PinpointQA-Qwen3-VL-8B-Instruct-LoRA

LoRA adapter weights for Qwen/Qwen3-VL-8B-Instruct, fine-tuned on PinpointQA.

Overview

• Base model: Qwen/Qwen3-VL-8B-Instruct
• Dataset: RainChow/PinpointQA
• Paper: https://huggingface.co/papers/2604.08991
• Method: LoRA
• Rank (r): 64
• LoRA alpha: 128
• LoRA dropout: 0.05

Notes

• This repository contains adapter weights only.
• It does not include merged full model weights.
• The adapter should be used together with the original base model and its corresponding processor/tokenizer.
• The supervised training data was derived from the official PinpointQA training split and further converted into the model-specific instruction/chat format used for SFT. Therefore, the raw train.jsonl in the dataset repository is not necessarily identical to the final serialized training samples used in training.

Related Resources

• Base model: https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct
• Dataset: https://huggingface.co/datasets/RainChow/PinpointQA
• Paper: https://huggingface.co/papers/2604.08991