You can deploy this model and start earning money today!

yunfengwang/
TVP-Pretrain-Qwen2VL-2B

Base model: Qwen/Qwen2-VL-2B-Instruct
LoRA: r=16, alpha=32
Training data: ~146K COCO grounding samples
Coordinates: Normalized to [0, 999]

Name: TVP-Pretrain-Qwen2VL-2B
Availability: InStock

Views

No views yet

TVP-Pretrain-Qwen2VL-2B

Stage 1: Pretraining — Teaches the base model to output visual primitive tokens (<|box|>, <|point|>, <|ref|>).

python

from model import VisualPrimitiveVLM
model = VisualPrimitiveVLM.from_pretrained("yunfengwang/TVP-Pretrain-Qwen2VL-2B", device_map="cuda")

See the project repo for full instructions.