FaithEyes-8B-SFT-GGUF

FaithEyes-8B-SFT is the cold-start supervised-fine-tuning checkpoint of FaithEyes on Qwen3-VL-8B-Instruct, the first stage of the two-stage SFT + RL pipeline introduced in the paper "FaithEyes: Towards Faithful Tool Use via Multi-Agent Process-Image Self-Verification." FaithEyes is a multi-agent self-judging framework in which a single VLM plays two roles β€” a main agent that solves visual questions by interleaving reasoning with executable code-based tool calls (like image crops, zooms, rotations, or contrast adjustments), and a subagent that judges whether each process image the main agent produces is actually helpful, emitting a {"is_helpful": true/false, "reasons": ...} verdict that both feeds back into the tool observation and scales the tool reward. This SFT checkpoint equips the base model with three prerequisite capabilities before reinforcement learning: writing executable Python for image manipulation, faithfulness judging (assessing whether a process image helps answer the question and outputting the verdict as JSON), and feedback-driven reasoning that conditions subsequent steps on the subagent's judgement rather than ignoring it β€” with the main agent and subagent sharing the same VLM and trained jointly, differing only in their prompts. As reported in the paper, this SFT-only model merely imitates demonstrations without yet autonomously discriminating helpful from unhelpful tool calls; that discrimination ability is added by the subsequent FaithEyes-8B-RL checkpoint via GRPO. Training built on the verl framework alongside DeepEyes, Thyme, ms_swift, and VLMEvalKit, and the model is released under the Apache 2.0 license. FaithEyes-8B-SFT on Hugging Face

Model Files

File Name Quant Type File Size File Link Description
FaithEyes-8B-SFT.BF16.gguf BF16 16.4 GB Link Full BF16 weights. Highest quality, largest file size.
FaithEyes-8B-SFT.Q3_K_L.gguf Q3_K_L 4.43 GB Link Lower quality but usable, good for low RAM availability.
FaithEyes-8B-SFT.Q3_K_M.gguf Q3_K_M 4.12 GB Link Low quality.
FaithEyes-8B-SFT.Q4_K_M.gguf Q4_K_M 5.03 GB Link Good quality, default size for most use cases, recommended.
FaithEyes-8B-SFT.Q4_K_S.gguf Q4_K_S 4.8 GB Link Slightly lower quality with more space savings, recommended.
FaithEyes-8B-SFT.Q5_K_M.gguf Q5_K_M 5.85 GB Link High quality, recommended.
FaithEyes-8B-SFT.Q5_K_S.gguf Q5_K_S 5.72 GB Link High quality, recommended.
FaithEyes-8B-SFT.mmproj-bf16.gguf mmproj-bf16 1.16 GB Link Multimodal projection file in BF16 format. Used for vision/language models.

llama.cpp

LLM inference in C/C++ β€” https://github.com/ggml-org/llama.cpp

Downloads last month
515
GGUF
Model size
8B params
Architecture
qwen3vl
Hardware compatibility
Log In to add your hardware

3-bit

4-bit

5-bit

16-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for prithivMLmods/FaithEyes-8B-SFT-GGUF

Quantized
(1)
this model

Collections including prithivMLmods/FaithEyes-8B-SFT-GGUF