Downloads · 30 days
10
23% of all-time downloads
lzy-vlm-lab-opd/qwen3-vl-8b-pure-rl-seed42-step100
qwen3-vl-8b-pure-rl-seed42-step100 is a image-text-to-text model from lzy-vlm-lab-opd. Use it for the image-text-to-text task on the model card, and read the license before you ship it in a product. It is set up for transformers.
This model is a GRPO-finetuned checkpoint based on Qwen3-VL-8B-Instruct. It was trained for multimodal visual reasoning and question answering with seed 42. This repository contains the inference-ready weights saved a…
Downloads · 30 days
10
23% of all-time downloads
All-time downloads
44
Public
Parameters
770K
17.5 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors17.5 GB · 100%
From the Hugging Face model README
This model is a GRPO-finetuned checkpoint based on Qwen3-VL-8B-Instruct. It was trained for multimodal visual reasoning and question answering with seed 42. This repository contains the inference-ready weights saved at training step 100.
The training prompt asks the model to finish each response with exactly one final answer inside <answer>...</answer>. For multiple-choice questions, only the option letter should appear inside the answer tag.
The checkpoint was saved with Transformers 4.57.3 in bfloat16 format.
from transformers import AutoProcessor, Qwen3VLForConditionalGeneration
model_id = "YOUR_ORG/YOUR_MODEL_NAME"
model = Qwen3VLForConditionalGeneration.from_pretrained(
model_id,
torch_dtype="auto",
device_map="auto",
)
processor = AutoProcessor.from_pretrained(model_id)
Use the processor's chat template when preparing multimodal messages.
The training mix covered visual reasoning, diagrams, charts, documents, OCR, science questions, spatial reasoning, and general visual question answering.
This is a research checkpoint. Its outputs may be incorrect, may not consistently follow the requested answer format, and should be independently evaluated before deployment or use in high-stakes settings.