Fourbitfourbit
Request access
VLMpopular

Qwen2.5-VL 72B

Alibaba · 72B · Image · video · OCR

Strongest open-weights VLM at the 72B tier. Excellent OCR and structured-document grounding — pair with a few hundred K labeled pages and it will outperform GPT-4o on your domain.

Train this model →Read the recipe
Quickstart

Launch a Qwen2.5-VL 72B fine-tune

~ fourbit
$ fourbit train \
  --base qwen2-5-vl-72b \
  --data s3://acme/dataset.jsonl \
  --recipe lora
More VLMs