llava-hf

/

LLaVA-NeXT-Video-7B-hf

like 56

Follow

Llava Hugging Face 150

in the config.json. the scaling factor is 2.5. But no matter in the source code or in the convert weight python file(https://github.com/huggingface/transformers/blob/main/src/transformers/models/llava_next_video/convert_llava_next_video_weights_to_hf.py), all of them showed that the factor shoud be 2.0 when predict. And also the inference result is way much better in 2.0