yongxianwei/Qwen2.5-VL-32B-OCR

VISIONPricing:Input $1.06 / Cached $0.053 / Output $2.6Concurrent Unit Cost:2Model Size:32BQuant:FP8Context Size:32kPublished:Nov 18, 2025License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

yongxianwei/Qwen2.5-VL-32B-OCR is a 32 billion parameter vision-language model, fine-tuned by Yongxianwei from the Qwen2.5-VL-32B-Instruct base model. It is specifically optimized as an OCR expert, designed to excel in optical character recognition tasks within the OptMerge MLLM merging benchmark. This model specializes in extracting and understanding text from images, making it suitable for applications requiring high-accuracy text recognition in visual contexts.

Loading preview...

Overview

yongxianwei/Qwen2.5-VL-32B-OCR is a 32 billion parameter vision-language model, developed by Yongxianwei, specifically fine-tuned for Optical Character Recognition (OCR) tasks. It serves as a dedicated OCR expert within the OptMerge MLLM merging benchmark, designed to be merged with other specialized experts without requiring additional training data. This model is built upon the powerful Qwen/Qwen2.5-VL-32B-Instruct base model.

Key Capabilities

  • Specialized OCR Performance: Achieves strong results on various OCR-related benchmarks, including TextVQA and OCRVQA, demonstrating its proficiency in extracting and understanding text from images.
  • OptMerge Framework Component: Designed as one of five capability experts (VQA, Geometry, Chart, OCR, Grounding) that can be unified into a single model using the OptMerge methodology, enabling flexible multimodal LLM development.
  • Full Weights Fine-tuning: The model provides full weights from supervised fine-tuning, eliminating the need for PEFT adapters and simplifying deployment.

Good For

  • Applications requiring high-accuracy text extraction from images: Ideal for tasks such as document analysis, visual question answering involving text, and scene text recognition.
  • Researchers exploring model merging techniques: Particularly useful for those working with the OptMerge framework to combine specialized multimodal LLM capabilities.
  • Developers building multimodal systems: Offers a robust OCR component that can be integrated into larger vision-language pipelines.

Finetunes
1 models