steven0226/qwen3vl-8b-chartqa-gguf

Qwen3-VL-8B ChartQA — GGUF Q4KM

Persistent CPU artifact converted from the pinned fine-tuned merged checkpoint `steven0226/qwen3vl-8b-chartqa-merged-16bit@519060ef43df3261e0512e5ae4c82a4d4e675f32`.

Files

• —Qwen3VL-8B-ChartQA-Q4_K_M.gguf — fine-tuned language model, Q4KM
• —mmproj-Qwen3VL-8B-ChartQA-Q8_0.gguf — vision encoder/projector, Q8_0

Conversion used llama.cpp commit 79bba02a6741de194912d370015866414faa83ad. conversion_metadata.json records byte sizes, digests, lineage, and a content-free smoke-test outcome.

Verification scope

One pinned ChartQA test sample fetched at runtime passed an exact-match check. The chart, question, label, and model output are not published. This is a smoke verification, not a complete GGUF quality evaluation.

The formal 2,500-question quality result belongs to the separately evaluated AWQ/vLLM artifact: 85.52%, -0.72 pp versus merged 16-bit, passing the predefined 2 pp gate.

llama.cpp

llama-server \
  --model Qwen3VL-8B-ChartQA-Q4_K_M.gguf \
  --mmproj mmproj-Qwen3VL-8B-ChartQA-Q8_0.gguf \
  --ctx-size 4096 --jinja

Publication boundary

This model repository contains weights and content-free metadata. It does not redistribute ChartQA images, questions, labels, or raw model predictions.

Source repository

程式碼、訓練與評估流程、測試與完整證據都在 GitHub：<https://github.com/kuotunyu/qwen3-vl-chartqa>。GitHub kuotunyu 與 Hugging Face steven0226 為同一人。

Source code, the training/evaluation pipeline, tests and the full evidence trail live at <https://github.com/kuotunyu/qwen3-vl-chartqa>. GitHub kuotunyu and Hugging Face steven0226 are the same author.