Document GPTQ compression and size comparison

Rename-facing model card update with GPTQ settings, calibration, artifact sizes, and base-model comparison.

README.md

@@ -57,18 +57,15 @@ retrieval evaluation, the candidate caption is never provided to the image branc
 Clone the repository and install the tested runtime dependencies:
 
 ```bash
-git clone https://huggingface.co/konic-labs/lfm2.5-encoder-230m-siglip2-monet-gptq-int4
-cd lfm2.5-encoder-230m-siglip2-monet-gptq-int4
+git clone https://huggingface.co/konic-labs/LFM2.5-multimodal-encoder-230M
+cd LFM2.5-multimodal-encoder-230M
 
-pip install torch transformers pillow accelerate safetensors pydantic loguru psutil
+pip install torch transformers pillow accelerate safetensors compressed-tensors
 ```
 
-The release includes the compressed-tensors runtime used to validate the GPTQ artifact.
-Expose it together with the repository code:
-
-```bash
-export PYTHONPATH="$PWD/vendor:$PWD:$PYTHONPATH"
-```
+GPTQ loading requires a `compressed-tensors` installation compatible with the model
+configuration. The tested quantization details are recorded in `recipe.yaml` and
+`compression_manifest.json`.
 
 The SigLIP2 vision model is downloaded from Hugging Face on first use. Run one image-text
 matching example:
@@ -111,7 +108,6 @@ experiments.
 | `recipe.yaml` | Quantization recipe |
 | `run_multimodal.py` | Local image-text inference example |
 | `evaluation/` | Saved metrics and deep-evaluation summaries |
-| `vendor/` | Tested compressed-tensors runtime source |
 
 ## Evaluation
 
@@ -155,17 +151,48 @@ Cyclic-negative matching AUROC is `0.9733` on MONET. Text-nearest hard-negative
 is `0.4961`, so the current model is substantially better at coarse compatibility than
 fine-grained semantic discrimination.
 
-## Compression and runtime
+## Compression details and size comparison
+
+The released LFM2.5 body uses GPTQ INT4 W4A16 compression:
+
+| Compression setting | Value |
+|---|---|
+| Method | GPTQ INT4 |
+| Weight/activation format | W4A16 |
+| Quantized layers | Linear layers; `lm_head` excluded |
+| Block size | 128 |
+| Activation order | Static |
+| Dampening fraction | 0.01 |
+| Calibration | 256 fused multimodal `inputs_embeds` samples |
+| Maximum calibration sequence length | 160 tokens |
+| Compressor revision | `8cec0acc1931de6f8f73257151ab7007c14dbf4e` |
+
+The calibration samples contained actual projected visual tokens rather than text-only
+inputs, so the quantization pass reflects the multimodal path.
+
+### Storage comparison
+
+| Artifact | Model weights | Full package | Relative to original package |
+|---|---:|---:|---:|
+| Original LFM2.5 Encoder 230M FP32 | 918.79 MB | 923.65 MB | 100% |
+| Clean BF16 reference | 459.40 MB | 474.81 MB | 51.4% |
+| Clean GPTQ INT4 release | 355.03 MB | 370.46 MB | 40.1% |
+
+The GPTQ release is approximately **59.9% smaller than the original package** and
+**22.0% smaller than the clean BF16 package**. The external SigLIP2 vision tower is not
+included in these sizes and is downloaded separately.
+
+### Fidelity and runtime
 
 | Metric | GPTQ INT4 |
 |---|---:|
-| Artifact size | approximately 370 MB |
 | Hidden cosine versus BF16 | 0.9427 |
 | Logit cosine versus BF16 | 0.9654 |
 | Multimodal latency, batch 4 | 45.7 ms on NVIDIA L4 |
+| Peak multimodal VRAM | 2.38 GiB on NVIDIA L4 |
 
 The current GPTQ backend is intended primarily for weight-size reduction. Native BF16
-is faster in the tested environment.
+is faster and uses less runtime VRAM in the tested environment.
 
 ## Limitations