Skip to content

getting_started.md

Latest commit

 

History

History
26 lines (14 loc) · 802 Bytes

getting_started.md

File metadata and controls

26 lines (14 loc) · 802 Bytes
Raw

Getting Started

Preparation of dataset

We perform the distillation recipe on VLM-SFT dataset curated by SOLO.

huggingface-cli download --repo-type dataset yangyiy/VLM-SFT --local-dir ./data/vlm_sft

Then you should unzip all the chunk.tar.gz files into the 'images' folder.

Preparation of teacher model HoVLE

We use the pre-trained HoVLE model as the teacher model.

huggingface-cli download  OpenGVLab/HoVLE --local-dir ./weights/HoVLE

Note

If you do not have the access to huggingface, you can use the mirror site by setting the environment variable export HF_ENDPOINT=https://hf-mirror.com.