sft_2b — LoRA SFT adapter (r=16, LM-only) for Qwen3-VL-2B on 40k VQAv2+GQA
Part of vqa-rlvr: post-training Qwen3-VL for visual question answering on a single RTX 4090 (QLoRA SFT + GRPO with verifiable rewards).
Results (full eval sets; VQAv2 / GQA / CLEVR / TextVQA)
- Short template: 79.9 / 62.3 / 99.8 / 79.6
- Reasoning template: 71.8 / 57.9 / 99.8 / 71.3
Metrics: official VQA accuracy (VQAv2/TextVQA), normalized EM (GQA/CLEVR); harness cross-checked against lmms-eval. Full tables, configs, and per-run JSONs: https://github.com/guangboyu/vqa-rlvr.
Usage
from peft import PeftModel
from transformers import AutoModelForImageTextToText
base = AutoModelForImageTextToText.from_pretrained("Qwen/Qwen3-VL-2B-Instruct", dtype="bfloat16")
model = PeftModel.from_pretrained(base, "omnifish123/vqa-rlvr-sft-2b").merge_and_unload()
Training config is in run_config.json in this repo.
- Downloads last month
- 33
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for omnifish123/vqa-rlvr-sft-2b
Base model
Qwen/Qwen3-VL-2B-Instruct