Qwen3-VL Histogram QA Adapter

1. Predict a numeric answer from a histogram image and a natural-language question.
2. Generate a short explanation conditioned on that predicted answer.

Intended Space Pairing

• Model repo: YifanQiao/qwen3vl4b-hist-qa-checkpoint-723
• Space repo: YifanQiao/histogram-vqa-demo

Files To Include In This Model Repo

• adapter_model.safetensors
• adapter_config.json
• processor_config.json
• tokenizer.json
• tokenizer_config.json
• chat_template.jinja
• README.md

Local Source Mapping

• deploy_bundle/adapter_model.safetensors -> adapter_model.safetensors
• deploy_bundle/adapter_config.json -> adapter_config.json
• deploy_bundle/processor_config.json -> processor_config.json
• deploy_bundle/tokenizer.json -> tokenizer.json
• deploy_bundle/tokenizer_config.json -> tokenizer_config.json
• deploy_bundle/chat_template.jinja -> chat_template.jinja

Runtime Notes

• Base model: Qwen/Qwen3-VL-4B-Instruct
• Adapter type: LoRA / PEFT
• Expected interface: one uploaded histogram image, one question, one numeric answer, one short explanation
• Free CPU Spaces may be slow for this model size, but the paired Gradio Space is structured so you can test the product flow first and upgrade hardware later if needed.

Example Inference Pattern

1. AutoProcessor.from_pretrained(<this-repo>)
2. AutoModelForImageTextToText.from_pretrained("Qwen/Qwen3-VL-4B-Instruct")
3. PeftModel.from_pretrained(base_model, <this-repo>)