ChartVerse-2B-GGUF

ChartVerse-2B from opendatalab is a compact 2B-parameter vision-language model fine-tuned from Qwen3-VL-2B-Instruct, specialized for complex chart reasoning as part of the ChartVerse project, achieving superior performance (54.3% average across 6 challenging benchmarks) over larger 7B chart-specific models like ECD-7B (50.0%), START-7B (52.5%), and Chart-R1-7B (53.6%) through high-quality training on ChartVerse-SFT-600K (412K unique charts, 603K QA pairs with 3.9B CoT tokens, rollout posterior entropy 0.44) and ChartVerse-RL-40K (40K hardest samples filtered by 0<r(Q)<1 failure rate) using LLaMA-Factory SFT (lr=1e-5, batch=128, ctx=22k) followed by veRL GSPO RL (lr=1e-6, rollout=16). Despite its small size, it demonstrates that data quality offsets scale limitations with strong Chain-of-Thought reasoning for multi-step analysis, truth-anchored QA verification via code execution, and Apache 2.0 licensing for deployment via Transformers/Qwen3VLForConditionalGeneration (max_new_tokens=16384) on standard GPUs, outperforming baselines in visual understanding while supporting project evaluation tools like VLMEvalKit.

ChartVerse-2B [GGUF]

Quants Usage

(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)

Here is a handy graph by ikawrakow comparing some lower-quality quant types (lower is better):