Hugging Face's logo

Wiselnn
/
Qwen2-VL-videorope-128frames-8k-context-330k-llava-video

Safetensors
qwen2_vl
Model card Files
xet
 Community

VideoRoPE: What Makes for Good Video Rotary Position Embedding?

Trained model: Qwen2VL Vision Tower + Qwen2 Language Model
RoPE type: VideoRoPE

To use this model, simply set which_type='videorope' and scale_factor=2.0.
For more details, please refer to the code implementation.

Downloads last month
3
Safetensors
Model size
8B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Dataset used to train Wiselnn/Qwen2-VL-videorope-128frames-8k-context-330k-llava-video

Collection including Wiselnn/Qwen2-VL-videorope-128frames-8k-context-330k-llava-video

Paper for Wiselnn/Qwen2-VL-videorope-128frames-8k-context-330k-llava-video