |
+
"load_seconds": 2.204909392632544,
|
| 20 |
"audio_seconds": 140.67866666666666,
|
| 21 |
+
"audio_per_wall": 9.38756127939619,
|
| 22 |
+
"abc_s": 3.2614384908229113,
|
| 23 |
"abc_tokens": 1249,
|
| 24 |
+
"sem_s": 7.348635695874691,
|
| 25 |
"sem_tokens": 3518,
|
| 26 |
+
"abc_ms_per_tok": 2.6112397844859174,
|
| 27 |
+
"sem_ms_per_tok": 2.0888674519257227,
|
| 28 |
+
"nar_s": 2.046467389911413,
|
| 29 |
+
"vae_s": 2.3237511478364468,
|
| 30 |
+
"nvml_peak_gib": 4.486572265625,
|
| 31 |
+
"torch_peak_gib": 2.735057830810547,
|
| 32 |
+
"torch_reserved_peak_gib": 3.517578125,
|
| 33 |
"semantic_sha256": "bcb32247a0bd840b8b568746f93e93f823a87796b1f87fb55492da8eb741b473",
|
| 34 |
+
"waveform_sha256": "e0984b3364956b3d27d98703ee965d578951863b358ccff6769e544a15a77648",
|
| 35 |
+
"latent_sha256": "1763ac7941cbedffae6c43ca33e947ca9c8c390ae0d290db8b6c426e97439b36",
|
| 36 |
"timing": {
|
| 37 |
"abc": {
|
| 38 |
+
"seconds": 3.2614384908229113,
|
| 39 |
+
"prefill_seconds": 0.4902397319674492,
|
| 40 |
+
"ttft_seconds": 0.49129611160606146,
|
| 41 |
"output_tokens": 1249,
|
| 42 |
"content_tokens": 1248,
|
| 43 |
+
"output_tps": 382.9598514626158,
|
| 44 |
"prefix_tokens": 423,
|
| 45 |
"cfg_branches": 1,
|
| 46 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 51 |
"lookahead": true,
|
| 52 |
"unroll2": false,
|
| 53 |
"first_step_seconds": null,
|
| 54 |
+
"capture_seconds": 0.039790925569832325,
|
| 55 |
+
"close_seconds": 0.0003131115809082985
|
| 56 |
},
|
| 57 |
"semantic": {
|
| 58 |
+
"seconds": 7.348635695874691,
|
| 59 |
+
"prefill_seconds": 0.056065828539431095,
|
| 60 |
+
"ttft_seconds": 0.05608602799475193,
|
| 61 |
"output_tokens": 3518,
|
| 62 |
"content_tokens": 3517,
|
| 63 |
+
"output_tps": 478.7283171453039,
|
| 64 |
"prefix_tokens": 1673,
|
| 65 |
"cfg_branches": 1,
|
| 66 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 71 |
"lookahead": true,
|
| 72 |
"unroll2": false,
|
| 73 |
"first_step_seconds": null,
|
| 74 |
+
"capture_seconds": 0.03171390853822231,
|
| 75 |
+
"close_seconds": 0.00040434859693050385
|
| 76 |
},
|
| 77 |
+
"nar_seconds": 2.046467389911413,
|
| 78 |
+
"vae_seconds": 2.3237511478364468,
|
| 79 |
"load": {
|
| 80 |
+
"resolve_and_integrity_seconds": 0.19479286018759012,
|
| 81 |
+
"mot_load_seconds": 0.6600089315325022
|
| 82 |
},
|
| 83 |
"stages": {
|
| 84 |
+
"t0->plan_done": 3.2638021977618337,
|
| 85 |
+
"plan_done->semantic_done": 7.3499299036338925,
|
| 86 |
+
"semantic_done->nar_done": 2.046788734383881,
|
| 87 |
+
"nar_done->vae_done": 2.3241178588941693,
|
| 88 |
+
"plan_outside_timer": 0.0023637069389224052,
|
| 89 |
+
"semantic_outside_timer": 0.0012942077592015266,
|
| 90 |
+
"semantic_close_seconds": 0.00040434859693050385,
|
| 91 |
"vae": {
|
| 92 |
+
"load_s": 0.34615388605743647,
|
| 93 |
+
"decode_s": 1.9768154453486204,
|
| 94 |
+
"post_s": 1.2327916920185089e-05,
|
| 95 |
"tiles": [
|
| 96 |
+
1.7687,
|
| 97 |
+
0.0144,
|
| 98 |
+
0.0141,
|
| 99 |
+
0.0141,
|
| 100 |
+
0.014,
|
| 101 |
+
0.014,
|
| 102 |
+
0.014,
|
| 103 |
+
0.0139,
|
| 104 |
+
0.0138,
|
| 105 |
+
0.0138,
|
| 106 |
+
0.0138,
|
| 107 |
+
0.0137,
|
| 108 |
+
0.0138,
|
| 109 |
+
0.0404
|
| 110 |
],
|
| 111 |
"cudnn_benchmark": false,
|
| 112 |
"cudnn_deterministic": true,
|
| 113 |
+
"core_frames": 256,
|
| 114 |
"input_device": "cpu",
|
| 115 |
"input_contiguous": false
|
| 116 |
},
|
| 117 |
"nar_prefill_source": "handoff",
|
| 118 |
+
"abc_close_seconds": 0.0003131115809082985,
|
| 119 |
+
"memory": {
|
| 120 |
+
"t0": {
|
| 121 |
+
"peak_gib": 2.479,
|
| 122 |
+
"alloc_gib": 2.479,
|
| 123 |
+
"reserved_gib": 2.756,
|
| 124 |
+
"nvml_used_gib": 3.674
|
| 125 |
+
},
|
| 126 |
+
"plan_done": {
|
| 127 |
+
"peak_gib": 2.754,
|
| 128 |
+
"alloc_gib": 2.479,
|
| 129 |
+
"reserved_gib": 2.98,
|
| 130 |
+
"nvml_used_gib": 3.922
|
| 131 |
+
},
|
| 132 |
+
"semantic_done": {
|
| 133 |
+
"peak_gib": 3.17,
|
| 134 |
+
"alloc_gib": 3.067,
|
| 135 |
+
"reserved_gib": 3.309,
|
| 136 |
+
"nvml_used_gib": 4.252
|
| 137 |
+
},
|
| 138 |
+
"nar_done": {
|
| 139 |
+
"peak_gib": 3.12,
|
| 140 |
+
"alloc_gib": 2.488,
|
| 141 |
+
"reserved_gib": 3.4,
|
| 142 |
+
"nvml_used_gib": 4.36
|
| 143 |
+
},
|
| 144 |
+
"vae_done": {
|
| 145 |
+
"peak_gib": 3.216,
|
| 146 |
+
"alloc_gib": 2.735,
|
| 147 |
+
"reserved_gib": 3.518,
|
| 148 |
+
"nvml_used_gib": 4.487
|
| 149 |
+
}
|
| 150 |
+
}
|
| 151 |
},
|
| 152 |
+
"e2e_seconds": 14.984941340982914
|
| 153 |
},
|
| 154 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 155 |
"cfg": 1.0,
|
|
|
|
| 159 |
"kv_cache_dtype": "int8",
|
| 160 |
"captured_sampling": true,
|
| 161 |
"fuse_rms_quant": true,
|
| 162 |
+
"orbitquant_version": "0.9.8",
|
| 163 |
"profile": "fast",
|
| 164 |
"allocator_backend": "native",
|
| 165 |
"modules": 224,
|
evaluation/profiles/fast/run-00/plan.json
CHANGED
|
@@ -9,12 +9,12 @@
|
|
| 9 |
"id": "song"
|
| 10 |
},
|
| 11 |
"timing": {
|
| 12 |
-
"seconds":
|
| 13 |
-
"prefill_seconds":
|
| 14 |
-
"ttft_seconds":
|
| 15 |
"output_tokens": 1249,
|
| 16 |
"content_tokens": 1248,
|
| 17 |
-
"output_tps":
|
| 18 |
"prefix_tokens": 423,
|
| 19 |
"cfg_branches": 1,
|
| 20 |
"sampling": "combined_cuda_graph",
|
|
@@ -25,8 +25,8 @@
|
|
| 25 |
"lookahead": true,
|
| 26 |
"unroll2": false,
|
| 27 |
"first_step_seconds": null,
|
| 28 |
-
"capture_seconds": 0.
|
| 29 |
-
"close_seconds": 0.
|
| 30 |
},
|
| 31 |
"truncated": false,
|
| 32 |
"prefix": [
|
|
|
|
| 9 |
"id": "song"
|
| 10 |
},
|
| 11 |
"timing": {
|
| 12 |
+
"seconds": 3.2614384908229113,
|
| 13 |
+
"prefill_seconds": 0.4902397319674492,
|
| 14 |
+
"ttft_seconds": 0.49129611160606146,
|
| 15 |
"output_tokens": 1249,
|
| 16 |
"content_tokens": 1248,
|
| 17 |
+
"output_tps": 382.9598514626158,
|
| 18 |
"prefix_tokens": 423,
|
| 19 |
"cfg_branches": 1,
|
| 20 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 25 |
"lookahead": true,
|
| 26 |
"unroll2": false,
|
| 27 |
"first_step_seconds": null,
|
| 28 |
+
"capture_seconds": 0.039790925569832325,
|
| 29 |
+
"close_seconds": 0.0003131115809082985
|
| 30 |
},
|
| 31 |
"truncated": false,
|
| 32 |
"prefix": [
|
evaluation/profiles/fast/run-00/plan_manifest.json
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
{
|
| 2 |
-
"plan.json": "
|
| 3 |
"abc_tokens.npy": "2a3f7d12e43955d2b4d6ec2c08f926be3a76a2fdf22a2952fa37148f6ca9c196",
|
| 4 |
"prefix.npy": "b542a03e835df9d2f05a4c2ffdaaf9eca5e2c49b510e51b6560f0d985952bf85",
|
| 5 |
"score.abc": "8d7ed2e1687e35072e41ff1b64b852068ff81d496bb8142d83019b53b66498ab"
|
|
|
|
| 1 |
{
|
| 2 |
+
"plan.json": "1979add7ff86e1abccc806a796464b646f2418bdeaa3367c907974dfde143234",
|
| 3 |
"abc_tokens.npy": "2a3f7d12e43955d2b4d6ec2c08f926be3a76a2fdf22a2952fa37148f6ca9c196",
|
| 4 |
"prefix.npy": "b542a03e835df9d2f05a4c2ffdaaf9eca5e2c49b510e51b6560f0d985952bf85",
|
| 5 |
"score.abc": "8d7ed2e1687e35072e41ff1b64b852068ff81d496bb8142d83019b53b66498ab"
|
evaluation/profiles/fast/run-00/result.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"status": "complete",
|
| 3 |
-
"identity": "
|
| 4 |
"truncated": {
|
| 5 |
"abc": false,
|
| 6 |
"semantic": false
|
|
@@ -29,12 +29,12 @@
|
|
| 29 |
},
|
| 30 |
"timing": {
|
| 31 |
"abc": {
|
| 32 |
-
"seconds":
|
| 33 |
-
"prefill_seconds":
|
| 34 |
-
"ttft_seconds":
|
| 35 |
"output_tokens": 1249,
|
| 36 |
"content_tokens": 1248,
|
| 37 |
-
"output_tps":
|
| 38 |
"prefix_tokens": 423,
|
| 39 |
"cfg_branches": 1,
|
| 40 |
"sampling": "combined_cuda_graph",
|
|
@@ -45,16 +45,16 @@
|
|
| 45 |
"lookahead": true,
|
| 46 |
"unroll2": false,
|
| 47 |
"first_step_seconds": null,
|
| 48 |
-
"capture_seconds": 0.
|
| 49 |
-
"close_seconds": 0.
|
| 50 |
},
|
| 51 |
"semantic": {
|
| 52 |
-
"seconds": 7.
|
| 53 |
-
"prefill_seconds": 0.
|
| 54 |
-
"ttft_seconds": 0.
|
| 55 |
"output_tokens": 3518,
|
| 56 |
"content_tokens": 3517,
|
| 57 |
-
"output_tps":
|
| 58 |
"prefix_tokens": 1673,
|
| 59 |
"cfg_branches": 1,
|
| 60 |
"sampling": "combined_cuda_graph",
|
|
@@ -65,46 +65,85 @@
|
|
| 65 |
"lookahead": true,
|
| 66 |
"unroll2": false,
|
| 67 |
"first_step_seconds": null,
|
| 68 |
-
"capture_seconds": 0.
|
| 69 |
-
"close_seconds": 0.
|
| 70 |
},
|
| 71 |
-
"nar_seconds":
|
| 72 |
-
"vae_seconds":
|
| 73 |
"load": {
|
| 74 |
-
"resolve_and_integrity_seconds":
|
| 75 |
-
"mot_load_seconds":
|
| 76 |
},
|
| 77 |
"stages": {
|
| 78 |
-
"t0->plan_done":
|
| 79 |
-
"plan_done->semantic_done": 7.
|
| 80 |
-
"semantic_done->nar_done":
|
| 81 |
-
"nar_done->vae_done":
|
| 82 |
-
"plan_outside_timer": 0.
|
| 83 |
-
"semantic_outside_timer": 0.
|
| 84 |
-
"semantic_close_seconds": 0.
|
| 85 |
"vae": {
|
| 86 |
-
"load_s":
|
| 87 |
-
"decode_s":
|
| 88 |
-
"post_s":
|
| 89 |
"tiles": [
|
| 90 |
-
|
| 91 |
-
0.
|
| 92 |
-
0.
|
| 93 |
-
0.
|
| 94 |
-
0.
|
| 95 |
-
0.
|
| 96 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 97 |
],
|
| 98 |
"cudnn_benchmark": false,
|
| 99 |
"cudnn_deterministic": true,
|
| 100 |
-
"core_frames":
|
| 101 |
"input_device": "cpu",
|
| 102 |
"input_contiguous": false
|
| 103 |
},
|
| 104 |
"nar_prefill_source": "handoff",
|
| 105 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 106 |
},
|
| 107 |
-
"e2e_seconds":
|
| 108 |
},
|
| 109 |
"artifacts": {
|
| 110 |
"abc_tokens.npy": {
|
|
@@ -112,23 +151,23 @@
|
|
| 112 |
"bytes": 5120
|
| 113 |
},
|
| 114 |
"audio.flac": {
|
| 115 |
-
"sha256": "
|
| 116 |
-
"bytes":
|
| 117 |
},
|
| 118 |
"config.json": {
|
| 119 |
-
"sha256": "
|
| 120 |
"bytes": 2476
|
| 121 |
},
|
| 122 |
"latent.npy": {
|
| 123 |
-
"sha256": "
|
| 124 |
"bytes": 900480
|
| 125 |
},
|
| 126 |
"plan.json": {
|
| 127 |
-
"sha256": "
|
| 128 |
-
"bytes":
|
| 129 |
},
|
| 130 |
"plan_manifest.json": {
|
| 131 |
-
"sha256": "
|
| 132 |
"bytes": 341
|
| 133 |
},
|
| 134 |
"prefix.npy": {
|
|
|
|
| 1 |
{
|
| 2 |
"status": "complete",
|
| 3 |
+
"identity": "5b78b1460d137eb9aa61003dc8c71bf596658015622424138dc69547e7213123",
|
| 4 |
"truncated": {
|
| 5 |
"abc": false,
|
| 6 |
"semantic": false
|
|
|
|
| 29 |
},
|
| 30 |
"timing": {
|
| 31 |
"abc": {
|
| 32 |
+
"seconds": 3.2614384908229113,
|
| 33 |
+
"prefill_seconds": 0.4902397319674492,
|
| 34 |
+
"ttft_seconds": 0.49129611160606146,
|
| 35 |
"output_tokens": 1249,
|
| 36 |
"content_tokens": 1248,
|
| 37 |
+
"output_tps": 382.9598514626158,
|
| 38 |
"prefix_tokens": 423,
|
| 39 |
"cfg_branches": 1,
|
| 40 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 45 |
"lookahead": true,
|
| 46 |
"unroll2": false,
|
| 47 |
"first_step_seconds": null,
|
| 48 |
+
"capture_seconds": 0.039790925569832325,
|
| 49 |
+
"close_seconds": 0.0003131115809082985
|
| 50 |
},
|
| 51 |
"semantic": {
|
| 52 |
+
"seconds": 7.348635695874691,
|
| 53 |
+
"prefill_seconds": 0.056065828539431095,
|
| 54 |
+
"ttft_seconds": 0.05608602799475193,
|
| 55 |
"output_tokens": 3518,
|
| 56 |
"content_tokens": 3517,
|
| 57 |
+
"output_tps": 478.7283171453039,
|
| 58 |
"prefix_tokens": 1673,
|
| 59 |
"cfg_branches": 1,
|
| 60 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 65 |
"lookahead": true,
|
| 66 |
"unroll2": false,
|
| 67 |
"first_step_seconds": null,
|
| 68 |
+
"capture_seconds": 0.03171390853822231,
|
| 69 |
+
"close_seconds": 0.00040434859693050385
|
| 70 |
},
|
| 71 |
+
"nar_seconds": 2.046467389911413,
|
| 72 |
+
"vae_seconds": 2.3237511478364468,
|
| 73 |
"load": {
|
| 74 |
+
"resolve_and_integrity_seconds": 0.19479286018759012,
|
| 75 |
+
"mot_load_seconds": 0.6600089315325022
|
| 76 |
},
|
| 77 |
"stages": {
|
| 78 |
+
"t0->plan_done": 3.2638021977618337,
|
| 79 |
+
"plan_done->semantic_done": 7.3499299036338925,
|
| 80 |
+
"semantic_done->nar_done": 2.046788734383881,
|
| 81 |
+
"nar_done->vae_done": 2.3241178588941693,
|
| 82 |
+
"plan_outside_timer": 0.0023637069389224052,
|
| 83 |
+
"semantic_outside_timer": 0.0012942077592015266,
|
| 84 |
+
"semantic_close_seconds": 0.00040434859693050385,
|
| 85 |
"vae": {
|
| 86 |
+
"load_s": 0.34615388605743647,
|
| 87 |
+
"decode_s": 1.9768154453486204,
|
| 88 |
+
"post_s": 1.2327916920185089e-05,
|
| 89 |
"tiles": [
|
| 90 |
+
1.7687,
|
| 91 |
+
0.0144,
|
| 92 |
+
0.0141,
|
| 93 |
+
0.0141,
|
| 94 |
+
0.014,
|
| 95 |
+
0.014,
|
| 96 |
+
0.014,
|
| 97 |
+
0.0139,
|
| 98 |
+
0.0138,
|
| 99 |
+
0.0138,
|
| 100 |
+
0.0138,
|
| 101 |
+
0.0137,
|
| 102 |
+
0.0138,
|
| 103 |
+
0.0404
|
| 104 |
],
|
| 105 |
"cudnn_benchmark": false,
|
| 106 |
"cudnn_deterministic": true,
|
| 107 |
+
"core_frames": 256,
|
| 108 |
"input_device": "cpu",
|
| 109 |
"input_contiguous": false
|
| 110 |
},
|
| 111 |
"nar_prefill_source": "handoff",
|
| 112 |
+
"abc_close_seconds": 0.0003131115809082985,
|
| 113 |
+
"memory": {
|
| 114 |
+
"t0": {
|
| 115 |
+
"peak_gib": 2.479,
|
| 116 |
+
"alloc_gib": 2.479,
|
| 117 |
+
"reserved_gib": 2.756,
|
| 118 |
+
"nvml_used_gib": 3.674
|
| 119 |
+
},
|
| 120 |
+
"plan_done": {
|
| 121 |
+
"peak_gib": 2.754,
|
| 122 |
+
"alloc_gib": 2.479,
|
| 123 |
+
"reserved_gib": 2.98,
|
| 124 |
+
"nvml_used_gib": 3.922
|
| 125 |
+
},
|
| 126 |
+
"semantic_done": {
|
| 127 |
+
"peak_gib": 3.17,
|
| 128 |
+
"alloc_gib": 3.067,
|
| 129 |
+
"reserved_gib": 3.309,
|
| 130 |
+
"nvml_used_gib": 4.252
|
| 131 |
+
},
|
| 132 |
+
"nar_done": {
|
| 133 |
+
"peak_gib": 3.12,
|
| 134 |
+
"alloc_gib": 2.488,
|
| 135 |
+
"reserved_gib": 3.4,
|
| 136 |
+
"nvml_used_gib": 4.36
|
| 137 |
+
},
|
| 138 |
+
"vae_done": {
|
| 139 |
+
"peak_gib": 3.216,
|
| 140 |
+
"alloc_gib": 2.735,
|
| 141 |
+
"reserved_gib": 3.518,
|
| 142 |
+
"nvml_used_gib": 4.487
|
| 143 |
+
}
|
| 144 |
+
}
|
| 145 |
},
|
| 146 |
+
"e2e_seconds": 14.984941340982914
|
| 147 |
},
|
| 148 |
"artifacts": {
|
| 149 |
"abc_tokens.npy": {
|
|
|
|
| 151 |
"bytes": 5120
|
| 152 |
},
|
| 153 |
"audio.flac": {
|
| 154 |
+
"sha256": "fc46780bd611a5d5becc6393b16947edae5fe7bfcbfc408a371cc8b99356c247",
|
| 155 |
+
"bytes": 28760548
|
| 156 |
},
|
| 157 |
"config.json": {
|
| 158 |
+
"sha256": "1ca79e079b519059f4656324481ca68ddd6138f12dd333ba544811f339cb9c40",
|
| 159 |
"bytes": 2476
|
| 160 |
},
|
| 161 |
"latent.npy": {
|
| 162 |
+
"sha256": "20e1f3f90823599e0aed7c2aebdeb8873194cce116ade44bebd4c1921d2b62b4",
|
| 163 |
"bytes": 900480
|
| 164 |
},
|
| 165 |
"plan.json": {
|
| 166 |
+
"sha256": "1979add7ff86e1abccc806a796464b646f2418bdeaa3367c907974dfde143234",
|
| 167 |
+
"bytes": 30339
|
| 168 |
},
|
| 169 |
"plan_manifest.json": {
|
| 170 |
+
"sha256": "4c2599cdbf93d309218efb019039189bc67984ccb9360d6827ec88a58008e512",
|
| 171 |
"bytes": 341
|
| 172 |
},
|
| 173 |
"prefix.npy": {
|
evaluation/profiles/fast/run-01/config.json
CHANGED
|
@@ -34,12 +34,12 @@
|
|
| 34 |
"model_dtype": "bfloat16",
|
| 35 |
"vae_dtype": "float32",
|
| 36 |
"vae_decode": "halo_crop",
|
| 37 |
-
"vae_core_frames":
|
| 38 |
"vae_halo_frames": 16,
|
| 39 |
"device": "cuda:0",
|
| 40 |
"memory_budget_gib": 18.0,
|
| 41 |
"offload_ar": false,
|
| 42 |
-
"runtime_sha256": "
|
| 43 |
"decoder_release": "standard",
|
| 44 |
"validation_status": "unvalidated",
|
| 45 |
"runtime_profile": "fast",
|
|
@@ -47,7 +47,7 @@
|
|
| 47 |
"captured_sampling": true,
|
| 48 |
"kv_cache_dtype": "int8",
|
| 49 |
"int8_attention": "ordered",
|
| 50 |
-
"runtime_bundle_sha256": "
|
| 51 |
"vae_compute_dtype": "torch.float16",
|
| 52 |
"expert_staging": false,
|
| 53 |
"orbitquant": {
|
|
|
|
| 34 |
"model_dtype": "bfloat16",
|
| 35 |
"vae_dtype": "float32",
|
| 36 |
"vae_decode": "halo_crop",
|
| 37 |
+
"vae_core_frames": 256,
|
| 38 |
"vae_halo_frames": 16,
|
| 39 |
"device": "cuda:0",
|
| 40 |
"memory_budget_gib": 18.0,
|
| 41 |
"offload_ar": false,
|
| 42 |
+
"runtime_sha256": "bb189091c79b42dd3eca4f24f6239294df3c517fb2b149fc3b30ada3be287119",
|
| 43 |
"decoder_release": "standard",
|
| 44 |
"validation_status": "unvalidated",
|
| 45 |
"runtime_profile": "fast",
|
|
|
|
| 47 |
"captured_sampling": true,
|
| 48 |
"kv_cache_dtype": "int8",
|
| 49 |
"int8_attention": "ordered",
|
| 50 |
+
"runtime_bundle_sha256": "269214c6459739a0f08889532d5fc76f3e6f48f3908807466e14e5da4ecb1c98",
|
| 51 |
"vae_compute_dtype": "torch.float16",
|
| 52 |
"expert_staging": false,
|
| 53 |
"orbitquant": {
|
evaluation/profiles/fast/run-01/metrics.json
CHANGED
|
@@ -1,45 +1,46 @@
|
|
| 1 |
{
|
| 2 |
"bundle": "v11",
|
| 3 |
-
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,kv_handoff,
|
| 4 |
"flags": [
|
| 5 |
"abc_crop",
|
| 6 |
"int8_heads",
|
| 7 |
"kv_handoff",
|
|
|
|
| 8 |
"lookahead",
|
| 9 |
-
"nar_compile",
|
| 10 |
"resident",
|
| 11 |
"vae_compile",
|
|
|
|
| 12 |
"vae_mode=default",
|
| 13 |
"w8a8"
|
| 14 |
],
|
| 15 |
-
"native_sha256": "
|
| 16 |
"repeat": 1,
|
| 17 |
-
"wall_seconds":
|
| 18 |
-
"load_seconds":
|
| 19 |
"audio_seconds": 140.67866666666666,
|
| 20 |
-
"audio_per_wall":
|
| 21 |
-
"abc_s":
|
| 22 |
"abc_tokens": 1249,
|
| 23 |
-
"sem_s": 7.
|
| 24 |
"sem_tokens": 3518,
|
| 25 |
-
"abc_ms_per_tok": 2.
|
| 26 |
-
"sem_ms_per_tok": 2.
|
| 27 |
-
"nar_s":
|
| 28 |
-
"vae_s": 0.
|
| 29 |
-
"nvml_peak_gib": 4.
|
| 30 |
-
"torch_peak_gib":
|
| 31 |
-
"torch_reserved_peak_gib": 3.
|
| 32 |
"semantic_sha256": "bcb32247a0bd840b8b568746f93e93f823a87796b1f87fb55492da8eb741b473",
|
| 33 |
-
"waveform_sha256": "
|
| 34 |
-
"latent_sha256": "
|
| 35 |
"timing": {
|
| 36 |
"abc": {
|
| 37 |
-
"seconds":
|
| 38 |
-
"prefill_seconds": 0.
|
| 39 |
-
"ttft_seconds": 0.
|
| 40 |
"output_tokens": 1249,
|
| 41 |
"content_tokens": 1248,
|
| 42 |
-
"output_tps":
|
| 43 |
"prefix_tokens": 423,
|
| 44 |
"cfg_branches": 1,
|
| 45 |
"sampling": "combined_cuda_graph",
|
|
@@ -50,16 +51,16 @@
|
|
| 50 |
"lookahead": true,
|
| 51 |
"unroll2": false,
|
| 52 |
"first_step_seconds": null,
|
| 53 |
-
"capture_seconds": 0.
|
| 54 |
-
"close_seconds": 0.
|
| 55 |
},
|
| 56 |
"semantic": {
|
| 57 |
-
"seconds": 7.
|
| 58 |
-
"prefill_seconds": 0.
|
| 59 |
-
"ttft_seconds": 0.
|
| 60 |
"output_tokens": 3518,
|
| 61 |
"content_tokens": 3517,
|
| 62 |
-
"output_tps":
|
| 63 |
"prefix_tokens": 1673,
|
| 64 |
"cfg_branches": 1,
|
| 65 |
"sampling": "combined_cuda_graph",
|
|
@@ -70,46 +71,85 @@
|
|
| 70 |
"lookahead": true,
|
| 71 |
"unroll2": false,
|
| 72 |
"first_step_seconds": null,
|
| 73 |
-
"capture_seconds": 0.
|
| 74 |
-
"close_seconds": 0.
|
| 75 |
},
|
| 76 |
-
"nar_seconds":
|
| 77 |
-
"vae_seconds": 0.
|
| 78 |
"load": {
|
| 79 |
-
"resolve_and_integrity_seconds":
|
| 80 |
-
"mot_load_seconds":
|
| 81 |
},
|
| 82 |
"stages": {
|
| 83 |
-
"t0->plan_done":
|
| 84 |
-
"plan_done->semantic_done": 7.
|
| 85 |
-
"semantic_done->nar_done":
|
| 86 |
-
"nar_done->vae_done": 0.
|
| 87 |
-
"plan_outside_timer": 0.
|
| 88 |
-
"semantic_outside_timer": 0.
|
| 89 |
-
"semantic_close_seconds": 0.
|
| 90 |
"vae": {
|
| 91 |
-
"load_s": 0.
|
| 92 |
-
"decode_s": 0.
|
| 93 |
-
"post_s":
|
| 94 |
"tiles": [
|
| 95 |
-
0.
|
| 96 |
-
0.
|
| 97 |
-
0.
|
| 98 |
-
0.
|
| 99 |
-
0.
|
| 100 |
-
0.
|
| 101 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 102 |
],
|
| 103 |
"cudnn_benchmark": false,
|
| 104 |
"cudnn_deterministic": true,
|
| 105 |
-
"core_frames":
|
| 106 |
"input_device": "cpu",
|
| 107 |
"input_contiguous": false
|
| 108 |
},
|
| 109 |
"nar_prefill_source": "handoff",
|
| 110 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 111 |
},
|
| 112 |
-
"e2e_seconds":
|
| 113 |
},
|
| 114 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 115 |
"cfg": 1.0,
|
|
@@ -119,7 +159,7 @@
|
|
| 119 |
"kv_cache_dtype": "int8",
|
| 120 |
"captured_sampling": true,
|
| 121 |
"fuse_rms_quant": true,
|
| 122 |
-
"orbitquant_version": "0.9.
|
| 123 |
"profile": "fast",
|
| 124 |
"allocator_backend": "native",
|
| 125 |
"modules": 224,
|
|
|
|
| 1 |
{
|
| 2 |
"bundle": "v11",
|
| 3 |
+
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,vae_core=256,kv_handoff,lazy_handoff,int8_heads,w8a8",
|
| 4 |
"flags": [
|
| 5 |
"abc_crop",
|
| 6 |
"int8_heads",
|
| 7 |
"kv_handoff",
|
| 8 |
+
"lazy_handoff",
|
| 9 |
"lookahead",
|
|
|
|
| 10 |
"resident",
|
| 11 |
"vae_compile",
|
| 12 |
+
"vae_core=256",
|
| 13 |
"vae_mode=default",
|
| 14 |
"w8a8"
|
| 15 |
],
|
| 16 |
+
"native_sha256": "118d64db97c3200ef785e83b8d703ebc3b480a799bb17a6affb104d1b4faa28d",
|
| 17 |
"repeat": 1,
|
| 18 |
+
"wall_seconds": 12.26123733073473,
|
| 19 |
+
"load_seconds": 2.204909392632544,
|
| 20 |
"audio_seconds": 140.67866666666666,
|
| 21 |
+
"audio_per_wall": 11.473447815420172,
|
| 22 |
+
"abc_s": 2.8048877101391554,
|
| 23 |
"abc_tokens": 1249,
|
| 24 |
+
"sem_s": 7.337649843655527,
|
| 25 |
"sem_tokens": 3518,
|
| 26 |
+
"abc_ms_per_tok": 2.245706733498123,
|
| 27 |
+
"sem_ms_per_tok": 2.085744696889007,
|
| 28 |
+
"nar_s": 1.9093166394159198,
|
| 29 |
+
"vae_s": 0.20481676515191793,
|
| 30 |
+
"nvml_peak_gib": 4.635009765625,
|
| 31 |
+
"torch_peak_gib": 2.735057830810547,
|
| 32 |
+
"torch_reserved_peak_gib": 3.654296875,
|
| 33 |
"semantic_sha256": "bcb32247a0bd840b8b568746f93e93f823a87796b1f87fb55492da8eb741b473",
|
| 34 |
+
"waveform_sha256": "e0984b3364956b3d27d98703ee965d578951863b358ccff6769e544a15a77648",
|
| 35 |
+
"latent_sha256": "1763ac7941cbedffae6c43ca33e947ca9c8c390ae0d290db8b6c426e97439b36",
|
| 36 |
"timing": {
|
| 37 |
"abc": {
|
| 38 |
+
"seconds": 2.8048877101391554,
|
| 39 |
+
"prefill_seconds": 0.0390327638015151,
|
| 40 |
+
"ttft_seconds": 0.0390746770426631,
|
| 41 |
"output_tokens": 1249,
|
| 42 |
"content_tokens": 1248,
|
| 43 |
+
"output_tps": 445.29411836527134,
|
| 44 |
"prefix_tokens": 423,
|
| 45 |
"cfg_branches": 1,
|
| 46 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 51 |
"lookahead": true,
|
| 52 |
"unroll2": false,
|
| 53 |
"first_step_seconds": null,
|
| 54 |
+
"capture_seconds": 0.035048287361860275,
|
| 55 |
+
"close_seconds": 0.00037193018943071365
|
| 56 |
},
|
| 57 |
"semantic": {
|
| 58 |
+
"seconds": 7.337649843655527,
|
| 59 |
+
"prefill_seconds": 0.048011607490479946,
|
| 60 |
+
"ttft_seconds": 0.048030185513198376,
|
| 61 |
"output_tokens": 3518,
|
| 62 |
"content_tokens": 3517,
|
| 63 |
+
"output_tps": 479.4450641497736,
|
| 64 |
"prefix_tokens": 1673,
|
| 65 |
"cfg_branches": 1,
|
| 66 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 71 |
"lookahead": true,
|
| 72 |
"unroll2": false,
|
| 73 |
"first_step_seconds": null,
|
| 74 |
+
"capture_seconds": 0.030972903594374657,
|
| 75 |
+
"close_seconds": 0.0003538327291607857
|
| 76 |
},
|
| 77 |
+
"nar_seconds": 1.9093166394159198,
|
| 78 |
+
"vae_seconds": 0.20481676515191793,
|
| 79 |
"load": {
|
| 80 |
+
"resolve_and_integrity_seconds": 0.19479286018759012,
|
| 81 |
+
"mot_load_seconds": 0.6600089315325022
|
| 82 |
},
|
| 83 |
"stages": {
|
| 84 |
+
"t0->plan_done": 2.806506736204028,
|
| 85 |
+
"plan_done->semantic_done": 7.338900384493172,
|
| 86 |
+
"semantic_done->nar_done": 1.9096503229811788,
|
| 87 |
+
"nar_done->vae_done": 0.2051340825855732,
|
| 88 |
+
"plan_outside_timer": 0.0016190260648727417,
|
| 89 |
+
"semantic_outside_timer": 0.0012505408376455307,
|
| 90 |
+
"semantic_close_seconds": 0.0003538327291607857,
|
| 91 |
"vae": {
|
| 92 |
+
"load_s": 0.0008637299761176109,
|
| 93 |
+
"decode_s": 0.20389806292951107,
|
| 94 |
+
"post_s": 1.0485760867595673e-05,
|
| 95 |
"tiles": [
|
| 96 |
+
0.0141,
|
| 97 |
+
0.0137,
|
| 98 |
+
0.0136,
|
| 99 |
+
0.0137,
|
| 100 |
+
0.0136,
|
| 101 |
+
0.0136,
|
| 102 |
+
0.0136,
|
| 103 |
+
0.0136,
|
| 104 |
+
0.0137,
|
| 105 |
+
0.0136,
|
| 106 |
+
0.0135,
|
| 107 |
+
0.0135,
|
| 108 |
+
0.0135,
|
| 109 |
+
0.0264
|
| 110 |
],
|
| 111 |
"cudnn_benchmark": false,
|
| 112 |
"cudnn_deterministic": true,
|
| 113 |
+
"core_frames": 256,
|
| 114 |
"input_device": "cpu",
|
| 115 |
"input_contiguous": false
|
| 116 |
},
|
| 117 |
"nar_prefill_source": "handoff",
|
| 118 |
+
"abc_close_seconds": 0.00037193018943071365,
|
| 119 |
+
"memory": {
|
| 120 |
+
"t0": {
|
| 121 |
+
"peak_gib": 2.735,
|
| 122 |
+
"alloc_gib": 2.735,
|
| 123 |
+
"reserved_gib": 3.518,
|
| 124 |
+
"nvml_used_gib": 4.487
|
| 125 |
+
},
|
| 126 |
+
"plan_done": {
|
| 127 |
+
"peak_gib": 3.01,
|
| 128 |
+
"alloc_gib": 2.735,
|
| 129 |
+
"reserved_gib": 3.215,
|
| 130 |
+
"nvml_used_gib": 4.184
|
| 131 |
+
},
|
| 132 |
+
"semantic_done": {
|
| 133 |
+
"peak_gib": 3.426,
|
| 134 |
+
"alloc_gib": 3.323,
|
| 135 |
+
"reserved_gib": 3.543,
|
| 136 |
+
"nvml_used_gib": 4.512
|
| 137 |
+
},
|
| 138 |
+
"nar_done": {
|
| 139 |
+
"peak_gib": 3.368,
|
| 140 |
+
"alloc_gib": 2.735,
|
| 141 |
+
"reserved_gib": 3.654,
|
| 142 |
+
"nvml_used_gib": 4.623
|
| 143 |
+
},
|
| 144 |
+
"vae_done": {
|
| 145 |
+
"peak_gib": 3.216,
|
| 146 |
+
"alloc_gib": 2.735,
|
| 147 |
+
"reserved_gib": 3.654,
|
| 148 |
+
"nvml_used_gib": 4.623
|
| 149 |
+
}
|
| 150 |
+
}
|
| 151 |
},
|
| 152 |
+
"e2e_seconds": 12.260506701655686
|
| 153 |
},
|
| 154 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 155 |
"cfg": 1.0,
|
|
|
|
| 159 |
"kv_cache_dtype": "int8",
|
| 160 |
"captured_sampling": true,
|
| 161 |
"fuse_rms_quant": true,
|
| 162 |
+
"orbitquant_version": "0.9.8",
|
| 163 |
"profile": "fast",
|
| 164 |
"allocator_backend": "native",
|
| 165 |
"modules": 224,
|
evaluation/profiles/fast/run-01/plan.json
CHANGED
|
@@ -9,12 +9,12 @@
|
|
| 9 |
"id": "song"
|
| 10 |
},
|
| 11 |
"timing": {
|
| 12 |
-
"seconds":
|
| 13 |
-
"prefill_seconds": 0.
|
| 14 |
-
"ttft_seconds": 0.
|
| 15 |
"output_tokens": 1249,
|
| 16 |
"content_tokens": 1248,
|
| 17 |
-
"output_tps":
|
| 18 |
"prefix_tokens": 423,
|
| 19 |
"cfg_branches": 1,
|
| 20 |
"sampling": "combined_cuda_graph",
|
|
@@ -25,8 +25,8 @@
|
|
| 25 |
"lookahead": true,
|
| 26 |
"unroll2": false,
|
| 27 |
"first_step_seconds": null,
|
| 28 |
-
"capture_seconds": 0.
|
| 29 |
-
"close_seconds": 0.
|
| 30 |
},
|
| 31 |
"truncated": false,
|
| 32 |
"prefix": [
|
|
|
|
| 9 |
"id": "song"
|
| 10 |
},
|
| 11 |
"timing": {
|
| 12 |
+
"seconds": 2.8048877101391554,
|
| 13 |
+
"prefill_seconds": 0.0390327638015151,
|
| 14 |
+
"ttft_seconds": 0.0390746770426631,
|
| 15 |
"output_tokens": 1249,
|
| 16 |
"content_tokens": 1248,
|
| 17 |
+
"output_tps": 445.29411836527134,
|
| 18 |
"prefix_tokens": 423,
|
| 19 |
"cfg_branches": 1,
|
| 20 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 25 |
"lookahead": true,
|
| 26 |
"unroll2": false,
|
| 27 |
"first_step_seconds": null,
|
| 28 |
+
"capture_seconds": 0.035048287361860275,
|
| 29 |
+
"close_seconds": 0.00037193018943071365
|
| 30 |
},
|
| 31 |
"truncated": false,
|
| 32 |
"prefix": [
|
evaluation/profiles/fast/run-01/plan_manifest.json
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
{
|
| 2 |
-
"plan.json": "
|
| 3 |
"abc_tokens.npy": "2a3f7d12e43955d2b4d6ec2c08f926be3a76a2fdf22a2952fa37148f6ca9c196",
|
| 4 |
"prefix.npy": "b542a03e835df9d2f05a4c2ffdaaf9eca5e2c49b510e51b6560f0d985952bf85",
|
| 5 |
"score.abc": "8d7ed2e1687e35072e41ff1b64b852068ff81d496bb8142d83019b53b66498ab"
|
|
|
|
| 1 |
{
|
| 2 |
+
"plan.json": "5094ca63eae20eaec46f49a19e87a1c4f2668d5a8a7c799aacf408055659a9c1",
|
| 3 |
"abc_tokens.npy": "2a3f7d12e43955d2b4d6ec2c08f926be3a76a2fdf22a2952fa37148f6ca9c196",
|
| 4 |
"prefix.npy": "b542a03e835df9d2f05a4c2ffdaaf9eca5e2c49b510e51b6560f0d985952bf85",
|
| 5 |
"score.abc": "8d7ed2e1687e35072e41ff1b64b852068ff81d496bb8142d83019b53b66498ab"
|
evaluation/profiles/fast/run-01/result.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"status": "complete",
|
| 3 |
-
"identity": "
|
| 4 |
"truncated": {
|
| 5 |
"abc": false,
|
| 6 |
"semantic": false
|
|
@@ -29,12 +29,12 @@
|
|
| 29 |
},
|
| 30 |
"timing": {
|
| 31 |
"abc": {
|
| 32 |
-
"seconds":
|
| 33 |
-
"prefill_seconds": 0.
|
| 34 |
-
"ttft_seconds": 0.
|
| 35 |
"output_tokens": 1249,
|
| 36 |
"content_tokens": 1248,
|
| 37 |
-
"output_tps":
|
| 38 |
"prefix_tokens": 423,
|
| 39 |
"cfg_branches": 1,
|
| 40 |
"sampling": "combined_cuda_graph",
|
|
@@ -45,16 +45,16 @@
|
|
| 45 |
"lookahead": true,
|
| 46 |
"unroll2": false,
|
| 47 |
"first_step_seconds": null,
|
| 48 |
-
"capture_seconds": 0.
|
| 49 |
-
"close_seconds": 0.
|
| 50 |
},
|
| 51 |
"semantic": {
|
| 52 |
-
"seconds": 7.
|
| 53 |
-
"prefill_seconds": 0.
|
| 54 |
-
"ttft_seconds": 0.
|
| 55 |
"output_tokens": 3518,
|
| 56 |
"content_tokens": 3517,
|
| 57 |
-
"output_tps":
|
| 58 |
"prefix_tokens": 1673,
|
| 59 |
"cfg_branches": 1,
|
| 60 |
"sampling": "combined_cuda_graph",
|
|
@@ -65,46 +65,85 @@
|
|
| 65 |
"lookahead": true,
|
| 66 |
"unroll2": false,
|
| 67 |
"first_step_seconds": null,
|
| 68 |
-
"capture_seconds": 0.
|
| 69 |
-
"close_seconds": 0.
|
| 70 |
},
|
| 71 |
-
"nar_seconds":
|
| 72 |
-
"vae_seconds": 0.
|
| 73 |
"load": {
|
| 74 |
-
"resolve_and_integrity_seconds":
|
| 75 |
-
"mot_load_seconds":
|
| 76 |
},
|
| 77 |
"stages": {
|
| 78 |
-
"t0->plan_done":
|
| 79 |
-
"plan_done->semantic_done": 7.
|
| 80 |
-
"semantic_done->nar_done":
|
| 81 |
-
"nar_done->vae_done": 0.
|
| 82 |
-
"plan_outside_timer": 0.
|
| 83 |
-
"semantic_outside_timer": 0.
|
| 84 |
-
"semantic_close_seconds": 0.
|
| 85 |
"vae": {
|
| 86 |
-
"load_s": 0.
|
| 87 |
-
"decode_s": 0.
|
| 88 |
-
"post_s":
|
| 89 |
"tiles": [
|
| 90 |
-
0.
|
| 91 |
-
0.
|
| 92 |
-
0.
|
| 93 |
-
0.
|
| 94 |
-
0.
|
| 95 |
-
0.
|
| 96 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 97 |
],
|
| 98 |
"cudnn_benchmark": false,
|
| 99 |
"cudnn_deterministic": true,
|
| 100 |
-
"core_frames":
|
| 101 |
"input_device": "cpu",
|
| 102 |
"input_contiguous": false
|
| 103 |
},
|
| 104 |
"nar_prefill_source": "handoff",
|
| 105 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 106 |
},
|
| 107 |
-
"e2e_seconds":
|
| 108 |
},
|
| 109 |
"artifacts": {
|
| 110 |
"abc_tokens.npy": {
|
|
@@ -112,23 +151,23 @@
|
|
| 112 |
"bytes": 5120
|
| 113 |
},
|
| 114 |
"audio.flac": {
|
| 115 |
-
"sha256": "
|
| 116 |
-
"bytes":
|
| 117 |
},
|
| 118 |
"config.json": {
|
| 119 |
-
"sha256": "
|
| 120 |
"bytes": 2476
|
| 121 |
},
|
| 122 |
"latent.npy": {
|
| 123 |
-
"sha256": "
|
| 124 |
"bytes": 900480
|
| 125 |
},
|
| 126 |
"plan.json": {
|
| 127 |
-
"sha256": "
|
| 128 |
-
"bytes":
|
| 129 |
},
|
| 130 |
"plan_manifest.json": {
|
| 131 |
-
"sha256": "
|
| 132 |
"bytes": 341
|
| 133 |
},
|
| 134 |
"prefix.npy": {
|
|
|
|
| 1 |
{
|
| 2 |
"status": "complete",
|
| 3 |
+
"identity": "5b78b1460d137eb9aa61003dc8c71bf596658015622424138dc69547e7213123",
|
| 4 |
"truncated": {
|
| 5 |
"abc": false,
|
| 6 |
"semantic": false
|
|
|
|
| 29 |
},
|
| 30 |
"timing": {
|
| 31 |
"abc": {
|
| 32 |
+
"seconds": 2.8048877101391554,
|
| 33 |
+
"prefill_seconds": 0.0390327638015151,
|
| 34 |
+
"ttft_seconds": 0.0390746770426631,
|
| 35 |
"output_tokens": 1249,
|
| 36 |
"content_tokens": 1248,
|
| 37 |
+
"output_tps": 445.29411836527134,
|
| 38 |
"prefix_tokens": 423,
|
| 39 |
"cfg_branches": 1,
|
| 40 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 45 |
"lookahead": true,
|
| 46 |
"unroll2": false,
|
| 47 |
"first_step_seconds": null,
|
| 48 |
+
"capture_seconds": 0.035048287361860275,
|
| 49 |
+
"close_seconds": 0.00037193018943071365
|
| 50 |
},
|
| 51 |
"semantic": {
|
| 52 |
+
"seconds": 7.337649843655527,
|
| 53 |
+
"prefill_seconds": 0.048011607490479946,
|
| 54 |
+
"ttft_seconds": 0.048030185513198376,
|
| 55 |
"output_tokens": 3518,
|
| 56 |
"content_tokens": 3517,
|
| 57 |
+
"output_tps": 479.4450641497736,
|
| 58 |
"prefix_tokens": 1673,
|
| 59 |
"cfg_branches": 1,
|
| 60 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 65 |
"lookahead": true,
|
| 66 |
"unroll2": false,
|
| 67 |
"first_step_seconds": null,
|
| 68 |
+
"capture_seconds": 0.030972903594374657,
|
| 69 |
+
"close_seconds": 0.0003538327291607857
|
| 70 |
},
|
| 71 |
+
"nar_seconds": 1.9093166394159198,
|
| 72 |
+
"vae_seconds": 0.20481676515191793,
|
| 73 |
"load": {
|
| 74 |
+
"resolve_and_integrity_seconds": 0.19479286018759012,
|
| 75 |
+
"mot_load_seconds": 0.6600089315325022
|
| 76 |
},
|
| 77 |
"stages": {
|
| 78 |
+
"t0->plan_done": 2.806506736204028,
|
| 79 |
+
"plan_done->semantic_done": 7.338900384493172,
|
| 80 |
+
"semantic_done->nar_done": 1.9096503229811788,
|
| 81 |
+
"nar_done->vae_done": 0.2051340825855732,
|
| 82 |
+
"plan_outside_timer": 0.0016190260648727417,
|
| 83 |
+
"semantic_outside_timer": 0.0012505408376455307,
|
| 84 |
+
"semantic_close_seconds": 0.0003538327291607857,
|
| 85 |
"vae": {
|
| 86 |
+
"load_s": 0.0008637299761176109,
|
| 87 |
+
"decode_s": 0.20389806292951107,
|
| 88 |
+
"post_s": 1.0485760867595673e-05,
|
| 89 |
"tiles": [
|
| 90 |
+
0.0141,
|
| 91 |
+
0.0137,
|
| 92 |
+
0.0136,
|
| 93 |
+
0.0137,
|
| 94 |
+
0.0136,
|
| 95 |
+
0.0136,
|
| 96 |
+
0.0136,
|
| 97 |
+
0.0136,
|
| 98 |
+
0.0137,
|
| 99 |
+
0.0136,
|
| 100 |
+
0.0135,
|
| 101 |
+
0.0135,
|
| 102 |
+
0.0135,
|
| 103 |
+
0.0264
|
| 104 |
],
|
| 105 |
"cudnn_benchmark": false,
|
| 106 |
"cudnn_deterministic": true,
|
| 107 |
+
"core_frames": 256,
|
| 108 |
"input_device": "cpu",
|
| 109 |
"input_contiguous": false
|
| 110 |
},
|
| 111 |
"nar_prefill_source": "handoff",
|
| 112 |
+
"abc_close_seconds": 0.00037193018943071365,
|
| 113 |
+
"memory": {
|
| 114 |
+
"t0": {
|
| 115 |
+
"peak_gib": 2.735,
|
| 116 |
+
"alloc_gib": 2.735,
|
| 117 |
+
"reserved_gib": 3.518,
|
| 118 |
+
"nvml_used_gib": 4.487
|
| 119 |
+
},
|
| 120 |
+
"plan_done": {
|
| 121 |
+
"peak_gib": 3.01,
|
| 122 |
+
"alloc_gib": 2.735,
|
| 123 |
+
"reserved_gib": 3.215,
|
| 124 |
+
"nvml_used_gib": 4.184
|
| 125 |
+
},
|
| 126 |
+
"semantic_done": {
|
| 127 |
+
"peak_gib": 3.426,
|
| 128 |
+
"alloc_gib": 3.323,
|
| 129 |
+
"reserved_gib": 3.543,
|
| 130 |
+
"nvml_used_gib": 4.512
|
| 131 |
+
},
|
| 132 |
+
"nar_done": {
|
| 133 |
+
"peak_gib": 3.368,
|
| 134 |
+
"alloc_gib": 2.735,
|
| 135 |
+
"reserved_gib": 3.654,
|
| 136 |
+
"nvml_used_gib": 4.623
|
| 137 |
+
},
|
| 138 |
+
"vae_done": {
|
| 139 |
+
"peak_gib": 3.216,
|
| 140 |
+
"alloc_gib": 2.735,
|
| 141 |
+
"reserved_gib": 3.654,
|
| 142 |
+
"nvml_used_gib": 4.623
|
| 143 |
+
}
|
| 144 |
+
}
|
| 145 |
},
|
| 146 |
+
"e2e_seconds": 12.260506701655686
|
| 147 |
},
|
| 148 |
"artifacts": {
|
| 149 |
"abc_tokens.npy": {
|
|
|
|
| 151 |
"bytes": 5120
|
| 152 |
},
|
| 153 |
"audio.flac": {
|
| 154 |
+
"sha256": "fc46780bd611a5d5becc6393b16947edae5fe7bfcbfc408a371cc8b99356c247",
|
| 155 |
+
"bytes": 28760548
|
| 156 |
},
|
| 157 |
"config.json": {
|
| 158 |
+
"sha256": "1ca79e079b519059f4656324481ca68ddd6138f12dd333ba544811f339cb9c40",
|
| 159 |
"bytes": 2476
|
| 160 |
},
|
| 161 |
"latent.npy": {
|
| 162 |
+
"sha256": "20e1f3f90823599e0aed7c2aebdeb8873194cce116ade44bebd4c1921d2b62b4",
|
| 163 |
"bytes": 900480
|
| 164 |
},
|
| 165 |
"plan.json": {
|
| 166 |
+
"sha256": "5094ca63eae20eaec46f49a19e87a1c4f2668d5a8a7c799aacf408055659a9c1",
|
| 167 |
+
"bytes": 30340
|
| 168 |
},
|
| 169 |
"plan_manifest.json": {
|
| 170 |
+
"sha256": "b17db0a80a096b20a98aac9ffba004946f3677da2fb5b84efb3a8b75e838eed2",
|
| 171 |
"bytes": 341
|
| 172 |
},
|
| 173 |
"prefix.npy": {
|
evaluation/profiles/fast/summary.json
CHANGED
|
@@ -1,46 +1,47 @@
|
|
| 1 |
[
|
| 2 |
{
|
| 3 |
"bundle": "v11",
|
| 4 |
-
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,kv_handoff,
|
| 5 |
"flags": [
|
| 6 |
"abc_crop",
|
| 7 |
"int8_heads",
|
| 8 |
"kv_handoff",
|
|
|
|
| 9 |
"lookahead",
|
| 10 |
-
"nar_compile",
|
| 11 |
"resident",
|
| 12 |
"vae_compile",
|
|
|
|
| 13 |
"vae_mode=default",
|
| 14 |
"w8a8"
|
| 15 |
],
|
| 16 |
-
"native_sha256": "
|
| 17 |
"repeat": 0,
|
| 18 |
-
"wall_seconds":
|
| 19 |
-
"load_seconds":
|
| 20 |
"audio_seconds": 140.67866666666666,
|
| 21 |
-
"audio_per_wall":
|
| 22 |
-
"abc_s":
|
| 23 |
"abc_tokens": 1249,
|
| 24 |
-
"sem_s": 7.
|
| 25 |
"sem_tokens": 3518,
|
| 26 |
-
"abc_ms_per_tok":
|
| 27 |
-
"sem_ms_per_tok": 2.
|
| 28 |
-
"nar_s":
|
| 29 |
-
"vae_s":
|
| 30 |
-
"nvml_peak_gib": 4.
|
| 31 |
-
"torch_peak_gib":
|
| 32 |
-
"torch_reserved_peak_gib": 3.
|
| 33 |
"semantic_sha256": "bcb32247a0bd840b8b568746f93e93f823a87796b1f87fb55492da8eb741b473",
|
| 34 |
-
"waveform_sha256": "
|
| 35 |
-
"latent_sha256": "
|
| 36 |
"timing": {
|
| 37 |
"abc": {
|
| 38 |
-
"seconds":
|
| 39 |
-
"prefill_seconds":
|
| 40 |
-
"ttft_seconds":
|
| 41 |
"output_tokens": 1249,
|
| 42 |
"content_tokens": 1248,
|
| 43 |
-
"output_tps":
|
| 44 |
"prefix_tokens": 423,
|
| 45 |
"cfg_branches": 1,
|
| 46 |
"sampling": "combined_cuda_graph",
|
|
@@ -51,16 +52,16 @@
|
|
| 51 |
"lookahead": true,
|
| 52 |
"unroll2": false,
|
| 53 |
"first_step_seconds": null,
|
| 54 |
-
"capture_seconds": 0.
|
| 55 |
-
"close_seconds": 0.
|
| 56 |
},
|
| 57 |
"semantic": {
|
| 58 |
-
"seconds": 7.
|
| 59 |
-
"prefill_seconds": 0.
|
| 60 |
-
"ttft_seconds": 0.
|
| 61 |
"output_tokens": 3518,
|
| 62 |
"content_tokens": 3517,
|
| 63 |
-
"output_tps":
|
| 64 |
"prefix_tokens": 1673,
|
| 65 |
"cfg_branches": 1,
|
| 66 |
"sampling": "combined_cuda_graph",
|
|
@@ -71,46 +72,85 @@
|
|
| 71 |
"lookahead": true,
|
| 72 |
"unroll2": false,
|
| 73 |
"first_step_seconds": null,
|
| 74 |
-
"capture_seconds": 0.
|
| 75 |
-
"close_seconds": 0.
|
| 76 |
},
|
| 77 |
-
"nar_seconds":
|
| 78 |
-
"vae_seconds":
|
| 79 |
"load": {
|
| 80 |
-
"resolve_and_integrity_seconds":
|
| 81 |
-
"mot_load_seconds":
|
| 82 |
},
|
| 83 |
"stages": {
|
| 84 |
-
"t0->plan_done":
|
| 85 |
-
"plan_done->semantic_done": 7.
|
| 86 |
-
"semantic_done->nar_done":
|
| 87 |
-
"nar_done->vae_done":
|
| 88 |
-
"plan_outside_timer": 0.
|
| 89 |
-
"semantic_outside_timer": 0.
|
| 90 |
-
"semantic_close_seconds": 0.
|
| 91 |
"vae": {
|
| 92 |
-
"load_s":
|
| 93 |
-
"decode_s":
|
| 94 |
-
"post_s":
|
| 95 |
"tiles": [
|
| 96 |
-
|
| 97 |
-
0.
|
| 98 |
-
0.
|
| 99 |
-
0.
|
| 100 |
-
0.
|
| 101 |
-
0.
|
| 102 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 103 |
],
|
| 104 |
"cudnn_benchmark": false,
|
| 105 |
"cudnn_deterministic": true,
|
| 106 |
-
"core_frames":
|
| 107 |
"input_device": "cpu",
|
| 108 |
"input_contiguous": false
|
| 109 |
},
|
| 110 |
"nar_prefill_source": "handoff",
|
| 111 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 112 |
},
|
| 113 |
-
"e2e_seconds":
|
| 114 |
},
|
| 115 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 116 |
"cfg": 1.0,
|
|
@@ -120,7 +160,7 @@
|
|
| 120 |
"kv_cache_dtype": "int8",
|
| 121 |
"captured_sampling": true,
|
| 122 |
"fuse_rms_quant": true,
|
| 123 |
-
"orbitquant_version": "0.9.
|
| 124 |
"profile": "fast",
|
| 125 |
"allocator_backend": "native",
|
| 126 |
"modules": 224,
|
|
@@ -143,46 +183,47 @@
|
|
| 143 |
},
|
| 144 |
{
|
| 145 |
"bundle": "v11",
|
| 146 |
-
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,kv_handoff,
|
| 147 |
"flags": [
|
| 148 |
"abc_crop",
|
| 149 |
"int8_heads",
|
| 150 |
"kv_handoff",
|
|
|
|
| 151 |
"lookahead",
|
| 152 |
-
"nar_compile",
|
| 153 |
"resident",
|
| 154 |
"vae_compile",
|
|
|
|
| 155 |
"vae_mode=default",
|
| 156 |
"w8a8"
|
| 157 |
],
|
| 158 |
-
"native_sha256": "
|
| 159 |
"repeat": 1,
|
| 160 |
-
"wall_seconds":
|
| 161 |
-
"load_seconds":
|
| 162 |
"audio_seconds": 140.67866666666666,
|
| 163 |
-
"audio_per_wall":
|
| 164 |
-
"abc_s":
|
| 165 |
"abc_tokens": 1249,
|
| 166 |
-
"sem_s": 7.
|
| 167 |
"sem_tokens": 3518,
|
| 168 |
-
"abc_ms_per_tok": 2.
|
| 169 |
-
"sem_ms_per_tok": 2.
|
| 170 |
-
"nar_s":
|
| 171 |
-
"vae_s": 0.
|
| 172 |
-
"nvml_peak_gib": 4.
|
| 173 |
-
"torch_peak_gib":
|
| 174 |
-
"torch_reserved_peak_gib": 3.
|
| 175 |
"semantic_sha256": "bcb32247a0bd840b8b568746f93e93f823a87796b1f87fb55492da8eb741b473",
|
| 176 |
-
"waveform_sha256": "
|
| 177 |
-
"latent_sha256": "
|
| 178 |
"timing": {
|
| 179 |
"abc": {
|
| 180 |
-
"seconds":
|
| 181 |
-
"prefill_seconds": 0.
|
| 182 |
-
"ttft_seconds": 0.
|
| 183 |
"output_tokens": 1249,
|
| 184 |
"content_tokens": 1248,
|
| 185 |
-
"output_tps":
|
| 186 |
"prefix_tokens": 423,
|
| 187 |
"cfg_branches": 1,
|
| 188 |
"sampling": "combined_cuda_graph",
|
|
@@ -193,16 +234,16 @@
|
|
| 193 |
"lookahead": true,
|
| 194 |
"unroll2": false,
|
| 195 |
"first_step_seconds": null,
|
| 196 |
-
"capture_seconds": 0.
|
| 197 |
-
"close_seconds": 0.
|
| 198 |
},
|
| 199 |
"semantic": {
|
| 200 |
-
"seconds": 7.
|
| 201 |
-
"prefill_seconds": 0.
|
| 202 |
-
"ttft_seconds": 0.
|
| 203 |
"output_tokens": 3518,
|
| 204 |
"content_tokens": 3517,
|
| 205 |
-
"output_tps":
|
| 206 |
"prefix_tokens": 1673,
|
| 207 |
"cfg_branches": 1,
|
| 208 |
"sampling": "combined_cuda_graph",
|
|
@@ -213,46 +254,85 @@
|
|
| 213 |
"lookahead": true,
|
| 214 |
"unroll2": false,
|
| 215 |
"first_step_seconds": null,
|
| 216 |
-
"capture_seconds": 0.
|
| 217 |
-
"close_seconds": 0.
|
| 218 |
},
|
| 219 |
-
"nar_seconds":
|
| 220 |
-
"vae_seconds": 0.
|
| 221 |
"load": {
|
| 222 |
-
"resolve_and_integrity_seconds":
|
| 223 |
-
"mot_load_seconds":
|
| 224 |
},
|
| 225 |
"stages": {
|
| 226 |
-
"t0->plan_done":
|
| 227 |
-
"plan_done->semantic_done": 7.
|
| 228 |
-
"semantic_done->nar_done":
|
| 229 |
-
"nar_done->vae_done": 0.
|
| 230 |
-
"plan_outside_timer": 0.
|
| 231 |
-
"semantic_outside_timer": 0.
|
| 232 |
-
"semantic_close_seconds": 0.
|
| 233 |
"vae": {
|
| 234 |
-
"load_s": 0.
|
| 235 |
-
"decode_s": 0.
|
| 236 |
-
"post_s":
|
| 237 |
"tiles": [
|
| 238 |
-
0.
|
| 239 |
-
0.
|
| 240 |
-
0.
|
| 241 |
-
0.
|
| 242 |
-
0.
|
| 243 |
-
0.
|
| 244 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 245 |
],
|
| 246 |
"cudnn_benchmark": false,
|
| 247 |
"cudnn_deterministic": true,
|
| 248 |
-
"core_frames":
|
| 249 |
"input_device": "cpu",
|
| 250 |
"input_contiguous": false
|
| 251 |
},
|
| 252 |
"nar_prefill_source": "handoff",
|
| 253 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 254 |
},
|
| 255 |
-
"e2e_seconds":
|
| 256 |
},
|
| 257 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 258 |
"cfg": 1.0,
|
|
@@ -262,7 +342,7 @@
|
|
| 262 |
"kv_cache_dtype": "int8",
|
| 263 |
"captured_sampling": true,
|
| 264 |
"fuse_rms_quant": true,
|
| 265 |
-
"orbitquant_version": "0.9.
|
| 266 |
"profile": "fast",
|
| 267 |
"allocator_backend": "native",
|
| 268 |
"modules": 224,
|
|
|
|
| 1 |
[
|
| 2 |
{
|
| 3 |
"bundle": "v11",
|
| 4 |
+
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,vae_core=256,kv_handoff,lazy_handoff,int8_heads,w8a8",
|
| 5 |
"flags": [
|
| 6 |
"abc_crop",
|
| 7 |
"int8_heads",
|
| 8 |
"kv_handoff",
|
| 9 |
+
"lazy_handoff",
|
| 10 |
"lookahead",
|
|
|
|
| 11 |
"resident",
|
| 12 |
"vae_compile",
|
| 13 |
+
"vae_core=256",
|
| 14 |
"vae_mode=default",
|
| 15 |
"w8a8"
|
| 16 |
],
|
| 17 |
+
"native_sha256": "118d64db97c3200ef785e83b8d703ebc3b480a799bb17a6affb104d1b4faa28d",
|
| 18 |
"repeat": 0,
|
| 19 |
+
"wall_seconds": 14.985645630396903,
|
| 20 |
+
"load_seconds": 2.204909392632544,
|
| 21 |
"audio_seconds": 140.67866666666666,
|
| 22 |
+
"audio_per_wall": 9.38756127939619,
|
| 23 |
+
"abc_s": 3.2614384908229113,
|
| 24 |
"abc_tokens": 1249,
|
| 25 |
+
"sem_s": 7.348635695874691,
|
| 26 |
"sem_tokens": 3518,
|
| 27 |
+
"abc_ms_per_tok": 2.6112397844859174,
|
| 28 |
+
"sem_ms_per_tok": 2.0888674519257227,
|
| 29 |
+
"nar_s": 2.046467389911413,
|
| 30 |
+
"vae_s": 2.3237511478364468,
|
| 31 |
+
"nvml_peak_gib": 4.486572265625,
|
| 32 |
+
"torch_peak_gib": 2.735057830810547,
|
| 33 |
+
"torch_reserved_peak_gib": 3.517578125,
|
| 34 |
"semantic_sha256": "bcb32247a0bd840b8b568746f93e93f823a87796b1f87fb55492da8eb741b473",
|
| 35 |
+
"waveform_sha256": "e0984b3364956b3d27d98703ee965d578951863b358ccff6769e544a15a77648",
|
| 36 |
+
"latent_sha256": "1763ac7941cbedffae6c43ca33e947ca9c8c390ae0d290db8b6c426e97439b36",
|
| 37 |
"timing": {
|
| 38 |
"abc": {
|
| 39 |
+
"seconds": 3.2614384908229113,
|
| 40 |
+
"prefill_seconds": 0.4902397319674492,
|
| 41 |
+
"ttft_seconds": 0.49129611160606146,
|
| 42 |
"output_tokens": 1249,
|
| 43 |
"content_tokens": 1248,
|
| 44 |
+
"output_tps": 382.9598514626158,
|
| 45 |
"prefix_tokens": 423,
|
| 46 |
"cfg_branches": 1,
|
| 47 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 52 |
"lookahead": true,
|
| 53 |
"unroll2": false,
|
| 54 |
"first_step_seconds": null,
|
| 55 |
+
"capture_seconds": 0.039790925569832325,
|
| 56 |
+
"close_seconds": 0.0003131115809082985
|
| 57 |
},
|
| 58 |
"semantic": {
|
| 59 |
+
"seconds": 7.348635695874691,
|
| 60 |
+
"prefill_seconds": 0.056065828539431095,
|
| 61 |
+
"ttft_seconds": 0.05608602799475193,
|
| 62 |
"output_tokens": 3518,
|
| 63 |
"content_tokens": 3517,
|
| 64 |
+
"output_tps": 478.7283171453039,
|
| 65 |
"prefix_tokens": 1673,
|
| 66 |
"cfg_branches": 1,
|
| 67 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 72 |
"lookahead": true,
|
| 73 |
"unroll2": false,
|
| 74 |
"first_step_seconds": null,
|
| 75 |
+
"capture_seconds": 0.03171390853822231,
|
| 76 |
+
"close_seconds": 0.00040434859693050385
|
| 77 |
},
|
| 78 |
+
"nar_seconds": 2.046467389911413,
|
| 79 |
+
"vae_seconds": 2.3237511478364468,
|
| 80 |
"load": {
|
| 81 |
+
"resolve_and_integrity_seconds": 0.19479286018759012,
|
| 82 |
+
"mot_load_seconds": 0.6600089315325022
|
| 83 |
},
|
| 84 |
"stages": {
|
| 85 |
+
"t0->plan_done": 3.2638021977618337,
|
| 86 |
+
"plan_done->semantic_done": 7.3499299036338925,
|
| 87 |
+
"semantic_done->nar_done": 2.046788734383881,
|
| 88 |
+
"nar_done->vae_done": 2.3241178588941693,
|
| 89 |
+
"plan_outside_timer": 0.0023637069389224052,
|
| 90 |
+
"semantic_outside_timer": 0.0012942077592015266,
|
| 91 |
+
"semantic_close_seconds": 0.00040434859693050385,
|
| 92 |
"vae": {
|
| 93 |
+
"load_s": 0.34615388605743647,
|
| 94 |
+
"decode_s": 1.9768154453486204,
|
| 95 |
+
"post_s": 1.2327916920185089e-05,
|
| 96 |
"tiles": [
|
| 97 |
+
1.7687,
|
| 98 |
+
0.0144,
|
| 99 |
+
0.0141,
|
| 100 |
+
0.0141,
|
| 101 |
+
0.014,
|
| 102 |
+
0.014,
|
| 103 |
+
0.014,
|
| 104 |
+
0.0139,
|
| 105 |
+
0.0138,
|
| 106 |
+
0.0138,
|
| 107 |
+
0.0138,
|
| 108 |
+
0.0137,
|
| 109 |
+
0.0138,
|
| 110 |
+
0.0404
|
| 111 |
],
|
| 112 |
"cudnn_benchmark": false,
|
| 113 |
"cudnn_deterministic": true,
|
| 114 |
+
"core_frames": 256,
|
| 115 |
"input_device": "cpu",
|
| 116 |
"input_contiguous": false
|
| 117 |
},
|
| 118 |
"nar_prefill_source": "handoff",
|
| 119 |
+
"abc_close_seconds": 0.0003131115809082985,
|
| 120 |
+
"memory": {
|
| 121 |
+
"t0": {
|
| 122 |
+
"peak_gib": 2.479,
|
| 123 |
+
"alloc_gib": 2.479,
|
| 124 |
+
"reserved_gib": 2.756,
|
| 125 |
+
"nvml_used_gib": 3.674
|
| 126 |
+
},
|
| 127 |
+
"plan_done": {
|
| 128 |
+
"peak_gib": 2.754,
|
| 129 |
+
"alloc_gib": 2.479,
|
| 130 |
+
"reserved_gib": 2.98,
|
| 131 |
+
"nvml_used_gib": 3.922
|
| 132 |
+
},
|
| 133 |
+
"semantic_done": {
|
| 134 |
+
"peak_gib": 3.17,
|
| 135 |
+
"alloc_gib": 3.067,
|
| 136 |
+
"reserved_gib": 3.309,
|
| 137 |
+
"nvml_used_gib": 4.252
|
| 138 |
+
},
|
| 139 |
+
"nar_done": {
|
| 140 |
+
"peak_gib": 3.12,
|
| 141 |
+
"alloc_gib": 2.488,
|
| 142 |
+
"reserved_gib": 3.4,
|
| 143 |
+
"nvml_used_gib": 4.36
|
| 144 |
+
},
|
| 145 |
+
"vae_done": {
|
| 146 |
+
"peak_gib": 3.216,
|
| 147 |
+
"alloc_gib": 2.735,
|
| 148 |
+
"reserved_gib": 3.518,
|
| 149 |
+
"nvml_used_gib": 4.487
|
| 150 |
+
}
|
| 151 |
+
}
|
| 152 |
},
|
| 153 |
+
"e2e_seconds": 14.984941340982914
|
| 154 |
},
|
| 155 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 156 |
"cfg": 1.0,
|
|
|
|
| 160 |
"kv_cache_dtype": "int8",
|
| 161 |
"captured_sampling": true,
|
| 162 |
"fuse_rms_quant": true,
|
| 163 |
+
"orbitquant_version": "0.9.8",
|
| 164 |
"profile": "fast",
|
| 165 |
"allocator_backend": "native",
|
| 166 |
"modules": 224,
|
|
|
|
| 183 |
},
|
| 184 |
{
|
| 185 |
"bundle": "v11",
|
| 186 |
+
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,vae_core=256,kv_handoff,lazy_handoff,int8_heads,w8a8",
|
| 187 |
"flags": [
|
| 188 |
"abc_crop",
|
| 189 |
"int8_heads",
|
| 190 |
"kv_handoff",
|
| 191 |
+
"lazy_handoff",
|
| 192 |
"lookahead",
|
|
|
|
| 193 |
"resident",
|
| 194 |
"vae_compile",
|
| 195 |
+
"vae_core=256",
|
| 196 |
"vae_mode=default",
|
| 197 |
"w8a8"
|
| 198 |
],
|
| 199 |
+
"native_sha256": "118d64db97c3200ef785e83b8d703ebc3b480a799bb17a6affb104d1b4faa28d",
|
| 200 |
"repeat": 1,
|
| 201 |
+
"wall_seconds": 12.26123733073473,
|
| 202 |
+
"load_seconds": 2.204909392632544,
|
| 203 |
"audio_seconds": 140.67866666666666,
|
| 204 |
+
"audio_per_wall": 11.473447815420172,
|
| 205 |
+
"abc_s": 2.8048877101391554,
|
| 206 |
"abc_tokens": 1249,
|
| 207 |
+
"sem_s": 7.337649843655527,
|
| 208 |
"sem_tokens": 3518,
|
| 209 |
+
"abc_ms_per_tok": 2.245706733498123,
|
| 210 |
+
"sem_ms_per_tok": 2.085744696889007,
|
| 211 |
+
"nar_s": 1.9093166394159198,
|
| 212 |
+
"vae_s": 0.20481676515191793,
|
| 213 |
+
"nvml_peak_gib": 4.635009765625,
|
| 214 |
+
"torch_peak_gib": 2.735057830810547,
|
| 215 |
+
"torch_reserved_peak_gib": 3.654296875,
|
| 216 |
"semantic_sha256": "bcb32247a0bd840b8b568746f93e93f823a87796b1f87fb55492da8eb741b473",
|
| 217 |
+
"waveform_sha256": "e0984b3364956b3d27d98703ee965d578951863b358ccff6769e544a15a77648",
|
| 218 |
+
"latent_sha256": "1763ac7941cbedffae6c43ca33e947ca9c8c390ae0d290db8b6c426e97439b36",
|
| 219 |
"timing": {
|
| 220 |
"abc": {
|
| 221 |
+
"seconds": 2.8048877101391554,
|
| 222 |
+
"prefill_seconds": 0.0390327638015151,
|
| 223 |
+
"ttft_seconds": 0.0390746770426631,
|
| 224 |
"output_tokens": 1249,
|
| 225 |
"content_tokens": 1248,
|
| 226 |
+
"output_tps": 445.29411836527134,
|
| 227 |
"prefix_tokens": 423,
|
| 228 |
"cfg_branches": 1,
|
| 229 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 234 |
"lookahead": true,
|
| 235 |
"unroll2": false,
|
| 236 |
"first_step_seconds": null,
|
| 237 |
+
"capture_seconds": 0.035048287361860275,
|
| 238 |
+
"close_seconds": 0.00037193018943071365
|
| 239 |
},
|
| 240 |
"semantic": {
|
| 241 |
+
"seconds": 7.337649843655527,
|
| 242 |
+
"prefill_seconds": 0.048011607490479946,
|
| 243 |
+
"ttft_seconds": 0.048030185513198376,
|
| 244 |
"output_tokens": 3518,
|
| 245 |
"content_tokens": 3517,
|
| 246 |
+
"output_tps": 479.4450641497736,
|
| 247 |
"prefix_tokens": 1673,
|
| 248 |
"cfg_branches": 1,
|
| 249 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 254 |
"lookahead": true,
|
| 255 |
"unroll2": false,
|
| 256 |
"first_step_seconds": null,
|
| 257 |
+
"capture_seconds": 0.030972903594374657,
|
| 258 |
+
"close_seconds": 0.0003538327291607857
|
| 259 |
},
|
| 260 |
+
"nar_seconds": 1.9093166394159198,
|
| 261 |
+
"vae_seconds": 0.20481676515191793,
|
| 262 |
"load": {
|
| 263 |
+
"resolve_and_integrity_seconds": 0.19479286018759012,
|
| 264 |
+
"mot_load_seconds": 0.6600089315325022
|
| 265 |
},
|
| 266 |
"stages": {
|
| 267 |
+
"t0->plan_done": 2.806506736204028,
|
| 268 |
+
"plan_done->semantic_done": 7.338900384493172,
|
| 269 |
+
"semantic_done->nar_done": 1.9096503229811788,
|
| 270 |
+
"nar_done->vae_done": 0.2051340825855732,
|
| 271 |
+
"plan_outside_timer": 0.0016190260648727417,
|
| 272 |
+
"semantic_outside_timer": 0.0012505408376455307,
|
| 273 |
+
"semantic_close_seconds": 0.0003538327291607857,
|
| 274 |
"vae": {
|
| 275 |
+
"load_s": 0.0008637299761176109,
|
| 276 |
+
"decode_s": 0.20389806292951107,
|
| 277 |
+
"post_s": 1.0485760867595673e-05,
|
| 278 |
"tiles": [
|
| 279 |
+
0.0141,
|
| 280 |
+
0.0137,
|
| 281 |
+
0.0136,
|
| 282 |
+
0.0137,
|
| 283 |
+
0.0136,
|
| 284 |
+
0.0136,
|
| 285 |
+
0.0136,
|
| 286 |
+
0.0136,
|
| 287 |
+
0.0137,
|
| 288 |
+
0.0136,
|
| 289 |
+
0.0135,
|
| 290 |
+
0.0135,
|
| 291 |
+
0.0135,
|
| 292 |
+
0.0264
|
| 293 |
],
|
| 294 |
"cudnn_benchmark": false,
|
| 295 |
"cudnn_deterministic": true,
|
| 296 |
+
"core_frames": 256,
|
| 297 |
"input_device": "cpu",
|
| 298 |
"input_contiguous": false
|
| 299 |
},
|
| 300 |
"nar_prefill_source": "handoff",
|
| 301 |
+
"abc_close_seconds": 0.00037193018943071365,
|
| 302 |
+
"memory": {
|
| 303 |
+
"t0": {
|
| 304 |
+
"peak_gib": 2.735,
|
| 305 |
+
"alloc_gib": 2.735,
|
| 306 |
+
"reserved_gib": 3.518,
|
| 307 |
+
"nvml_used_gib": 4.487
|
| 308 |
+
},
|
| 309 |
+
"plan_done": {
|
| 310 |
+
"peak_gib": 3.01,
|
| 311 |
+
"alloc_gib": 2.735,
|
| 312 |
+
"reserved_gib": 3.215,
|
| 313 |
+
"nvml_used_gib": 4.184
|
| 314 |
+
},
|
| 315 |
+
"semantic_done": {
|
| 316 |
+
"peak_gib": 3.426,
|
| 317 |
+
"alloc_gib": 3.323,
|
| 318 |
+
"reserved_gib": 3.543,
|
| 319 |
+
"nvml_used_gib": 4.512
|
| 320 |
+
},
|
| 321 |
+
"nar_done": {
|
| 322 |
+
"peak_gib": 3.368,
|
| 323 |
+
"alloc_gib": 2.735,
|
| 324 |
+
"reserved_gib": 3.654,
|
| 325 |
+
"nvml_used_gib": 4.623
|
| 326 |
+
},
|
| 327 |
+
"vae_done": {
|
| 328 |
+
"peak_gib": 3.216,
|
| 329 |
+
"alloc_gib": 2.735,
|
| 330 |
+
"reserved_gib": 3.654,
|
| 331 |
+
"nvml_used_gib": 4.623
|
| 332 |
+
}
|
| 333 |
+
}
|
| 334 |
},
|
| 335 |
+
"e2e_seconds": 12.260506701655686
|
| 336 |
},
|
| 337 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 338 |
"cfg": 1.0,
|
|
|
|
| 342 |
"kv_cache_dtype": "int8",
|
| 343 |
"captured_sampling": true,
|
| 344 |
"fuse_rms_quant": true,
|
| 345 |
+
"orbitquant_version": "0.9.8",
|
| 346 |
"profile": "fast",
|
| 347 |
"allocator_backend": "native",
|
| 348 |
"modules": 224,
|
evaluation/profiles/lowmem/run-00/config.json
CHANGED
|
@@ -34,12 +34,12 @@
|
|
| 34 |
"model_dtype": "bfloat16",
|
| 35 |
"vae_dtype": "float32",
|
| 36 |
"vae_decode": "halo_crop",
|
| 37 |
-
"vae_core_frames":
|
| 38 |
"vae_halo_frames": 16,
|
| 39 |
"device": "cuda:0",
|
| 40 |
"memory_budget_gib": 18.0,
|
| 41 |
"offload_ar": false,
|
| 42 |
-
"runtime_sha256": "
|
| 43 |
"decoder_release": "standard",
|
| 44 |
"validation_status": "unvalidated",
|
| 45 |
"runtime_profile": "fast",
|
|
@@ -47,7 +47,7 @@
|
|
| 47 |
"captured_sampling": true,
|
| 48 |
"kv_cache_dtype": "int8",
|
| 49 |
"int8_attention": "ordered",
|
| 50 |
-
"runtime_bundle_sha256": "
|
| 51 |
"vae_compute_dtype": "torch.float16",
|
| 52 |
"expert_staging": false,
|
| 53 |
"orbitquant": {
|
|
|
|
| 34 |
"model_dtype": "bfloat16",
|
| 35 |
"vae_dtype": "float32",
|
| 36 |
"vae_decode": "halo_crop",
|
| 37 |
+
"vae_core_frames": 256,
|
| 38 |
"vae_halo_frames": 16,
|
| 39 |
"device": "cuda:0",
|
| 40 |
"memory_budget_gib": 18.0,
|
| 41 |
"offload_ar": false,
|
| 42 |
+
"runtime_sha256": "bb189091c79b42dd3eca4f24f6239294df3c517fb2b149fc3b30ada3be287119",
|
| 43 |
"decoder_release": "standard",
|
| 44 |
"validation_status": "unvalidated",
|
| 45 |
"runtime_profile": "fast",
|
|
|
|
| 47 |
"captured_sampling": true,
|
| 48 |
"kv_cache_dtype": "int8",
|
| 49 |
"int8_attention": "ordered",
|
| 50 |
+
"runtime_bundle_sha256": "269214c6459739a0f08889532d5fc76f3e6f48f3908807466e14e5da4ecb1c98",
|
| 51 |
"vae_compute_dtype": "torch.float16",
|
| 52 |
"expert_staging": false,
|
| 53 |
"orbitquant": {
|
evaluation/profiles/lowmem/run-00/metrics.json
CHANGED
|
@@ -1,46 +1,47 @@
|
|
| 1 |
{
|
| 2 |
"bundle": "v11",
|
| 3 |
-
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,kv_handoff,
|
| 4 |
"flags": [
|
| 5 |
"abc_crop",
|
| 6 |
"int8_embed",
|
| 7 |
"int8_heads",
|
| 8 |
"kv_handoff",
|
|
|
|
| 9 |
"lookahead",
|
| 10 |
-
"nar_compile",
|
| 11 |
"resident",
|
| 12 |
"vae_compile",
|
|
|
|
| 13 |
"vae_mode=default",
|
| 14 |
"w8a8"
|
| 15 |
],
|
| 16 |
-
"native_sha256": "
|
| 17 |
"repeat": 0,
|
| 18 |
-
"wall_seconds":
|
| 19 |
-
"load_seconds":
|
| 20 |
"audio_seconds": 152.15866666666668,
|
| 21 |
-
"audio_per_wall":
|
| 22 |
-
"abc_s":
|
| 23 |
"abc_tokens": 1400,
|
| 24 |
-
"sem_s": 8.
|
| 25 |
"sem_tokens": 3805,
|
| 26 |
-
"abc_ms_per_tok":
|
| 27 |
-
"sem_ms_per_tok": 2.
|
| 28 |
-
"nar_s":
|
| 29 |
-
"vae_s":
|
| 30 |
-
"nvml_peak_gib": 4.
|
| 31 |
-
"torch_peak_gib":
|
| 32 |
-
"torch_reserved_peak_gib": 3.
|
| 33 |
"semantic_sha256": "2e35b32e83f3613db2825d4879ffb1a09798f37b8f85b20ce42f7c7624bae895",
|
| 34 |
-
"waveform_sha256": "
|
| 35 |
-
"latent_sha256": "
|
| 36 |
"timing": {
|
| 37 |
"abc": {
|
| 38 |
-
"seconds":
|
| 39 |
-
"prefill_seconds":
|
| 40 |
-
"ttft_seconds":
|
| 41 |
"output_tokens": 1400,
|
| 42 |
"content_tokens": 1399,
|
| 43 |
-
"output_tps":
|
| 44 |
"prefix_tokens": 423,
|
| 45 |
"cfg_branches": 1,
|
| 46 |
"sampling": "combined_cuda_graph",
|
|
@@ -51,16 +52,16 @@
|
|
| 51 |
"lookahead": true,
|
| 52 |
"unroll2": false,
|
| 53 |
"first_step_seconds": null,
|
| 54 |
-
"capture_seconds": 0.
|
| 55 |
-
"close_seconds": 0.
|
| 56 |
},
|
| 57 |
"semantic": {
|
| 58 |
-
"seconds": 8.
|
| 59 |
-
"prefill_seconds": 0.
|
| 60 |
-
"ttft_seconds": 0.
|
| 61 |
"output_tokens": 3805,
|
| 62 |
"content_tokens": 3804,
|
| 63 |
-
"output_tps":
|
| 64 |
"prefix_tokens": 1824,
|
| 65 |
"cfg_branches": 1,
|
| 66 |
"sampling": "combined_cuda_graph",
|
|
@@ -71,47 +72,86 @@
|
|
| 71 |
"lookahead": true,
|
| 72 |
"unroll2": false,
|
| 73 |
"first_step_seconds": null,
|
| 74 |
-
"capture_seconds": 0.
|
| 75 |
-
"close_seconds": 0.
|
| 76 |
},
|
| 77 |
-
"nar_seconds":
|
| 78 |
-
"vae_seconds":
|
| 79 |
"load": {
|
| 80 |
-
"resolve_and_integrity_seconds":
|
| 81 |
-
"mot_load_seconds":
|
| 82 |
},
|
| 83 |
"stages": {
|
| 84 |
-
"t0->plan_done":
|
| 85 |
-
"plan_done->semantic_done": 8.
|
| 86 |
-
"semantic_done->nar_done":
|
| 87 |
-
"nar_done->vae_done":
|
| 88 |
-
"plan_outside_timer": 0.
|
| 89 |
-
"semantic_outside_timer": 0.
|
| 90 |
-
"semantic_close_seconds": 0.
|
| 91 |
"vae": {
|
| 92 |
-
"load_s":
|
| 93 |
-
"decode_s":
|
| 94 |
-
"post_s":
|
| 95 |
"tiles": [
|
| 96 |
-
|
| 97 |
-
0.
|
| 98 |
-
0.
|
| 99 |
-
0.
|
| 100 |
-
0.
|
| 101 |
-
0.
|
| 102 |
-
0.
|
| 103 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 104 |
],
|
| 105 |
"cudnn_benchmark": false,
|
| 106 |
"cudnn_deterministic": true,
|
| 107 |
-
"core_frames":
|
| 108 |
"input_device": "cpu",
|
| 109 |
"input_contiguous": false
|
| 110 |
},
|
| 111 |
"nar_prefill_source": "handoff",
|
| 112 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 113 |
},
|
| 114 |
-
"e2e_seconds":
|
| 115 |
},
|
| 116 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 117 |
"cfg": 1.0,
|
|
@@ -121,7 +161,7 @@
|
|
| 121 |
"kv_cache_dtype": "int8",
|
| 122 |
"captured_sampling": true,
|
| 123 |
"fuse_rms_quant": true,
|
| 124 |
-
"orbitquant_version": "0.9.
|
| 125 |
"profile": "fast",
|
| 126 |
"allocator_backend": "native",
|
| 127 |
"modules": 224,
|
|
|
|
| 1 |
{
|
| 2 |
"bundle": "v11",
|
| 3 |
+
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,vae_core=256,kv_handoff,lazy_handoff,int8_heads,w8a8,int8_embed",
|
| 4 |
"flags": [
|
| 5 |
"abc_crop",
|
| 6 |
"int8_embed",
|
| 7 |
"int8_heads",
|
| 8 |
"kv_handoff",
|
| 9 |
+
"lazy_handoff",
|
| 10 |
"lookahead",
|
|
|
|
| 11 |
"resident",
|
| 12 |
"vae_compile",
|
| 13 |
+
"vae_core=256",
|
| 14 |
"vae_mode=default",
|
| 15 |
"w8a8"
|
| 16 |
],
|
| 17 |
+
"native_sha256": "118d64db97c3200ef785e83b8d703ebc3b480a799bb17a6affb104d1b4faa28d",
|
| 18 |
"repeat": 0,
|
| 19 |
+
"wall_seconds": 16.067145991139114,
|
| 20 |
+
"load_seconds": 2.229018185287714,
|
| 21 |
"audio_seconds": 152.15866666666668,
|
| 22 |
+
"audio_per_wall": 9.470173903354137,
|
| 23 |
+
"abc_s": 3.7763062892481685,
|
| 24 |
"abc_tokens": 1400,
|
| 25 |
+
"sem_s": 8.044851490296423,
|
| 26 |
"sem_tokens": 3805,
|
| 27 |
+
"abc_ms_per_tok": 2.697361635177263,
|
| 28 |
+
"sem_ms_per_tok": 2.1142842287244217,
|
| 29 |
+
"nar_s": 2.187400832772255,
|
| 30 |
+
"vae_s": 2.0492115300148726,
|
| 31 |
+
"nvml_peak_gib": 4.213134765625,
|
| 32 |
+
"torch_peak_gib": 2.3834547996520996,
|
| 33 |
+
"torch_reserved_peak_gib": 3.244140625,
|
| 34 |
"semantic_sha256": "2e35b32e83f3613db2825d4879ffb1a09798f37b8f85b20ce42f7c7624bae895",
|
| 35 |
+
"waveform_sha256": "d1ce52cdce9991784376a9bc1334abd85a3eb8b0870a914ecfdab0dabd87fad5",
|
| 36 |
+
"latent_sha256": "877e8cc0fa28b4515b39b8b3f642e1c9f4db1524825b47eb0e6991772fe1ade6",
|
| 37 |
"timing": {
|
| 38 |
"abc": {
|
| 39 |
+
"seconds": 3.7763062892481685,
|
| 40 |
+
"prefill_seconds": 0.6570839313790202,
|
| 41 |
+
"ttft_seconds": 0.6582063101232052,
|
| 42 |
"output_tokens": 1400,
|
| 43 |
"content_tokens": 1399,
|
| 44 |
+
"output_tps": 370.73263998369384,
|
| 45 |
"prefix_tokens": 423,
|
| 46 |
"cfg_branches": 1,
|
| 47 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 52 |
"lookahead": true,
|
| 53 |
"unroll2": false,
|
| 54 |
"first_step_seconds": null,
|
| 55 |
+
"capture_seconds": 0.04039210919290781,
|
| 56 |
+
"close_seconds": 0.0003549950197339058
|
| 57 |
},
|
| 58 |
"semantic": {
|
| 59 |
+
"seconds": 8.044851490296423,
|
| 60 |
+
"prefill_seconds": 0.05773051269352436,
|
| 61 |
+
"ttft_seconds": 0.05774910096079111,
|
| 62 |
"output_tokens": 3805,
|
| 63 |
"content_tokens": 3804,
|
| 64 |
+
"output_tps": 472.97330529836785,
|
| 65 |
"prefix_tokens": 1824,
|
| 66 |
"cfg_branches": 1,
|
| 67 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 72 |
"lookahead": true,
|
| 73 |
"unroll2": false,
|
| 74 |
"first_step_seconds": null,
|
| 75 |
+
"capture_seconds": 0.032243385910987854,
|
| 76 |
+
"close_seconds": 0.0004120608791708946
|
| 77 |
},
|
| 78 |
+
"nar_seconds": 2.187400832772255,
|
| 79 |
+
"vae_seconds": 2.0492115300148726,
|
| 80 |
"load": {
|
| 81 |
+
"resolve_and_integrity_seconds": 0.20540913753211498,
|
| 82 |
+
"mot_load_seconds": 0.6672966200858355
|
| 83 |
},
|
| 84 |
"stages": {
|
| 85 |
+
"t0->plan_done": 3.7815310433506966,
|
| 86 |
+
"plan_done->semantic_done": 8.046199827454984,
|
| 87 |
+
"semantic_done->nar_done": 2.1876858919858932,
|
| 88 |
+
"nar_done->vae_done": 2.049526182934642,
|
| 89 |
+
"plan_outside_timer": 0.005224754102528095,
|
| 90 |
+
"semantic_outside_timer": 0.0013483371585607529,
|
| 91 |
+
"semantic_close_seconds": 0.0004120608791708946,
|
| 92 |
"vae": {
|
| 93 |
+
"load_s": 0.026218397542834282,
|
| 94 |
+
"decode_s": 2.0229244586080313,
|
| 95 |
+
"post_s": 1.2959353625774384e-05,
|
| 96 |
"tiles": [
|
| 97 |
+
1.7943,
|
| 98 |
+
0.0139,
|
| 99 |
+
0.014,
|
| 100 |
+
0.0142,
|
| 101 |
+
0.0141,
|
| 102 |
+
0.0141,
|
| 103 |
+
0.0139,
|
| 104 |
+
0.014,
|
| 105 |
+
0.0141,
|
| 106 |
+
0.0144,
|
| 107 |
+
0.0141,
|
| 108 |
+
0.0142,
|
| 109 |
+
0.014,
|
| 110 |
+
0.0142,
|
| 111 |
+
0.0452
|
| 112 |
],
|
| 113 |
"cudnn_benchmark": false,
|
| 114 |
"cudnn_deterministic": true,
|
| 115 |
+
"core_frames": 256,
|
| 116 |
"input_device": "cpu",
|
| 117 |
"input_contiguous": false
|
| 118 |
},
|
| 119 |
"nar_prefill_source": "handoff",
|
| 120 |
+
"abc_close_seconds": 0.0003549950197339058,
|
| 121 |
+
"memory": {
|
| 122 |
+
"t0": {
|
| 123 |
+
"peak_gib": 2.127,
|
| 124 |
+
"alloc_gib": 2.127,
|
| 125 |
+
"reserved_gib": 2.424,
|
| 126 |
+
"nvml_used_gib": 3.342
|
| 127 |
+
},
|
| 128 |
+
"plan_done": {
|
| 129 |
+
"peak_gib": 2.402,
|
| 130 |
+
"alloc_gib": 2.127,
|
| 131 |
+
"reserved_gib": 2.629,
|
| 132 |
+
"nvml_used_gib": 3.573
|
| 133 |
+
},
|
| 134 |
+
"semantic_done": {
|
| 135 |
+
"peak_gib": 2.836,
|
| 136 |
+
"alloc_gib": 2.723,
|
| 137 |
+
"reserved_gib": 2.977,
|
| 138 |
+
"nvml_used_gib": 3.92
|
| 139 |
+
},
|
| 140 |
+
"nar_done": {
|
| 141 |
+
"peak_gib": 2.82,
|
| 142 |
+
"alloc_gib": 2.136,
|
| 143 |
+
"reserved_gib": 3.127,
|
| 144 |
+
"nvml_used_gib": 4.086
|
| 145 |
+
},
|
| 146 |
+
"vae_done": {
|
| 147 |
+
"peak_gib": 2.918,
|
| 148 |
+
"alloc_gib": 2.383,
|
| 149 |
+
"reserved_gib": 3.244,
|
| 150 |
+
"nvml_used_gib": 4.213
|
| 151 |
+
}
|
| 152 |
+
}
|
| 153 |
},
|
| 154 |
+
"e2e_seconds": 16.065859724767506
|
| 155 |
},
|
| 156 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 157 |
"cfg": 1.0,
|
|
|
|
| 161 |
"kv_cache_dtype": "int8",
|
| 162 |
"captured_sampling": true,
|
| 163 |
"fuse_rms_quant": true,
|
| 164 |
+
"orbitquant_version": "0.9.8",
|
| 165 |
"profile": "fast",
|
| 166 |
"allocator_backend": "native",
|
| 167 |
"modules": 224,
|
evaluation/profiles/lowmem/run-00/plan.json
CHANGED
|
@@ -9,12 +9,12 @@
|
|
| 9 |
"id": "song"
|
| 10 |
},
|
| 11 |
"timing": {
|
| 12 |
-
"seconds":
|
| 13 |
-
"prefill_seconds":
|
| 14 |
-
"ttft_seconds":
|
| 15 |
"output_tokens": 1400,
|
| 16 |
"content_tokens": 1399,
|
| 17 |
-
"output_tps":
|
| 18 |
"prefix_tokens": 423,
|
| 19 |
"cfg_branches": 1,
|
| 20 |
"sampling": "combined_cuda_graph",
|
|
@@ -25,8 +25,8 @@
|
|
| 25 |
"lookahead": true,
|
| 26 |
"unroll2": false,
|
| 27 |
"first_step_seconds": null,
|
| 28 |
-
"capture_seconds": 0.
|
| 29 |
-
"close_seconds": 0.
|
| 30 |
},
|
| 31 |
"truncated": false,
|
| 32 |
"prefix": [
|
|
|
|
| 9 |
"id": "song"
|
| 10 |
},
|
| 11 |
"timing": {
|
| 12 |
+
"seconds": 3.7763062892481685,
|
| 13 |
+
"prefill_seconds": 0.6570839313790202,
|
| 14 |
+
"ttft_seconds": 0.6582063101232052,
|
| 15 |
"output_tokens": 1400,
|
| 16 |
"content_tokens": 1399,
|
| 17 |
+
"output_tps": 370.73263998369384,
|
| 18 |
"prefix_tokens": 423,
|
| 19 |
"cfg_branches": 1,
|
| 20 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 25 |
"lookahead": true,
|
| 26 |
"unroll2": false,
|
| 27 |
"first_step_seconds": null,
|
| 28 |
+
"capture_seconds": 0.04039210919290781,
|
| 29 |
+
"close_seconds": 0.0003549950197339058
|
| 30 |
},
|
| 31 |
"truncated": false,
|
| 32 |
"prefix": [
|
evaluation/profiles/lowmem/run-00/plan_manifest.json
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
{
|
| 2 |
-
"plan.json": "
|
| 3 |
"abc_tokens.npy": "7fa8ba96997683e2a3d126ea903cd325b32a6d1aeb835c4a57e46ccd334d647e",
|
| 4 |
"prefix.npy": "26b655d943e4b860b3da460c7551d630ed51dd5de8c79ec511eaaa99c680fef4",
|
| 5 |
"score.abc": "ff4ea2b1fab4dc0e10a8c3347b8a9106151d38845493e5a3e11b847a53606e13"
|
|
|
|
| 1 |
{
|
| 2 |
+
"plan.json": "b8b2dfaa50594f8fd427345295a13e1c2cd8acef36533d8b61c66f875480328d",
|
| 3 |
"abc_tokens.npy": "7fa8ba96997683e2a3d126ea903cd325b32a6d1aeb835c4a57e46ccd334d647e",
|
| 4 |
"prefix.npy": "26b655d943e4b860b3da460c7551d630ed51dd5de8c79ec511eaaa99c680fef4",
|
| 5 |
"score.abc": "ff4ea2b1fab4dc0e10a8c3347b8a9106151d38845493e5a3e11b847a53606e13"
|
evaluation/profiles/lowmem/run-00/result.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"status": "complete",
|
| 3 |
-
"identity": "
|
| 4 |
"truncated": {
|
| 5 |
"abc": false,
|
| 6 |
"semantic": false
|
|
@@ -29,12 +29,12 @@
|
|
| 29 |
},
|
| 30 |
"timing": {
|
| 31 |
"abc": {
|
| 32 |
-
"seconds":
|
| 33 |
-
"prefill_seconds":
|
| 34 |
-
"ttft_seconds":
|
| 35 |
"output_tokens": 1400,
|
| 36 |
"content_tokens": 1399,
|
| 37 |
-
"output_tps":
|
| 38 |
"prefix_tokens": 423,
|
| 39 |
"cfg_branches": 1,
|
| 40 |
"sampling": "combined_cuda_graph",
|
|
@@ -45,16 +45,16 @@
|
|
| 45 |
"lookahead": true,
|
| 46 |
"unroll2": false,
|
| 47 |
"first_step_seconds": null,
|
| 48 |
-
"capture_seconds": 0.
|
| 49 |
-
"close_seconds": 0.
|
| 50 |
},
|
| 51 |
"semantic": {
|
| 52 |
-
"seconds": 8.
|
| 53 |
-
"prefill_seconds": 0.
|
| 54 |
-
"ttft_seconds": 0.
|
| 55 |
"output_tokens": 3805,
|
| 56 |
"content_tokens": 3804,
|
| 57 |
-
"output_tps":
|
| 58 |
"prefix_tokens": 1824,
|
| 59 |
"cfg_branches": 1,
|
| 60 |
"sampling": "combined_cuda_graph",
|
|
@@ -65,47 +65,86 @@
|
|
| 65 |
"lookahead": true,
|
| 66 |
"unroll2": false,
|
| 67 |
"first_step_seconds": null,
|
| 68 |
-
"capture_seconds": 0.
|
| 69 |
-
"close_seconds": 0.
|
| 70 |
},
|
| 71 |
-
"nar_seconds":
|
| 72 |
-
"vae_seconds":
|
| 73 |
"load": {
|
| 74 |
-
"resolve_and_integrity_seconds":
|
| 75 |
-
"mot_load_seconds":
|
| 76 |
},
|
| 77 |
"stages": {
|
| 78 |
-
"t0->plan_done":
|
| 79 |
-
"plan_done->semantic_done": 8.
|
| 80 |
-
"semantic_done->nar_done":
|
| 81 |
-
"nar_done->vae_done":
|
| 82 |
-
"plan_outside_timer": 0.
|
| 83 |
-
"semantic_outside_timer": 0.
|
| 84 |
-
"semantic_close_seconds": 0.
|
| 85 |
"vae": {
|
| 86 |
-
"load_s":
|
| 87 |
-
"decode_s":
|
| 88 |
-
"post_s":
|
| 89 |
"tiles": [
|
| 90 |
-
|
| 91 |
-
0.
|
| 92 |
-
0.
|
| 93 |
-
0.
|
| 94 |
-
0.
|
| 95 |
-
0.
|
| 96 |
-
0.
|
| 97 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 98 |
],
|
| 99 |
"cudnn_benchmark": false,
|
| 100 |
"cudnn_deterministic": true,
|
| 101 |
-
"core_frames":
|
| 102 |
"input_device": "cpu",
|
| 103 |
"input_contiguous": false
|
| 104 |
},
|
| 105 |
"nar_prefill_source": "handoff",
|
| 106 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 107 |
},
|
| 108 |
-
"e2e_seconds":
|
| 109 |
},
|
| 110 |
"artifacts": {
|
| 111 |
"abc_tokens.npy": {
|
|
@@ -113,23 +152,23 @@
|
|
| 113 |
"bytes": 5724
|
| 114 |
},
|
| 115 |
"audio.flac": {
|
| 116 |
-
"sha256": "
|
| 117 |
-
"bytes":
|
| 118 |
},
|
| 119 |
"config.json": {
|
| 120 |
-
"sha256": "
|
| 121 |
"bytes": 2476
|
| 122 |
},
|
| 123 |
"latent.npy": {
|
| 124 |
-
"sha256": "
|
| 125 |
"bytes": 973952
|
| 126 |
},
|
| 127 |
"plan.json": {
|
| 128 |
-
"sha256": "
|
| 129 |
-
"bytes":
|
| 130 |
},
|
| 131 |
"plan_manifest.json": {
|
| 132 |
-
"sha256": "
|
| 133 |
"bytes": 341
|
| 134 |
},
|
| 135 |
"prefix.npy": {
|
|
|
|
| 1 |
{
|
| 2 |
"status": "complete",
|
| 3 |
+
"identity": "5b78b1460d137eb9aa61003dc8c71bf596658015622424138dc69547e7213123",
|
| 4 |
"truncated": {
|
| 5 |
"abc": false,
|
| 6 |
"semantic": false
|
|
|
|
| 29 |
},
|
| 30 |
"timing": {
|
| 31 |
"abc": {
|
| 32 |
+
"seconds": 3.7763062892481685,
|
| 33 |
+
"prefill_seconds": 0.6570839313790202,
|
| 34 |
+
"ttft_seconds": 0.6582063101232052,
|
| 35 |
"output_tokens": 1400,
|
| 36 |
"content_tokens": 1399,
|
| 37 |
+
"output_tps": 370.73263998369384,
|
| 38 |
"prefix_tokens": 423,
|
| 39 |
"cfg_branches": 1,
|
| 40 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 45 |
"lookahead": true,
|
| 46 |
"unroll2": false,
|
| 47 |
"first_step_seconds": null,
|
| 48 |
+
"capture_seconds": 0.04039210919290781,
|
| 49 |
+
"close_seconds": 0.0003549950197339058
|
| 50 |
},
|
| 51 |
"semantic": {
|
| 52 |
+
"seconds": 8.044851490296423,
|
| 53 |
+
"prefill_seconds": 0.05773051269352436,
|
| 54 |
+
"ttft_seconds": 0.05774910096079111,
|
| 55 |
"output_tokens": 3805,
|
| 56 |
"content_tokens": 3804,
|
| 57 |
+
"output_tps": 472.97330529836785,
|
| 58 |
"prefix_tokens": 1824,
|
| 59 |
"cfg_branches": 1,
|
| 60 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 65 |
"lookahead": true,
|
| 66 |
"unroll2": false,
|
| 67 |
"first_step_seconds": null,
|
| 68 |
+
"capture_seconds": 0.032243385910987854,
|
| 69 |
+
"close_seconds": 0.0004120608791708946
|
| 70 |
},
|
| 71 |
+
"nar_seconds": 2.187400832772255,
|
| 72 |
+
"vae_seconds": 2.0492115300148726,
|
| 73 |
"load": {
|
| 74 |
+
"resolve_and_integrity_seconds": 0.20540913753211498,
|
| 75 |
+
"mot_load_seconds": 0.6672966200858355
|
| 76 |
},
|
| 77 |
"stages": {
|
| 78 |
+
"t0->plan_done": 3.7815310433506966,
|
| 79 |
+
"plan_done->semantic_done": 8.046199827454984,
|
| 80 |
+
"semantic_done->nar_done": 2.1876858919858932,
|
| 81 |
+
"nar_done->vae_done": 2.049526182934642,
|
| 82 |
+
"plan_outside_timer": 0.005224754102528095,
|
| 83 |
+
"semantic_outside_timer": 0.0013483371585607529,
|
| 84 |
+
"semantic_close_seconds": 0.0004120608791708946,
|
| 85 |
"vae": {
|
| 86 |
+
"load_s": 0.026218397542834282,
|
| 87 |
+
"decode_s": 2.0229244586080313,
|
| 88 |
+
"post_s": 1.2959353625774384e-05,
|
| 89 |
"tiles": [
|
| 90 |
+
1.7943,
|
| 91 |
+
0.0139,
|
| 92 |
+
0.014,
|
| 93 |
+
0.0142,
|
| 94 |
+
0.0141,
|
| 95 |
+
0.0141,
|
| 96 |
+
0.0139,
|
| 97 |
+
0.014,
|
| 98 |
+
0.0141,
|
| 99 |
+
0.0144,
|
| 100 |
+
0.0141,
|
| 101 |
+
0.0142,
|
| 102 |
+
0.014,
|
| 103 |
+
0.0142,
|
| 104 |
+
0.0452
|
| 105 |
],
|
| 106 |
"cudnn_benchmark": false,
|
| 107 |
"cudnn_deterministic": true,
|
| 108 |
+
"core_frames": 256,
|
| 109 |
"input_device": "cpu",
|
| 110 |
"input_contiguous": false
|
| 111 |
},
|
| 112 |
"nar_prefill_source": "handoff",
|
| 113 |
+
"abc_close_seconds": 0.0003549950197339058,
|
| 114 |
+
"memory": {
|
| 115 |
+
"t0": {
|
| 116 |
+
"peak_gib": 2.127,
|
| 117 |
+
"alloc_gib": 2.127,
|
| 118 |
+
"reserved_gib": 2.424,
|
| 119 |
+
"nvml_used_gib": 3.342
|
| 120 |
+
},
|
| 121 |
+
"plan_done": {
|
| 122 |
+
"peak_gib": 2.402,
|
| 123 |
+
"alloc_gib": 2.127,
|
| 124 |
+
"reserved_gib": 2.629,
|
| 125 |
+
"nvml_used_gib": 3.573
|
| 126 |
+
},
|
| 127 |
+
"semantic_done": {
|
| 128 |
+
"peak_gib": 2.836,
|
| 129 |
+
"alloc_gib": 2.723,
|
| 130 |
+
"reserved_gib": 2.977,
|
| 131 |
+
"nvml_used_gib": 3.92
|
| 132 |
+
},
|
| 133 |
+
"nar_done": {
|
| 134 |
+
"peak_gib": 2.82,
|
| 135 |
+
"alloc_gib": 2.136,
|
| 136 |
+
"reserved_gib": 3.127,
|
| 137 |
+
"nvml_used_gib": 4.086
|
| 138 |
+
},
|
| 139 |
+
"vae_done": {
|
| 140 |
+
"peak_gib": 2.918,
|
| 141 |
+
"alloc_gib": 2.383,
|
| 142 |
+
"reserved_gib": 3.244,
|
| 143 |
+
"nvml_used_gib": 4.213
|
| 144 |
+
}
|
| 145 |
+
}
|
| 146 |
},
|
| 147 |
+
"e2e_seconds": 16.065859724767506
|
| 148 |
},
|
| 149 |
"artifacts": {
|
| 150 |
"abc_tokens.npy": {
|
|
|
|
| 152 |
"bytes": 5724
|
| 153 |
},
|
| 154 |
"audio.flac": {
|
| 155 |
+
"sha256": "3ff922c8d2c5d7959e572cadc3bcd94d04e0c53212c429938884c37ab9e11801",
|
| 156 |
+
"bytes": 30317169
|
| 157 |
},
|
| 158 |
"config.json": {
|
| 159 |
+
"sha256": "1ca79e079b519059f4656324481ca68ddd6138f12dd333ba544811f339cb9c40",
|
| 160 |
"bytes": 2476
|
| 161 |
},
|
| 162 |
"latent.npy": {
|
| 163 |
+
"sha256": "bbf7fbbfcbdbf5a4c758f297f190cd567bf3edfa5d15929da8ccd43838a02a2c",
|
| 164 |
"bytes": 973952
|
| 165 |
},
|
| 166 |
"plan.json": {
|
| 167 |
+
"sha256": "b8b2dfaa50594f8fd427345295a13e1c2cd8acef36533d8b61c66f875480328d",
|
| 168 |
+
"bytes": 33014
|
| 169 |
},
|
| 170 |
"plan_manifest.json": {
|
| 171 |
+
"sha256": "e8a900437b475ba87e9f5443420b49d3c586511b1062367efb426340be60e381",
|
| 172 |
"bytes": 341
|
| 173 |
},
|
| 174 |
"prefix.npy": {
|
evaluation/profiles/lowmem/run-01/config.json
CHANGED
|
@@ -34,12 +34,12 @@
|
|
| 34 |
"model_dtype": "bfloat16",
|
| 35 |
"vae_dtype": "float32",
|
| 36 |
"vae_decode": "halo_crop",
|
| 37 |
-
"vae_core_frames":
|
| 38 |
"vae_halo_frames": 16,
|
| 39 |
"device": "cuda:0",
|
| 40 |
"memory_budget_gib": 18.0,
|
| 41 |
"offload_ar": false,
|
| 42 |
-
"runtime_sha256": "
|
| 43 |
"decoder_release": "standard",
|
| 44 |
"validation_status": "unvalidated",
|
| 45 |
"runtime_profile": "fast",
|
|
@@ -47,7 +47,7 @@
|
|
| 47 |
"captured_sampling": true,
|
| 48 |
"kv_cache_dtype": "int8",
|
| 49 |
"int8_attention": "ordered",
|
| 50 |
-
"runtime_bundle_sha256": "
|
| 51 |
"vae_compute_dtype": "torch.float16",
|
| 52 |
"expert_staging": false,
|
| 53 |
"orbitquant": {
|
|
|
|
| 34 |
"model_dtype": "bfloat16",
|
| 35 |
"vae_dtype": "float32",
|
| 36 |
"vae_decode": "halo_crop",
|
| 37 |
+
"vae_core_frames": 256,
|
| 38 |
"vae_halo_frames": 16,
|
| 39 |
"device": "cuda:0",
|
| 40 |
"memory_budget_gib": 18.0,
|
| 41 |
"offload_ar": false,
|
| 42 |
+
"runtime_sha256": "bb189091c79b42dd3eca4f24f6239294df3c517fb2b149fc3b30ada3be287119",
|
| 43 |
"decoder_release": "standard",
|
| 44 |
"validation_status": "unvalidated",
|
| 45 |
"runtime_profile": "fast",
|
|
|
|
| 47 |
"captured_sampling": true,
|
| 48 |
"kv_cache_dtype": "int8",
|
| 49 |
"int8_attention": "ordered",
|
| 50 |
+
"runtime_bundle_sha256": "269214c6459739a0f08889532d5fc76f3e6f48f3908807466e14e5da4ecb1c98",
|
| 51 |
"vae_compute_dtype": "torch.float16",
|
| 52 |
"expert_staging": false,
|
| 53 |
"orbitquant": {
|
evaluation/profiles/lowmem/run-01/metrics.json
CHANGED
|
@@ -1,46 +1,47 @@
|
|
| 1 |
{
|
| 2 |
"bundle": "v11",
|
| 3 |
-
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,kv_handoff,
|
| 4 |
"flags": [
|
| 5 |
"abc_crop",
|
| 6 |
"int8_embed",
|
| 7 |
"int8_heads",
|
| 8 |
"kv_handoff",
|
|
|
|
| 9 |
"lookahead",
|
| 10 |
-
"nar_compile",
|
| 11 |
"resident",
|
| 12 |
"vae_compile",
|
|
|
|
| 13 |
"vae_mode=default",
|
| 14 |
"w8a8"
|
| 15 |
],
|
| 16 |
-
"native_sha256": "
|
| 17 |
"repeat": 1,
|
| 18 |
-
"wall_seconds":
|
| 19 |
-
"load_seconds":
|
| 20 |
"audio_seconds": 152.15866666666668,
|
| 21 |
-
"audio_per_wall":
|
| 22 |
-
"abc_s": 3.
|
| 23 |
"abc_tokens": 1400,
|
| 24 |
-
"sem_s": 8.
|
| 25 |
"sem_tokens": 3805,
|
| 26 |
-
"abc_ms_per_tok": 2.
|
| 27 |
-
"sem_ms_per_tok": 2.
|
| 28 |
-
"nar_s":
|
| 29 |
-
"vae_s": 0.
|
| 30 |
-
"nvml_peak_gib": 4.
|
| 31 |
-
"torch_peak_gib":
|
| 32 |
-
"torch_reserved_peak_gib": 3.
|
| 33 |
"semantic_sha256": "2e35b32e83f3613db2825d4879ffb1a09798f37b8f85b20ce42f7c7624bae895",
|
| 34 |
-
"waveform_sha256": "
|
| 35 |
-
"latent_sha256": "
|
| 36 |
"timing": {
|
| 37 |
"abc": {
|
| 38 |
-
"seconds": 3.
|
| 39 |
-
"prefill_seconds": 0.
|
| 40 |
-
"ttft_seconds": 0.
|
| 41 |
"output_tokens": 1400,
|
| 42 |
"content_tokens": 1399,
|
| 43 |
-
"output_tps":
|
| 44 |
"prefix_tokens": 423,
|
| 45 |
"cfg_branches": 1,
|
| 46 |
"sampling": "combined_cuda_graph",
|
|
@@ -51,16 +52,16 @@
|
|
| 51 |
"lookahead": true,
|
| 52 |
"unroll2": false,
|
| 53 |
"first_step_seconds": null,
|
| 54 |
-
"capture_seconds": 0.
|
| 55 |
-
"close_seconds": 0.
|
| 56 |
},
|
| 57 |
"semantic": {
|
| 58 |
-
"seconds": 8.
|
| 59 |
-
"prefill_seconds": 0.
|
| 60 |
-
"ttft_seconds": 0.
|
| 61 |
"output_tokens": 3805,
|
| 62 |
"content_tokens": 3804,
|
| 63 |
-
"output_tps":
|
| 64 |
"prefix_tokens": 1824,
|
| 65 |
"cfg_branches": 1,
|
| 66 |
"sampling": "combined_cuda_graph",
|
|
@@ -71,47 +72,86 @@
|
|
| 71 |
"lookahead": true,
|
| 72 |
"unroll2": false,
|
| 73 |
"first_step_seconds": null,
|
| 74 |
-
"capture_seconds": 0.
|
| 75 |
-
"close_seconds": 0.
|
| 76 |
},
|
| 77 |
-
"nar_seconds":
|
| 78 |
-
"vae_seconds": 0.
|
| 79 |
"load": {
|
| 80 |
-
"resolve_and_integrity_seconds":
|
| 81 |
-
"mot_load_seconds":
|
| 82 |
},
|
| 83 |
"stages": {
|
| 84 |
-
"t0->plan_done": 3.
|
| 85 |
-
"plan_done->semantic_done": 8.
|
| 86 |
-
"semantic_done->nar_done":
|
| 87 |
-
"nar_done->vae_done": 0.
|
| 88 |
-
"plan_outside_timer": 0.
|
| 89 |
-
"semantic_outside_timer": 0.
|
| 90 |
-
"semantic_close_seconds": 0.
|
| 91 |
"vae": {
|
| 92 |
-
"load_s": 0.
|
| 93 |
-
"decode_s": 0.
|
| 94 |
-
"post_s":
|
| 95 |
"tiles": [
|
| 96 |
-
0.
|
| 97 |
-
0.
|
| 98 |
-
0.
|
| 99 |
-
0.
|
| 100 |
-
0.
|
| 101 |
-
0.
|
| 102 |
-
0.
|
| 103 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 104 |
],
|
| 105 |
"cudnn_benchmark": false,
|
| 106 |
"cudnn_deterministic": true,
|
| 107 |
-
"core_frames":
|
| 108 |
"input_device": "cpu",
|
| 109 |
"input_contiguous": false
|
| 110 |
},
|
| 111 |
"nar_prefill_source": "handoff",
|
| 112 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 113 |
},
|
| 114 |
-
"e2e_seconds":
|
| 115 |
},
|
| 116 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 117 |
"cfg": 1.0,
|
|
@@ -121,7 +161,7 @@
|
|
| 121 |
"kv_cache_dtype": "int8",
|
| 122 |
"captured_sampling": true,
|
| 123 |
"fuse_rms_quant": true,
|
| 124 |
-
"orbitquant_version": "0.9.
|
| 125 |
"profile": "fast",
|
| 126 |
"allocator_backend": "native",
|
| 127 |
"modules": 224,
|
|
|
|
| 1 |
{
|
| 2 |
"bundle": "v11",
|
| 3 |
+
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,vae_core=256,kv_handoff,lazy_handoff,int8_heads,w8a8,int8_embed",
|
| 4 |
"flags": [
|
| 5 |
"abc_crop",
|
| 6 |
"int8_embed",
|
| 7 |
"int8_heads",
|
| 8 |
"kv_handoff",
|
| 9 |
+
"lazy_handoff",
|
| 10 |
"lookahead",
|
|
|
|
| 11 |
"resident",
|
| 12 |
"vae_compile",
|
| 13 |
+
"vae_core=256",
|
| 14 |
"vae_mode=default",
|
| 15 |
"w8a8"
|
| 16 |
],
|
| 17 |
+
"native_sha256": "118d64db97c3200ef785e83b8d703ebc3b480a799bb17a6affb104d1b4faa28d",
|
| 18 |
"repeat": 1,
|
| 19 |
+
"wall_seconds": 13.488805101253092,
|
| 20 |
+
"load_seconds": 2.229018185287714,
|
| 21 |
"audio_seconds": 152.15866666666668,
|
| 22 |
+
"audio_per_wall": 11.280366609532473,
|
| 23 |
+
"abc_s": 3.1498299119994044,
|
| 24 |
"abc_tokens": 1400,
|
| 25 |
+
"sem_s": 8.052766863256693,
|
| 26 |
"sem_tokens": 3805,
|
| 27 |
+
"abc_ms_per_tok": 2.2498785085710034,
|
| 28 |
+
"sem_ms_per_tok": 2.1163644844301426,
|
| 29 |
+
"nar_s": 2.0583013892173767,
|
| 30 |
+
"vae_s": 0.22332004271447659,
|
| 31 |
+
"nvml_peak_gib": 4.330322265625,
|
| 32 |
+
"torch_peak_gib": 2.3834547996520996,
|
| 33 |
+
"torch_reserved_peak_gib": 3.361328125,
|
| 34 |
"semantic_sha256": "2e35b32e83f3613db2825d4879ffb1a09798f37b8f85b20ce42f7c7624bae895",
|
| 35 |
+
"waveform_sha256": "d1ce52cdce9991784376a9bc1334abd85a3eb8b0870a914ecfdab0dabd87fad5",
|
| 36 |
+
"latent_sha256": "877e8cc0fa28b4515b39b8b3f642e1c9f4db1524825b47eb0e6991772fe1ade6",
|
| 37 |
"timing": {
|
| 38 |
"abc": {
|
| 39 |
+
"seconds": 3.1498299119994044,
|
| 40 |
+
"prefill_seconds": 0.03966649807989597,
|
| 41 |
+
"ttft_seconds": 0.0396859971806407,
|
| 42 |
"output_tokens": 1400,
|
| 43 |
"content_tokens": 1399,
|
| 44 |
+
"output_tps": 444.46844404729393,
|
| 45 |
"prefix_tokens": 423,
|
| 46 |
"cfg_branches": 1,
|
| 47 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 52 |
"lookahead": true,
|
| 53 |
"unroll2": false,
|
| 54 |
"first_step_seconds": null,
|
| 55 |
+
"capture_seconds": 0.03284061420708895,
|
| 56 |
+
"close_seconds": 0.0003672325983643532
|
| 57 |
},
|
| 58 |
"semantic": {
|
| 59 |
+
"seconds": 8.052766863256693,
|
| 60 |
+
"prefill_seconds": 0.049934471026062965,
|
| 61 |
+
"ttft_seconds": 0.04995417129248381,
|
| 62 |
"output_tokens": 3805,
|
| 63 |
"content_tokens": 3804,
|
| 64 |
+
"output_tps": 472.5084017223349,
|
| 65 |
"prefix_tokens": 1824,
|
| 66 |
"cfg_branches": 1,
|
| 67 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 72 |
"lookahead": true,
|
| 73 |
"unroll2": false,
|
| 74 |
"first_step_seconds": null,
|
| 75 |
+
"capture_seconds": 0.03139474894851446,
|
| 76 |
+
"close_seconds": 0.0003659715875983238
|
| 77 |
},
|
| 78 |
+
"nar_seconds": 2.0583013892173767,
|
| 79 |
+
"vae_seconds": 0.22332004271447659,
|
| 80 |
"load": {
|
| 81 |
+
"resolve_and_integrity_seconds": 0.20540913753211498,
|
| 82 |
+
"mot_load_seconds": 0.6672966200858355
|
| 83 |
},
|
| 84 |
"stages": {
|
| 85 |
+
"t0->plan_done": 3.1514371503144503,
|
| 86 |
+
"plan_done->semantic_done": 8.05413985811174,
|
| 87 |
+
"semantic_done->nar_done": 2.0585968447849154,
|
| 88 |
+
"nar_done->vae_done": 0.22365511767566204,
|
| 89 |
+
"plan_outside_timer": 0.0016072383150458336,
|
| 90 |
+
"semantic_outside_timer": 0.0013729948550462723,
|
| 91 |
+
"semantic_close_seconds": 0.0003659715875983238,
|
| 92 |
"vae": {
|
| 93 |
+
"load_s": 0.0010827388614416122,
|
| 94 |
+
"decode_s": 0.22217374853789806,
|
| 95 |
+
"post_s": 1.1587515473365784e-05,
|
| 96 |
"tiles": [
|
| 97 |
+
0.0142,
|
| 98 |
+
0.0136,
|
| 99 |
+
0.0135,
|
| 100 |
+
0.0135,
|
| 101 |
+
0.0135,
|
| 102 |
+
0.0135,
|
| 103 |
+
0.0136,
|
| 104 |
+
0.0135,
|
| 105 |
+
0.0135,
|
| 106 |
+
0.0135,
|
| 107 |
+
0.0135,
|
| 108 |
+
0.0136,
|
| 109 |
+
0.0135,
|
| 110 |
+
0.0135,
|
| 111 |
+
0.032
|
| 112 |
],
|
| 113 |
"cudnn_benchmark": false,
|
| 114 |
"cudnn_deterministic": true,
|
| 115 |
+
"core_frames": 256,
|
| 116 |
"input_device": "cpu",
|
| 117 |
"input_contiguous": false
|
| 118 |
},
|
| 119 |
"nar_prefill_source": "handoff",
|
| 120 |
+
"abc_close_seconds": 0.0003672325983643532,
|
| 121 |
+
"memory": {
|
| 122 |
+
"t0": {
|
| 123 |
+
"peak_gib": 2.383,
|
| 124 |
+
"alloc_gib": 2.383,
|
| 125 |
+
"reserved_gib": 3.244,
|
| 126 |
+
"nvml_used_gib": 4.213
|
| 127 |
+
},
|
| 128 |
+
"plan_done": {
|
| 129 |
+
"peak_gib": 2.658,
|
| 130 |
+
"alloc_gib": 2.383,
|
| 131 |
+
"reserved_gib": 2.883,
|
| 132 |
+
"nvml_used_gib": 3.852
|
| 133 |
+
},
|
| 134 |
+
"semantic_done": {
|
| 135 |
+
"peak_gib": 3.092,
|
| 136 |
+
"alloc_gib": 2.98,
|
| 137 |
+
"reserved_gib": 3.211,
|
| 138 |
+
"nvml_used_gib": 4.18
|
| 139 |
+
},
|
| 140 |
+
"nar_done": {
|
| 141 |
+
"peak_gib": 3.068,
|
| 142 |
+
"alloc_gib": 2.383,
|
| 143 |
+
"reserved_gib": 3.361,
|
| 144 |
+
"nvml_used_gib": 4.33
|
| 145 |
+
},
|
| 146 |
+
"vae_done": {
|
| 147 |
+
"peak_gib": 2.918,
|
| 148 |
+
"alloc_gib": 2.383,
|
| 149 |
+
"reserved_gib": 3.361,
|
| 150 |
+
"nvml_used_gib": 4.33
|
| 151 |
+
}
|
| 152 |
+
}
|
| 153 |
},
|
| 154 |
+
"e2e_seconds": 13.488125356845558
|
| 155 |
},
|
| 156 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 157 |
"cfg": 1.0,
|
|
|
|
| 161 |
"kv_cache_dtype": "int8",
|
| 162 |
"captured_sampling": true,
|
| 163 |
"fuse_rms_quant": true,
|
| 164 |
+
"orbitquant_version": "0.9.8",
|
| 165 |
"profile": "fast",
|
| 166 |
"allocator_backend": "native",
|
| 167 |
"modules": 224,
|
evaluation/profiles/lowmem/run-01/plan.json
CHANGED
|
@@ -9,12 +9,12 @@
|
|
| 9 |
"id": "song"
|
| 10 |
},
|
| 11 |
"timing": {
|
| 12 |
-
"seconds": 3.
|
| 13 |
-
"prefill_seconds": 0.
|
| 14 |
-
"ttft_seconds": 0.
|
| 15 |
"output_tokens": 1400,
|
| 16 |
"content_tokens": 1399,
|
| 17 |
-
"output_tps":
|
| 18 |
"prefix_tokens": 423,
|
| 19 |
"cfg_branches": 1,
|
| 20 |
"sampling": "combined_cuda_graph",
|
|
@@ -25,8 +25,8 @@
|
|
| 25 |
"lookahead": true,
|
| 26 |
"unroll2": false,
|
| 27 |
"first_step_seconds": null,
|
| 28 |
-
"capture_seconds": 0.
|
| 29 |
-
"close_seconds": 0.
|
| 30 |
},
|
| 31 |
"truncated": false,
|
| 32 |
"prefix": [
|
|
|
|
| 9 |
"id": "song"
|
| 10 |
},
|
| 11 |
"timing": {
|
| 12 |
+
"seconds": 3.1498299119994044,
|
| 13 |
+
"prefill_seconds": 0.03966649807989597,
|
| 14 |
+
"ttft_seconds": 0.0396859971806407,
|
| 15 |
"output_tokens": 1400,
|
| 16 |
"content_tokens": 1399,
|
| 17 |
+
"output_tps": 444.46844404729393,
|
| 18 |
"prefix_tokens": 423,
|
| 19 |
"cfg_branches": 1,
|
| 20 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 25 |
"lookahead": true,
|
| 26 |
"unroll2": false,
|
| 27 |
"first_step_seconds": null,
|
| 28 |
+
"capture_seconds": 0.03284061420708895,
|
| 29 |
+
"close_seconds": 0.0003672325983643532
|
| 30 |
},
|
| 31 |
"truncated": false,
|
| 32 |
"prefix": [
|
evaluation/profiles/lowmem/run-01/plan_manifest.json
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
{
|
| 2 |
-
"plan.json": "
|
| 3 |
"abc_tokens.npy": "7fa8ba96997683e2a3d126ea903cd325b32a6d1aeb835c4a57e46ccd334d647e",
|
| 4 |
"prefix.npy": "26b655d943e4b860b3da460c7551d630ed51dd5de8c79ec511eaaa99c680fef4",
|
| 5 |
"score.abc": "ff4ea2b1fab4dc0e10a8c3347b8a9106151d38845493e5a3e11b847a53606e13"
|
|
|
|
| 1 |
{
|
| 2 |
+
"plan.json": "69335439395268b5c586e26f8ea5caaff53490cfba3b69186bd16abe0e752542",
|
| 3 |
"abc_tokens.npy": "7fa8ba96997683e2a3d126ea903cd325b32a6d1aeb835c4a57e46ccd334d647e",
|
| 4 |
"prefix.npy": "26b655d943e4b860b3da460c7551d630ed51dd5de8c79ec511eaaa99c680fef4",
|
| 5 |
"score.abc": "ff4ea2b1fab4dc0e10a8c3347b8a9106151d38845493e5a3e11b847a53606e13"
|
evaluation/profiles/lowmem/run-01/result.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"status": "complete",
|
| 3 |
-
"identity": "
|
| 4 |
"truncated": {
|
| 5 |
"abc": false,
|
| 6 |
"semantic": false
|
|
@@ -29,12 +29,12 @@
|
|
| 29 |
},
|
| 30 |
"timing": {
|
| 31 |
"abc": {
|
| 32 |
-
"seconds": 3.
|
| 33 |
-
"prefill_seconds": 0.
|
| 34 |
-
"ttft_seconds": 0.
|
| 35 |
"output_tokens": 1400,
|
| 36 |
"content_tokens": 1399,
|
| 37 |
-
"output_tps":
|
| 38 |
"prefix_tokens": 423,
|
| 39 |
"cfg_branches": 1,
|
| 40 |
"sampling": "combined_cuda_graph",
|
|
@@ -45,16 +45,16 @@
|
|
| 45 |
"lookahead": true,
|
| 46 |
"unroll2": false,
|
| 47 |
"first_step_seconds": null,
|
| 48 |
-
"capture_seconds": 0.
|
| 49 |
-
"close_seconds": 0.
|
| 50 |
},
|
| 51 |
"semantic": {
|
| 52 |
-
"seconds": 8.
|
| 53 |
-
"prefill_seconds": 0.
|
| 54 |
-
"ttft_seconds": 0.
|
| 55 |
"output_tokens": 3805,
|
| 56 |
"content_tokens": 3804,
|
| 57 |
-
"output_tps":
|
| 58 |
"prefix_tokens": 1824,
|
| 59 |
"cfg_branches": 1,
|
| 60 |
"sampling": "combined_cuda_graph",
|
|
@@ -65,47 +65,86 @@
|
|
| 65 |
"lookahead": true,
|
| 66 |
"unroll2": false,
|
| 67 |
"first_step_seconds": null,
|
| 68 |
-
"capture_seconds": 0.
|
| 69 |
-
"close_seconds": 0.
|
| 70 |
},
|
| 71 |
-
"nar_seconds":
|
| 72 |
-
"vae_seconds": 0.
|
| 73 |
"load": {
|
| 74 |
-
"resolve_and_integrity_seconds":
|
| 75 |
-
"mot_load_seconds":
|
| 76 |
},
|
| 77 |
"stages": {
|
| 78 |
-
"t0->plan_done": 3.
|
| 79 |
-
"plan_done->semantic_done": 8.
|
| 80 |
-
"semantic_done->nar_done":
|
| 81 |
-
"nar_done->vae_done": 0.
|
| 82 |
-
"plan_outside_timer": 0.
|
| 83 |
-
"semantic_outside_timer": 0.
|
| 84 |
-
"semantic_close_seconds": 0.
|
| 85 |
"vae": {
|
| 86 |
-
"load_s": 0.
|
| 87 |
-
"decode_s": 0.
|
| 88 |
-
"post_s":
|
| 89 |
"tiles": [
|
| 90 |
-
0.
|
| 91 |
-
0.
|
| 92 |
-
0.
|
| 93 |
-
0.
|
| 94 |
-
0.
|
| 95 |
-
0.
|
| 96 |
-
0.
|
| 97 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 98 |
],
|
| 99 |
"cudnn_benchmark": false,
|
| 100 |
"cudnn_deterministic": true,
|
| 101 |
-
"core_frames":
|
| 102 |
"input_device": "cpu",
|
| 103 |
"input_contiguous": false
|
| 104 |
},
|
| 105 |
"nar_prefill_source": "handoff",
|
| 106 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 107 |
},
|
| 108 |
-
"e2e_seconds":
|
| 109 |
},
|
| 110 |
"artifacts": {
|
| 111 |
"abc_tokens.npy": {
|
|
@@ -113,23 +152,23 @@
|
|
| 113 |
"bytes": 5724
|
| 114 |
},
|
| 115 |
"audio.flac": {
|
| 116 |
-
"sha256": "
|
| 117 |
-
"bytes":
|
| 118 |
},
|
| 119 |
"config.json": {
|
| 120 |
-
"sha256": "
|
| 121 |
"bytes": 2476
|
| 122 |
},
|
| 123 |
"latent.npy": {
|
| 124 |
-
"sha256": "
|
| 125 |
"bytes": 973952
|
| 126 |
},
|
| 127 |
"plan.json": {
|
| 128 |
-
"sha256": "
|
| 129 |
-
"bytes":
|
| 130 |
},
|
| 131 |
"plan_manifest.json": {
|
| 132 |
-
"sha256": "
|
| 133 |
"bytes": 341
|
| 134 |
},
|
| 135 |
"prefix.npy": {
|
|
|
|
| 1 |
{
|
| 2 |
"status": "complete",
|
| 3 |
+
"identity": "5b78b1460d137eb9aa61003dc8c71bf596658015622424138dc69547e7213123",
|
| 4 |
"truncated": {
|
| 5 |
"abc": false,
|
| 6 |
"semantic": false
|
|
|
|
| 29 |
},
|
| 30 |
"timing": {
|
| 31 |
"abc": {
|
| 32 |
+
"seconds": 3.1498299119994044,
|
| 33 |
+
"prefill_seconds": 0.03966649807989597,
|
| 34 |
+
"ttft_seconds": 0.0396859971806407,
|
| 35 |
"output_tokens": 1400,
|
| 36 |
"content_tokens": 1399,
|
| 37 |
+
"output_tps": 444.46844404729393,
|
| 38 |
"prefix_tokens": 423,
|
| 39 |
"cfg_branches": 1,
|
| 40 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 45 |
"lookahead": true,
|
| 46 |
"unroll2": false,
|
| 47 |
"first_step_seconds": null,
|
| 48 |
+
"capture_seconds": 0.03284061420708895,
|
| 49 |
+
"close_seconds": 0.0003672325983643532
|
| 50 |
},
|
| 51 |
"semantic": {
|
| 52 |
+
"seconds": 8.052766863256693,
|
| 53 |
+
"prefill_seconds": 0.049934471026062965,
|
| 54 |
+
"ttft_seconds": 0.04995417129248381,
|
| 55 |
"output_tokens": 3805,
|
| 56 |
"content_tokens": 3804,
|
| 57 |
+
"output_tps": 472.5084017223349,
|
| 58 |
"prefix_tokens": 1824,
|
| 59 |
"cfg_branches": 1,
|
| 60 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 65 |
"lookahead": true,
|
| 66 |
"unroll2": false,
|
| 67 |
"first_step_seconds": null,
|
| 68 |
+
"capture_seconds": 0.03139474894851446,
|
| 69 |
+
"close_seconds": 0.0003659715875983238
|
| 70 |
},
|
| 71 |
+
"nar_seconds": 2.0583013892173767,
|
| 72 |
+
"vae_seconds": 0.22332004271447659,
|
| 73 |
"load": {
|
| 74 |
+
"resolve_and_integrity_seconds": 0.20540913753211498,
|
| 75 |
+
"mot_load_seconds": 0.6672966200858355
|
| 76 |
},
|
| 77 |
"stages": {
|
| 78 |
+
"t0->plan_done": 3.1514371503144503,
|
| 79 |
+
"plan_done->semantic_done": 8.05413985811174,
|
| 80 |
+
"semantic_done->nar_done": 2.0585968447849154,
|
| 81 |
+
"nar_done->vae_done": 0.22365511767566204,
|
| 82 |
+
"plan_outside_timer": 0.0016072383150458336,
|
| 83 |
+
"semantic_outside_timer": 0.0013729948550462723,
|
| 84 |
+
"semantic_close_seconds": 0.0003659715875983238,
|
| 85 |
"vae": {
|
| 86 |
+
"load_s": 0.0010827388614416122,
|
| 87 |
+
"decode_s": 0.22217374853789806,
|
| 88 |
+
"post_s": 1.1587515473365784e-05,
|
| 89 |
"tiles": [
|
| 90 |
+
0.0142,
|
| 91 |
+
0.0136,
|
| 92 |
+
0.0135,
|
| 93 |
+
0.0135,
|
| 94 |
+
0.0135,
|
| 95 |
+
0.0135,
|
| 96 |
+
0.0136,
|
| 97 |
+
0.0135,
|
| 98 |
+
0.0135,
|
| 99 |
+
0.0135,
|
| 100 |
+
0.0135,
|
| 101 |
+
0.0136,
|
| 102 |
+
0.0135,
|
| 103 |
+
0.0135,
|
| 104 |
+
0.032
|
| 105 |
],
|
| 106 |
"cudnn_benchmark": false,
|
| 107 |
"cudnn_deterministic": true,
|
| 108 |
+
"core_frames": 256,
|
| 109 |
"input_device": "cpu",
|
| 110 |
"input_contiguous": false
|
| 111 |
},
|
| 112 |
"nar_prefill_source": "handoff",
|
| 113 |
+
"abc_close_seconds": 0.0003672325983643532,
|
| 114 |
+
"memory": {
|
| 115 |
+
"t0": {
|
| 116 |
+
"peak_gib": 2.383,
|
| 117 |
+
"alloc_gib": 2.383,
|
| 118 |
+
"reserved_gib": 3.244,
|
| 119 |
+
"nvml_used_gib": 4.213
|
| 120 |
+
},
|
| 121 |
+
"plan_done": {
|
| 122 |
+
"peak_gib": 2.658,
|
| 123 |
+
"alloc_gib": 2.383,
|
| 124 |
+
"reserved_gib": 2.883,
|
| 125 |
+
"nvml_used_gib": 3.852
|
| 126 |
+
},
|
| 127 |
+
"semantic_done": {
|
| 128 |
+
"peak_gib": 3.092,
|
| 129 |
+
"alloc_gib": 2.98,
|
| 130 |
+
"reserved_gib": 3.211,
|
| 131 |
+
"nvml_used_gib": 4.18
|
| 132 |
+
},
|
| 133 |
+
"nar_done": {
|
| 134 |
+
"peak_gib": 3.068,
|
| 135 |
+
"alloc_gib": 2.383,
|
| 136 |
+
"reserved_gib": 3.361,
|
| 137 |
+
"nvml_used_gib": 4.33
|
| 138 |
+
},
|
| 139 |
+
"vae_done": {
|
| 140 |
+
"peak_gib": 2.918,
|
| 141 |
+
"alloc_gib": 2.383,
|
| 142 |
+
"reserved_gib": 3.361,
|
| 143 |
+
"nvml_used_gib": 4.33
|
| 144 |
+
}
|
| 145 |
+
}
|
| 146 |
},
|
| 147 |
+
"e2e_seconds": 13.488125356845558
|
| 148 |
},
|
| 149 |
"artifacts": {
|
| 150 |
"abc_tokens.npy": {
|
|
|
|
| 152 |
"bytes": 5724
|
| 153 |
},
|
| 154 |
"audio.flac": {
|
| 155 |
+
"sha256": "3ff922c8d2c5d7959e572cadc3bcd94d04e0c53212c429938884c37ab9e11801",
|
| 156 |
+
"bytes": 30317169
|
| 157 |
},
|
| 158 |
"config.json": {
|
| 159 |
+
"sha256": "1ca79e079b519059f4656324481ca68ddd6138f12dd333ba544811f339cb9c40",
|
| 160 |
"bytes": 2476
|
| 161 |
},
|
| 162 |
"latent.npy": {
|
| 163 |
+
"sha256": "bbf7fbbfcbdbf5a4c758f297f190cd567bf3edfa5d15929da8ccd43838a02a2c",
|
| 164 |
"bytes": 973952
|
| 165 |
},
|
| 166 |
"plan.json": {
|
| 167 |
+
"sha256": "69335439395268b5c586e26f8ea5caaff53490cfba3b69186bd16abe0e752542",
|
| 168 |
+
"bytes": 33015
|
| 169 |
},
|
| 170 |
"plan_manifest.json": {
|
| 171 |
+
"sha256": "5d1712e55cee8affd69a54d391d2edc20ba768790f28804baf9fd6541d8a8ed6",
|
| 172 |
"bytes": 341
|
| 173 |
},
|
| 174 |
"prefix.npy": {
|
evaluation/profiles/lowmem/summary.json
CHANGED
|
@@ -1,47 +1,48 @@
|
|
| 1 |
[
|
| 2 |
{
|
| 3 |
"bundle": "v11",
|
| 4 |
-
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,kv_handoff,
|
| 5 |
"flags": [
|
| 6 |
"abc_crop",
|
| 7 |
"int8_embed",
|
| 8 |
"int8_heads",
|
| 9 |
"kv_handoff",
|
|
|
|
| 10 |
"lookahead",
|
| 11 |
-
"nar_compile",
|
| 12 |
"resident",
|
| 13 |
"vae_compile",
|
|
|
|
| 14 |
"vae_mode=default",
|
| 15 |
"w8a8"
|
| 16 |
],
|
| 17 |
-
"native_sha256": "
|
| 18 |
"repeat": 0,
|
| 19 |
-
"wall_seconds":
|
| 20 |
-
"load_seconds":
|
| 21 |
"audio_seconds": 152.15866666666668,
|
| 22 |
-
"audio_per_wall":
|
| 23 |
-
"abc_s":
|
| 24 |
"abc_tokens": 1400,
|
| 25 |
-
"sem_s": 8.
|
| 26 |
"sem_tokens": 3805,
|
| 27 |
-
"abc_ms_per_tok":
|
| 28 |
-
"sem_ms_per_tok": 2.
|
| 29 |
-
"nar_s":
|
| 30 |
-
"vae_s":
|
| 31 |
-
"nvml_peak_gib": 4.
|
| 32 |
-
"torch_peak_gib":
|
| 33 |
-
"torch_reserved_peak_gib": 3.
|
| 34 |
"semantic_sha256": "2e35b32e83f3613db2825d4879ffb1a09798f37b8f85b20ce42f7c7624bae895",
|
| 35 |
-
"waveform_sha256": "
|
| 36 |
-
"latent_sha256": "
|
| 37 |
"timing": {
|
| 38 |
"abc": {
|
| 39 |
-
"seconds":
|
| 40 |
-
"prefill_seconds":
|
| 41 |
-
"ttft_seconds":
|
| 42 |
"output_tokens": 1400,
|
| 43 |
"content_tokens": 1399,
|
| 44 |
-
"output_tps":
|
| 45 |
"prefix_tokens": 423,
|
| 46 |
"cfg_branches": 1,
|
| 47 |
"sampling": "combined_cuda_graph",
|
|
@@ -52,16 +53,16 @@
|
|
| 52 |
"lookahead": true,
|
| 53 |
"unroll2": false,
|
| 54 |
"first_step_seconds": null,
|
| 55 |
-
"capture_seconds": 0.
|
| 56 |
-
"close_seconds": 0.
|
| 57 |
},
|
| 58 |
"semantic": {
|
| 59 |
-
"seconds": 8.
|
| 60 |
-
"prefill_seconds": 0.
|
| 61 |
-
"ttft_seconds": 0.
|
| 62 |
"output_tokens": 3805,
|
| 63 |
"content_tokens": 3804,
|
| 64 |
-
"output_tps":
|
| 65 |
"prefix_tokens": 1824,
|
| 66 |
"cfg_branches": 1,
|
| 67 |
"sampling": "combined_cuda_graph",
|
|
@@ -72,47 +73,86 @@
|
|
| 72 |
"lookahead": true,
|
| 73 |
"unroll2": false,
|
| 74 |
"first_step_seconds": null,
|
| 75 |
-
"capture_seconds": 0.
|
| 76 |
-
"close_seconds": 0.
|
| 77 |
},
|
| 78 |
-
"nar_seconds":
|
| 79 |
-
"vae_seconds":
|
| 80 |
"load": {
|
| 81 |
-
"resolve_and_integrity_seconds":
|
| 82 |
-
"mot_load_seconds":
|
| 83 |
},
|
| 84 |
"stages": {
|
| 85 |
-
"t0->plan_done":
|
| 86 |
-
"plan_done->semantic_done": 8.
|
| 87 |
-
"semantic_done->nar_done":
|
| 88 |
-
"nar_done->vae_done":
|
| 89 |
-
"plan_outside_timer": 0.
|
| 90 |
-
"semantic_outside_timer": 0.
|
| 91 |
-
"semantic_close_seconds": 0.
|
| 92 |
"vae": {
|
| 93 |
-
"load_s":
|
| 94 |
-
"decode_s":
|
| 95 |
-
"post_s":
|
| 96 |
"tiles": [
|
| 97 |
-
|
| 98 |
-
0.
|
| 99 |
-
0.
|
| 100 |
-
0.
|
| 101 |
-
0.
|
| 102 |
-
0.
|
| 103 |
-
0.
|
| 104 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 105 |
],
|
| 106 |
"cudnn_benchmark": false,
|
| 107 |
"cudnn_deterministic": true,
|
| 108 |
-
"core_frames":
|
| 109 |
"input_device": "cpu",
|
| 110 |
"input_contiguous": false
|
| 111 |
},
|
| 112 |
"nar_prefill_source": "handoff",
|
| 113 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 114 |
},
|
| 115 |
-
"e2e_seconds":
|
| 116 |
},
|
| 117 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 118 |
"cfg": 1.0,
|
|
@@ -122,7 +162,7 @@
|
|
| 122 |
"kv_cache_dtype": "int8",
|
| 123 |
"captured_sampling": true,
|
| 124 |
"fuse_rms_quant": true,
|
| 125 |
-
"orbitquant_version": "0.9.
|
| 126 |
"profile": "fast",
|
| 127 |
"allocator_backend": "native",
|
| 128 |
"modules": 224,
|
|
@@ -145,47 +185,48 @@
|
|
| 145 |
},
|
| 146 |
{
|
| 147 |
"bundle": "v11",
|
| 148 |
-
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,kv_handoff,
|
| 149 |
"flags": [
|
| 150 |
"abc_crop",
|
| 151 |
"int8_embed",
|
| 152 |
"int8_heads",
|
| 153 |
"kv_handoff",
|
|
|
|
| 154 |
"lookahead",
|
| 155 |
-
"nar_compile",
|
| 156 |
"resident",
|
| 157 |
"vae_compile",
|
|
|
|
| 158 |
"vae_mode=default",
|
| 159 |
"w8a8"
|
| 160 |
],
|
| 161 |
-
"native_sha256": "
|
| 162 |
"repeat": 1,
|
| 163 |
-
"wall_seconds":
|
| 164 |
-
"load_seconds":
|
| 165 |
"audio_seconds": 152.15866666666668,
|
| 166 |
-
"audio_per_wall":
|
| 167 |
-
"abc_s": 3.
|
| 168 |
"abc_tokens": 1400,
|
| 169 |
-
"sem_s": 8.
|
| 170 |
"sem_tokens": 3805,
|
| 171 |
-
"abc_ms_per_tok": 2.
|
| 172 |
-
"sem_ms_per_tok": 2.
|
| 173 |
-
"nar_s":
|
| 174 |
-
"vae_s": 0.
|
| 175 |
-
"nvml_peak_gib": 4.
|
| 176 |
-
"torch_peak_gib":
|
| 177 |
-
"torch_reserved_peak_gib": 3.
|
| 178 |
"semantic_sha256": "2e35b32e83f3613db2825d4879ffb1a09798f37b8f85b20ce42f7c7624bae895",
|
| 179 |
-
"waveform_sha256": "
|
| 180 |
-
"latent_sha256": "
|
| 181 |
"timing": {
|
| 182 |
"abc": {
|
| 183 |
-
"seconds": 3.
|
| 184 |
-
"prefill_seconds": 0.
|
| 185 |
-
"ttft_seconds": 0.
|
| 186 |
"output_tokens": 1400,
|
| 187 |
"content_tokens": 1399,
|
| 188 |
-
"output_tps":
|
| 189 |
"prefix_tokens": 423,
|
| 190 |
"cfg_branches": 1,
|
| 191 |
"sampling": "combined_cuda_graph",
|
|
@@ -196,16 +237,16 @@
|
|
| 196 |
"lookahead": true,
|
| 197 |
"unroll2": false,
|
| 198 |
"first_step_seconds": null,
|
| 199 |
-
"capture_seconds": 0.
|
| 200 |
-
"close_seconds": 0.
|
| 201 |
},
|
| 202 |
"semantic": {
|
| 203 |
-
"seconds": 8.
|
| 204 |
-
"prefill_seconds": 0.
|
| 205 |
-
"ttft_seconds": 0.
|
| 206 |
"output_tokens": 3805,
|
| 207 |
"content_tokens": 3804,
|
| 208 |
-
"output_tps":
|
| 209 |
"prefix_tokens": 1824,
|
| 210 |
"cfg_branches": 1,
|
| 211 |
"sampling": "combined_cuda_graph",
|
|
@@ -216,47 +257,86 @@
|
|
| 216 |
"lookahead": true,
|
| 217 |
"unroll2": false,
|
| 218 |
"first_step_seconds": null,
|
| 219 |
-
"capture_seconds": 0.
|
| 220 |
-
"close_seconds": 0.
|
| 221 |
},
|
| 222 |
-
"nar_seconds":
|
| 223 |
-
"vae_seconds": 0.
|
| 224 |
"load": {
|
| 225 |
-
"resolve_and_integrity_seconds":
|
| 226 |
-
"mot_load_seconds":
|
| 227 |
},
|
| 228 |
"stages": {
|
| 229 |
-
"t0->plan_done": 3.
|
| 230 |
-
"plan_done->semantic_done": 8.
|
| 231 |
-
"semantic_done->nar_done":
|
| 232 |
-
"nar_done->vae_done": 0.
|
| 233 |
-
"plan_outside_timer": 0.
|
| 234 |
-
"semantic_outside_timer": 0.
|
| 235 |
-
"semantic_close_seconds": 0.
|
| 236 |
"vae": {
|
| 237 |
-
"load_s": 0.
|
| 238 |
-
"decode_s": 0.
|
| 239 |
-
"post_s":
|
| 240 |
"tiles": [
|
| 241 |
-
0.
|
| 242 |
-
0.
|
| 243 |
-
0.
|
| 244 |
-
0.
|
| 245 |
-
0.
|
| 246 |
-
0.
|
| 247 |
-
0.
|
| 248 |
-
0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 249 |
],
|
| 250 |
"cudnn_benchmark": false,
|
| 251 |
"cudnn_deterministic": true,
|
| 252 |
-
"core_frames":
|
| 253 |
"input_device": "cpu",
|
| 254 |
"input_contiguous": false
|
| 255 |
},
|
| 256 |
"nar_prefill_source": "handoff",
|
| 257 |
-
"abc_close_seconds": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 258 |
},
|
| 259 |
-
"e2e_seconds":
|
| 260 |
},
|
| 261 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 262 |
"cfg": 1.0,
|
|
@@ -266,7 +346,7 @@
|
|
| 266 |
"kv_cache_dtype": "int8",
|
| 267 |
"captured_sampling": true,
|
| 268 |
"fuse_rms_quant": true,
|
| 269 |
-
"orbitquant_version": "0.9.
|
| 270 |
"profile": "fast",
|
| 271 |
"allocator_backend": "native",
|
| 272 |
"modules": 224,
|
|
|
|
| 1 |
[
|
| 2 |
{
|
| 3 |
"bundle": "v11",
|
| 4 |
+
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,vae_core=256,kv_handoff,lazy_handoff,int8_heads,w8a8,int8_embed",
|
| 5 |
"flags": [
|
| 6 |
"abc_crop",
|
| 7 |
"int8_embed",
|
| 8 |
"int8_heads",
|
| 9 |
"kv_handoff",
|
| 10 |
+
"lazy_handoff",
|
| 11 |
"lookahead",
|
|
|
|
| 12 |
"resident",
|
| 13 |
"vae_compile",
|
| 14 |
+
"vae_core=256",
|
| 15 |
"vae_mode=default",
|
| 16 |
"w8a8"
|
| 17 |
],
|
| 18 |
+
"native_sha256": "118d64db97c3200ef785e83b8d703ebc3b480a799bb17a6affb104d1b4faa28d",
|
| 19 |
"repeat": 0,
|
| 20 |
+
"wall_seconds": 16.067145991139114,
|
| 21 |
+
"load_seconds": 2.229018185287714,
|
| 22 |
"audio_seconds": 152.15866666666668,
|
| 23 |
+
"audio_per_wall": 9.470173903354137,
|
| 24 |
+
"abc_s": 3.7763062892481685,
|
| 25 |
"abc_tokens": 1400,
|
| 26 |
+
"sem_s": 8.044851490296423,
|
| 27 |
"sem_tokens": 3805,
|
| 28 |
+
"abc_ms_per_tok": 2.697361635177263,
|
| 29 |
+
"sem_ms_per_tok": 2.1142842287244217,
|
| 30 |
+
"nar_s": 2.187400832772255,
|
| 31 |
+
"vae_s": 2.0492115300148726,
|
| 32 |
+
"nvml_peak_gib": 4.213134765625,
|
| 33 |
+
"torch_peak_gib": 2.3834547996520996,
|
| 34 |
+
"torch_reserved_peak_gib": 3.244140625,
|
| 35 |
"semantic_sha256": "2e35b32e83f3613db2825d4879ffb1a09798f37b8f85b20ce42f7c7624bae895",
|
| 36 |
+
"waveform_sha256": "d1ce52cdce9991784376a9bc1334abd85a3eb8b0870a914ecfdab0dabd87fad5",
|
| 37 |
+
"latent_sha256": "877e8cc0fa28b4515b39b8b3f642e1c9f4db1524825b47eb0e6991772fe1ade6",
|
| 38 |
"timing": {
|
| 39 |
"abc": {
|
| 40 |
+
"seconds": 3.7763062892481685,
|
| 41 |
+
"prefill_seconds": 0.6570839313790202,
|
| 42 |
+
"ttft_seconds": 0.6582063101232052,
|
| 43 |
"output_tokens": 1400,
|
| 44 |
"content_tokens": 1399,
|
| 45 |
+
"output_tps": 370.73263998369384,
|
| 46 |
"prefix_tokens": 423,
|
| 47 |
"cfg_branches": 1,
|
| 48 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 53 |
"lookahead": true,
|
| 54 |
"unroll2": false,
|
| 55 |
"first_step_seconds": null,
|
| 56 |
+
"capture_seconds": 0.04039210919290781,
|
| 57 |
+
"close_seconds": 0.0003549950197339058
|
| 58 |
},
|
| 59 |
"semantic": {
|
| 60 |
+
"seconds": 8.044851490296423,
|
| 61 |
+
"prefill_seconds": 0.05773051269352436,
|
| 62 |
+
"ttft_seconds": 0.05774910096079111,
|
| 63 |
"output_tokens": 3805,
|
| 64 |
"content_tokens": 3804,
|
| 65 |
+
"output_tps": 472.97330529836785,
|
| 66 |
"prefix_tokens": 1824,
|
| 67 |
"cfg_branches": 1,
|
| 68 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 73 |
"lookahead": true,
|
| 74 |
"unroll2": false,
|
| 75 |
"first_step_seconds": null,
|
| 76 |
+
"capture_seconds": 0.032243385910987854,
|
| 77 |
+
"close_seconds": 0.0004120608791708946
|
| 78 |
},
|
| 79 |
+
"nar_seconds": 2.187400832772255,
|
| 80 |
+
"vae_seconds": 2.0492115300148726,
|
| 81 |
"load": {
|
| 82 |
+
"resolve_and_integrity_seconds": 0.20540913753211498,
|
| 83 |
+
"mot_load_seconds": 0.6672966200858355
|
| 84 |
},
|
| 85 |
"stages": {
|
| 86 |
+
"t0->plan_done": 3.7815310433506966,
|
| 87 |
+
"plan_done->semantic_done": 8.046199827454984,
|
| 88 |
+
"semantic_done->nar_done": 2.1876858919858932,
|
| 89 |
+
"nar_done->vae_done": 2.049526182934642,
|
| 90 |
+
"plan_outside_timer": 0.005224754102528095,
|
| 91 |
+
"semantic_outside_timer": 0.0013483371585607529,
|
| 92 |
+
"semantic_close_seconds": 0.0004120608791708946,
|
| 93 |
"vae": {
|
| 94 |
+
"load_s": 0.026218397542834282,
|
| 95 |
+
"decode_s": 2.0229244586080313,
|
| 96 |
+
"post_s": 1.2959353625774384e-05,
|
| 97 |
"tiles": [
|
| 98 |
+
1.7943,
|
| 99 |
+
0.0139,
|
| 100 |
+
0.014,
|
| 101 |
+
0.0142,
|
| 102 |
+
0.0141,
|
| 103 |
+
0.0141,
|
| 104 |
+
0.0139,
|
| 105 |
+
0.014,
|
| 106 |
+
0.0141,
|
| 107 |
+
0.0144,
|
| 108 |
+
0.0141,
|
| 109 |
+
0.0142,
|
| 110 |
+
0.014,
|
| 111 |
+
0.0142,
|
| 112 |
+
0.0452
|
| 113 |
],
|
| 114 |
"cudnn_benchmark": false,
|
| 115 |
"cudnn_deterministic": true,
|
| 116 |
+
"core_frames": 256,
|
| 117 |
"input_device": "cpu",
|
| 118 |
"input_contiguous": false
|
| 119 |
},
|
| 120 |
"nar_prefill_source": "handoff",
|
| 121 |
+
"abc_close_seconds": 0.0003549950197339058,
|
| 122 |
+
"memory": {
|
| 123 |
+
"t0": {
|
| 124 |
+
"peak_gib": 2.127,
|
| 125 |
+
"alloc_gib": 2.127,
|
| 126 |
+
"reserved_gib": 2.424,
|
| 127 |
+
"nvml_used_gib": 3.342
|
| 128 |
+
},
|
| 129 |
+
"plan_done": {
|
| 130 |
+
"peak_gib": 2.402,
|
| 131 |
+
"alloc_gib": 2.127,
|
| 132 |
+
"reserved_gib": 2.629,
|
| 133 |
+
"nvml_used_gib": 3.573
|
| 134 |
+
},
|
| 135 |
+
"semantic_done": {
|
| 136 |
+
"peak_gib": 2.836,
|
| 137 |
+
"alloc_gib": 2.723,
|
| 138 |
+
"reserved_gib": 2.977,
|
| 139 |
+
"nvml_used_gib": 3.92
|
| 140 |
+
},
|
| 141 |
+
"nar_done": {
|
| 142 |
+
"peak_gib": 2.82,
|
| 143 |
+
"alloc_gib": 2.136,
|
| 144 |
+
"reserved_gib": 3.127,
|
| 145 |
+
"nvml_used_gib": 4.086
|
| 146 |
+
},
|
| 147 |
+
"vae_done": {
|
| 148 |
+
"peak_gib": 2.918,
|
| 149 |
+
"alloc_gib": 2.383,
|
| 150 |
+
"reserved_gib": 3.244,
|
| 151 |
+
"nvml_used_gib": 4.213
|
| 152 |
+
}
|
| 153 |
+
}
|
| 154 |
},
|
| 155 |
+
"e2e_seconds": 16.065859724767506
|
| 156 |
},
|
| 157 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 158 |
"cfg": 1.0,
|
|
|
|
| 162 |
"kv_cache_dtype": "int8",
|
| 163 |
"captured_sampling": true,
|
| 164 |
"fuse_rms_quant": true,
|
| 165 |
+
"orbitquant_version": "0.9.8",
|
| 166 |
"profile": "fast",
|
| 167 |
"allocator_backend": "native",
|
| 168 |
"modules": 224,
|
|
|
|
| 185 |
},
|
| 186 |
{
|
| 187 |
"bundle": "v11",
|
| 188 |
+
"mode": "lookahead,resident,abc_crop,vae_compile,vae_mode=default,vae_core=256,kv_handoff,lazy_handoff,int8_heads,w8a8,int8_embed",
|
| 189 |
"flags": [
|
| 190 |
"abc_crop",
|
| 191 |
"int8_embed",
|
| 192 |
"int8_heads",
|
| 193 |
"kv_handoff",
|
| 194 |
+
"lazy_handoff",
|
| 195 |
"lookahead",
|
|
|
|
| 196 |
"resident",
|
| 197 |
"vae_compile",
|
| 198 |
+
"vae_core=256",
|
| 199 |
"vae_mode=default",
|
| 200 |
"w8a8"
|
| 201 |
],
|
| 202 |
+
"native_sha256": "118d64db97c3200ef785e83b8d703ebc3b480a799bb17a6affb104d1b4faa28d",
|
| 203 |
"repeat": 1,
|
| 204 |
+
"wall_seconds": 13.488805101253092,
|
| 205 |
+
"load_seconds": 2.229018185287714,
|
| 206 |
"audio_seconds": 152.15866666666668,
|
| 207 |
+
"audio_per_wall": 11.280366609532473,
|
| 208 |
+
"abc_s": 3.1498299119994044,
|
| 209 |
"abc_tokens": 1400,
|
| 210 |
+
"sem_s": 8.052766863256693,
|
| 211 |
"sem_tokens": 3805,
|
| 212 |
+
"abc_ms_per_tok": 2.2498785085710034,
|
| 213 |
+
"sem_ms_per_tok": 2.1163644844301426,
|
| 214 |
+
"nar_s": 2.0583013892173767,
|
| 215 |
+
"vae_s": 0.22332004271447659,
|
| 216 |
+
"nvml_peak_gib": 4.330322265625,
|
| 217 |
+
"torch_peak_gib": 2.3834547996520996,
|
| 218 |
+
"torch_reserved_peak_gib": 3.361328125,
|
| 219 |
"semantic_sha256": "2e35b32e83f3613db2825d4879ffb1a09798f37b8f85b20ce42f7c7624bae895",
|
| 220 |
+
"waveform_sha256": "d1ce52cdce9991784376a9bc1334abd85a3eb8b0870a914ecfdab0dabd87fad5",
|
| 221 |
+
"latent_sha256": "877e8cc0fa28b4515b39b8b3f642e1c9f4db1524825b47eb0e6991772fe1ade6",
|
| 222 |
"timing": {
|
| 223 |
"abc": {
|
| 224 |
+
"seconds": 3.1498299119994044,
|
| 225 |
+
"prefill_seconds": 0.03966649807989597,
|
| 226 |
+
"ttft_seconds": 0.0396859971806407,
|
| 227 |
"output_tokens": 1400,
|
| 228 |
"content_tokens": 1399,
|
| 229 |
+
"output_tps": 444.46844404729393,
|
| 230 |
"prefix_tokens": 423,
|
| 231 |
"cfg_branches": 1,
|
| 232 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 237 |
"lookahead": true,
|
| 238 |
"unroll2": false,
|
| 239 |
"first_step_seconds": null,
|
| 240 |
+
"capture_seconds": 0.03284061420708895,
|
| 241 |
+
"close_seconds": 0.0003672325983643532
|
| 242 |
},
|
| 243 |
"semantic": {
|
| 244 |
+
"seconds": 8.052766863256693,
|
| 245 |
+
"prefill_seconds": 0.049934471026062965,
|
| 246 |
+
"ttft_seconds": 0.04995417129248381,
|
| 247 |
"output_tokens": 3805,
|
| 248 |
"content_tokens": 3804,
|
| 249 |
+
"output_tps": 472.5084017223349,
|
| 250 |
"prefix_tokens": 1824,
|
| 251 |
"cfg_branches": 1,
|
| 252 |
"sampling": "combined_cuda_graph",
|
|
|
|
| 257 |
"lookahead": true,
|
| 258 |
"unroll2": false,
|
| 259 |
"first_step_seconds": null,
|
| 260 |
+
"capture_seconds": 0.03139474894851446,
|
| 261 |
+
"close_seconds": 0.0003659715875983238
|
| 262 |
},
|
| 263 |
+
"nar_seconds": 2.0583013892173767,
|
| 264 |
+
"vae_seconds": 0.22332004271447659,
|
| 265 |
"load": {
|
| 266 |
+
"resolve_and_integrity_seconds": 0.20540913753211498,
|
| 267 |
+
"mot_load_seconds": 0.6672966200858355
|
| 268 |
},
|
| 269 |
"stages": {
|
| 270 |
+
"t0->plan_done": 3.1514371503144503,
|
| 271 |
+
"plan_done->semantic_done": 8.05413985811174,
|
| 272 |
+
"semantic_done->nar_done": 2.0585968447849154,
|
| 273 |
+
"nar_done->vae_done": 0.22365511767566204,
|
| 274 |
+
"plan_outside_timer": 0.0016072383150458336,
|
| 275 |
+
"semantic_outside_timer": 0.0013729948550462723,
|
| 276 |
+
"semantic_close_seconds": 0.0003659715875983238,
|
| 277 |
"vae": {
|
| 278 |
+
"load_s": 0.0010827388614416122,
|
| 279 |
+
"decode_s": 0.22217374853789806,
|
| 280 |
+
"post_s": 1.1587515473365784e-05,
|
| 281 |
"tiles": [
|
| 282 |
+
0.0142,
|
| 283 |
+
0.0136,
|
| 284 |
+
0.0135,
|
| 285 |
+
0.0135,
|
| 286 |
+
0.0135,
|
| 287 |
+
0.0135,
|
| 288 |
+
0.0136,
|
| 289 |
+
0.0135,
|
| 290 |
+
0.0135,
|
| 291 |
+
0.0135,
|
| 292 |
+
0.0135,
|
| 293 |
+
0.0136,
|
| 294 |
+
0.0135,
|
| 295 |
+
0.0135,
|
| 296 |
+
0.032
|
| 297 |
],
|
| 298 |
"cudnn_benchmark": false,
|
| 299 |
"cudnn_deterministic": true,
|
| 300 |
+
"core_frames": 256,
|
| 301 |
"input_device": "cpu",
|
| 302 |
"input_contiguous": false
|
| 303 |
},
|
| 304 |
"nar_prefill_source": "handoff",
|
| 305 |
+
"abc_close_seconds": 0.0003672325983643532,
|
| 306 |
+
"memory": {
|
| 307 |
+
"t0": {
|
| 308 |
+
"peak_gib": 2.383,
|
| 309 |
+
"alloc_gib": 2.383,
|
| 310 |
+
"reserved_gib": 3.244,
|
| 311 |
+
"nvml_used_gib": 4.213
|
| 312 |
+
},
|
| 313 |
+
"plan_done": {
|
| 314 |
+
"peak_gib": 2.658,
|
| 315 |
+
"alloc_gib": 2.383,
|
| 316 |
+
"reserved_gib": 2.883,
|
| 317 |
+
"nvml_used_gib": 3.852
|
| 318 |
+
},
|
| 319 |
+
"semantic_done": {
|
| 320 |
+
"peak_gib": 3.092,
|
| 321 |
+
"alloc_gib": 2.98,
|
| 322 |
+
"reserved_gib": 3.211,
|
| 323 |
+
"nvml_used_gib": 4.18
|
| 324 |
+
},
|
| 325 |
+
"nar_done": {
|
| 326 |
+
"peak_gib": 3.068,
|
| 327 |
+
"alloc_gib": 2.383,
|
| 328 |
+
"reserved_gib": 3.361,
|
| 329 |
+
"nvml_used_gib": 4.33
|
| 330 |
+
},
|
| 331 |
+
"vae_done": {
|
| 332 |
+
"peak_gib": 2.918,
|
| 333 |
+
"alloc_gib": 2.383,
|
| 334 |
+
"reserved_gib": 3.361,
|
| 335 |
+
"nvml_used_gib": 4.33
|
| 336 |
+
}
|
| 337 |
+
}
|
| 338 |
},
|
| 339 |
+
"e2e_seconds": 13.488125356845558
|
| 340 |
},
|
| 341 |
"gpu": "NVIDIA GeForce RTX 4090",
|
| 342 |
"cfg": 1.0,
|
|
|
|
| 346 |
"kv_cache_dtype": "int8",
|
| 347 |
"captured_sampling": true,
|
| 348 |
"fuse_rms_quant": true,
|
| 349 |
+
"orbitquant_version": "0.9.8",
|
| 350 |
"profile": "fast",
|
| 351 |
"allocator_backend": "native",
|
| 352 |
"modules": 224,
|
evaluation/profiles/off/run-00/config.json
ADDED
|
@@ -0,0 +1,88 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"generation": {
|
| 3 |
+
"abc": {
|
| 4 |
+
"temperature": 0.7,
|
| 5 |
+
"top_p": 0.9,
|
| 6 |
+
"top_k": 30,
|
| 7 |
+
"repetition_penalty": 1.005,
|
| 8 |
+
"penalty_window": 100,
|
| 9 |
+
"min_tokens": 32,
|
| 10 |
+
"max_tokens": 4096
|
| 11 |
+
},
|
| 12 |
+
"semantic": {
|
| 13 |
+
"temperature": 1.0,
|
| 14 |
+
"top_p": 0.95,
|
| 15 |
+
"top_k": 100,
|
| 16 |
+
"repetition_penalty": 1.2,
|
| 17 |
+
"penalty_window": 50,
|
| 18 |
+
"min_tokens": 200,
|
| 19 |
+
"max_tokens": 9000
|
| 20 |
+
},
|
| 21 |
+
"ode_steps": 8,
|
| 22 |
+
"ode_method": "midpoint",
|
| 23 |
+
"context": 24576,
|
| 24 |
+
"version": "yue2-native-v1"
|
| 25 |
+
},
|
| 26 |
+
"overrides": {
|
| 27 |
+
"ode_steps": 8
|
| 28 |
+
},
|
| 29 |
+
"cot": "full",
|
| 30 |
+
"cfg_scale": 1.0,
|
| 31 |
+
"cfg_negative": "same_instruction_and_exact_abc",
|
| 32 |
+
"backend": "torch",
|
| 33 |
+
"quantization": "orbitquant",
|
| 34 |
+
"model_dtype": "bfloat16",
|
| 35 |
+
"vae_dtype": "float32",
|
| 36 |
+
"vae_decode": "halo_crop",
|
| 37 |
+
"vae_core_frames": 512,
|
| 38 |
+
"vae_halo_frames": 16,
|
| 39 |
+
"device": "cuda:0",
|
| 40 |
+
"memory_budget_gib": 18.0,
|
| 41 |
+
"offload_ar": false,
|
| 42 |
+
"runtime_sha256": "c354c01c9b04ef1dad51891857040d20790473fac7cb787042b70f6f9f451643",
|
| 43 |
+
"decoder_release": "standard",
|
| 44 |
+
"validation_status": "unvalidated",
|
| 45 |
+
"runtime_profile": "fast",
|
| 46 |
+
"fuse_rms_quant": true,
|
| 47 |
+
"captured_sampling": true,
|
| 48 |
+
"kv_cache_dtype": "int8",
|
| 49 |
+
"int8_attention": "ordered",
|
| 50 |
+
"runtime_bundle_sha256": "51c8dd94465183f85c9759831fba4318bd89780458923f2cd3a3a5daa1e37952",
|
| 51 |
+
"vae_compute_dtype": "torch.float16",
|
| 52 |
+
"expert_staging": false,
|
| 53 |
+
"orbitquant": {
|
| 54 |
+
"quant_method": "orbitquant",
|
| 55 |
+
"weight_bits": 4,
|
| 56 |
+
"activation_bits": 4,
|
| 57 |
+
"rotation": "rpbh",
|
| 58 |
+
"rotation_seed": 0,
|
| 59 |
+
"block_size": "paper",
|
| 60 |
+
"codebook": "lloyd_max",
|
| 61 |
+
"codebook_version": 2,
|
| 62 |
+
"codebook_dtype": "float32",
|
| 63 |
+
"row_norm_dtype": "bfloat16",
|
| 64 |
+
"activation_norm_dtype": "float32",
|
| 65 |
+
"activation_eps": 1e-10,
|
| 66 |
+
"weight_pack_dtype": "uint8",
|
| 67 |
+
"target_policy": "universal",
|
| 68 |
+
"adaln_policy": "int4_rtn",
|
| 69 |
+
"adaln_group_size": 64,
|
| 70 |
+
"modules_to_convert": [],
|
| 71 |
+
"modules_to_use_adaln": [],
|
| 72 |
+
"modules_to_not_convert": [],
|
| 73 |
+
"modules_dtype_dict": {},
|
| 74 |
+
"artifact_format_version": 1,
|
| 75 |
+
"runtime_mode": "auto_fused",
|
| 76 |
+
"activation_kernel_backend": "auto",
|
| 77 |
+
"packed_matmul_block_m": 64,
|
| 78 |
+
"packed_matmul_block_n": 64,
|
| 79 |
+
"packed_matmul_block_k": 128,
|
| 80 |
+
"packed_matmul_num_warps": 4,
|
| 81 |
+
"weight_row_tile_size": 256,
|
| 82 |
+
"w4a4_int8_weight_cache": false,
|
| 83 |
+
"lowbit_boundary_protection": "auto",
|
| 84 |
+
"lowbit_protected_blocks": 4,
|
| 85 |
+
"lowbit_protected_bits": 4,
|
| 86 |
+
"lowbit_interior_protection": "auto"
|
| 87 |
+
}
|
| 88 |
+
}
|
evaluation/profiles/off/run-00/metrics.json
ADDED
|
@@ -0,0 +1,165 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"bundle": "v11",
|
| 3 |
+
"mode": "base",
|
| 4 |
+
"flags": [],
|
| 5 |
+
"native_sha256": "118d64db97c3200ef785e83b8d703ebc3b480a799bb17a6affb104d1b4faa28d",
|
| 6 |
+
"repeat": 0,
|
| 7 |
+
"wall_seconds": 19.164482676424086,
|
| 8 |
+
"load_seconds": 2.261388538405299,
|
| 9 |
+
"audio_seconds": 168.91866666666667,
|
| 10 |
+
"audio_per_wall": 8.814152175078974,
|
| 11 |
+
"abc_s": 4.439309379085898,
|
| 12 |
+
"abc_tokens": 1424,
|
| 13 |
+
"sem_s": 9.244723854586482,
|
| 14 |
+
"sem_tokens": 4224,
|
| 15 |
+
"abc_ms_per_tok": 3.117492541492906,
|
| 16 |
+
"sem_ms_per_tok": 2.1886183367865724,
|
| 17 |
+
"nar_s": 2.7404489433392882,
|
| 18 |
+
"vae_s": 2.7345637548714876,
|
| 19 |
+
"nvml_peak_gib": 5.316650390625,
|
| 20 |
+
"torch_peak_gib": 0.2797279357910156,
|
| 21 |
+
"torch_reserved_peak_gib": 0.482421875,
|
| 22 |
+
"semantic_sha256": "36a231d034d554c12388e82b4cd4be2598079ae23f0e7fdfb14187e339adcb0c",
|
| 23 |
+
"waveform_sha256": "4580778f3e57663e34b77e329dd1bbf89ad647a6a250dbfbd8df6ffe8372f79f",
|
| 24 |
+
"latent_sha256": "5c566f70f058de65c5f3e8daf3dab9846cb354ad72b26df9680d472f193b6c7d",
|
| 25 |
+
"timing": {
|
| 26 |
+
"abc": {
|
| 27 |
+
"seconds": 4.439309379085898,
|
| 28 |
+
"prefill_seconds": 0.6292197797447443,
|
| 29 |
+
"ttft_seconds": 0.6302690282464027,
|
| 30 |
+
"output_tokens": 1424,
|
| 31 |
+
"content_tokens": 1423,
|
| 32 |
+
"output_tps": 320.77061506653024,
|
| 33 |
+
"prefix_tokens": 423,
|
| 34 |
+
"cfg_branches": 1,
|
| 35 |
+
"sampling": "combined_cuda_graph",
|
| 36 |
+
"execution": "cuda_graph",
|
| 37 |
+
"attention": "int8",
|
| 38 |
+
"combined_graph_replays": 1423,
|
| 39 |
+
"device_history": true,
|
| 40 |
+
"lookahead": false,
|
| 41 |
+
"unroll2": false,
|
| 42 |
+
"first_step_seconds": 0.0437213946133852,
|
| 43 |
+
"capture_seconds": null,
|
| 44 |
+
"close_seconds": 0.0002884352579712868
|
| 45 |
+
},
|
| 46 |
+
"semantic": {
|
| 47 |
+
"seconds": 9.244723854586482,
|
| 48 |
+
"prefill_seconds": 0.05751974321901798,
|
| 49 |
+
"ttft_seconds": 0.05754018295556307,
|
| 50 |
+
"output_tokens": 4224,
|
| 51 |
+
"content_tokens": 4223,
|
| 52 |
+
"output_tps": 456.9092669982126,
|
| 53 |
+
"prefix_tokens": 1848,
|
| 54 |
+
"cfg_branches": 1,
|
| 55 |
+
"sampling": "combined_cuda_graph",
|
| 56 |
+
"execution": "cuda_graph",
|
| 57 |
+
"attention": "int8",
|
| 58 |
+
"combined_graph_replays": 4223,
|
| 59 |
+
"device_history": true,
|
| 60 |
+
"lookahead": false,
|
| 61 |
+
"unroll2": false,
|
| 62 |
+
"first_step_seconds": 0.034588039852678776,
|
| 63 |
+
"capture_seconds": null,
|
| 64 |
+
"close_seconds": 0.00040249619632959366
|
| 65 |
+
},
|
| 66 |
+
"nar_seconds": 2.7404489433392882,
|
| 67 |
+
"vae_seconds": 2.7345637548714876,
|
| 68 |
+
"load": {
|
| 69 |
+
"resolve_and_integrity_seconds": 0.19746538531035185,
|
| 70 |
+
"mot_load_seconds": 0.6815848043188453
|
| 71 |
+
},
|
| 72 |
+
"stages": {
|
| 73 |
+
"t0->plan_done": 4.441724243573844,
|
| 74 |
+
"plan_done->semantic_done": 9.246008949354291,
|
| 75 |
+
"semantic_done->nar_done": 2.7407320588827133,
|
| 76 |
+
"nar_done->vae_done": 2.734832789748907,
|
| 77 |
+
"plan_outside_timer": 0.0024148644879460335,
|
| 78 |
+
"semantic_outside_timer": 0.0012850947678089142,
|
| 79 |
+
"semantic_close_seconds": 0.00040249619632959366,
|
| 80 |
+
"vae": {
|
| 81 |
+
"load_s": 1.6986743062734604,
|
| 82 |
+
"decode_s": 0.8483156505972147,
|
| 83 |
+
"post_s": 0.07316656317561865,
|
| 84 |
+
"tiles": [
|
| 85 |
+
0.2446,
|
| 86 |
+
0.0855,
|
| 87 |
+
0.0808,
|
| 88 |
+
0.0806,
|
| 89 |
+
0.0807,
|
| 90 |
+
0.081,
|
| 91 |
+
0.0804,
|
| 92 |
+
0.0809,
|
| 93 |
+
0.0338
|
| 94 |
+
],
|
| 95 |
+
"cudnn_benchmark": false,
|
| 96 |
+
"cudnn_deterministic": true,
|
| 97 |
+
"core_frames": 512,
|
| 98 |
+
"input_device": "cpu",
|
| 99 |
+
"input_contiguous": false
|
| 100 |
+
},
|
| 101 |
+
"nar_prefill_source": "prefill",
|
| 102 |
+
"abc_close_seconds": 0.0002884352579712868,
|
| 103 |
+
"memory": {
|
| 104 |
+
"t0": {
|
| 105 |
+
"peak_gib": 2.831,
|
| 106 |
+
"alloc_gib": 2.831,
|
| 107 |
+
"reserved_gib": 3.068,
|
| 108 |
+
"nvml_used_gib": 3.928
|
| 109 |
+
},
|
| 110 |
+
"plan_done": {
|
| 111 |
+
"peak_gib": 3.108,
|
| 112 |
+
"alloc_gib": 2.855,
|
| 113 |
+
"reserved_gib": 3.371,
|
| 114 |
+
"nvml_used_gib": 4.328
|
| 115 |
+
},
|
| 116 |
+
"semantic_done": {
|
| 117 |
+
"peak_gib": 3.566,
|
| 118 |
+
"alloc_gib": 2.863,
|
| 119 |
+
"reserved_gib": 3.719,
|
| 120 |
+
"nvml_used_gib": 4.676
|
| 121 |
+
},
|
| 122 |
+
"nar_done": {
|
| 123 |
+
"peak_gib": 4.048,
|
| 124 |
+
"alloc_gib": 2.864,
|
| 125 |
+
"reserved_gib": 4.357,
|
| 126 |
+
"nvml_used_gib": 5.317
|
| 127 |
+
},
|
| 128 |
+
"vae_done": {
|
| 129 |
+
"peak_gib": 2.864,
|
| 130 |
+
"alloc_gib": 0.28,
|
| 131 |
+
"reserved_gib": 0.482,
|
| 132 |
+
"nvml_used_gib": 1.451
|
| 133 |
+
}
|
| 134 |
+
}
|
| 135 |
+
},
|
| 136 |
+
"e2e_seconds": 19.163636380806565
|
| 137 |
+
},
|
| 138 |
+
"gpu": "NVIDIA GeForce RTX 4090",
|
| 139 |
+
"cfg": 1.0,
|
| 140 |
+
"steps": 8,
|
| 141 |
+
"runtime": {
|
| 142 |
+
"int8_attention": "ordered",
|
| 143 |
+
"kv_cache_dtype": "int8",
|
| 144 |
+
"captured_sampling": true,
|
| 145 |
+
"fuse_rms_quant": true,
|
| 146 |
+
"orbitquant_version": "0.9.8",
|
| 147 |
+
"profile": "fast",
|
| 148 |
+
"allocator_backend": "native",
|
| 149 |
+
"modules": 224,
|
| 150 |
+
"logical_projections": 392,
|
| 151 |
+
"fused": true,
|
| 152 |
+
"effective_modes": {
|
| 153 |
+
"native_packed_matmul": 224
|
| 154 |
+
},
|
| 155 |
+
"activation_backends": {
|
| 156 |
+
"native_cuda_int8_surrogate": 168,
|
| 157 |
+
"triton_cuda_packed_w4": 56
|
| 158 |
+
},
|
| 159 |
+
"decoded_weight_caches": 0,
|
| 160 |
+
"int8_weight_caches": 0,
|
| 161 |
+
"packed_weight_bytes": 1409286144
|
| 162 |
+
},
|
| 163 |
+
"allocator": "expandable_segments:True",
|
| 164 |
+
"measurement": "first process request"
|
| 165 |
+
}
|