• internlm/CapRL-Qwen3VL-4B

• text-generation-inference
• multimodal
• image caption
• captioning

• en

CapRL-Qwen3VL-4B-GGUF

CapRL-Qwen3VL-4B from internlm is a 4B-parameter vision-language model from the CapRL 2.0 series, fine-tuned from Qwen3-VL-4B using an upgraded Reinforcement Learning with Verifiable Rewards (RLVR) two-stage pipeline—LVLMs generate rich captions followed by vision-only LLM QA evaluation on a rigorously filtered, diverse image dataset—significantly outperforming CapRL-Qwen2.5VL-3B and Qwen2.5-VL-72B in captioning tasks while offering high performance and advanced abilities for charts, infographics, documents, and natural images with structured, hallucination-minimal outputs. As the top model in the CapRL series (vs. 2B for speed/efficiency), it delivers remarkable visual understanding, comprehensive information coverage, and well-organized descriptions via vLLM OpenAI-compatible API (gpumemoryutilization=0.95), supporting base64-encoded images for precise text extraction and detailed captioning without SFT memorization issues. Part of ongoing advancements with QA curation code and GGUF quantizations, it's ideal for research, annotation, and production deployment balancing compute cost and superior perception.

CapRL-Qwen3VL-4B [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):

!image.png