LoRA weights
#2
by lesan20 - opened
Hello,
Thank you for your work on InternVideo2 — the project is very interesting and useful.
I’m currently working with the InternVideo2-CLIP-1B-224p-f8 model and noticed that the provided 1B_clip.pth checkpoint seems to miss LoRA weights (lora_A / lora_B). As a result, the text encoder produces nearly identical embeddings for different inputs.
Could you please clarify whether a full checkpoint with trained LoRA adapters is available?
Thank you in advance!