which stage to train for videos?

[surbhi-pro-vigil]

I only have videos with description and I want to ado video understanding. can I directly start fine tuning from stage 3? if yes where will we get the weights till stage 2? with this particular command, no weights are downloaded..only some files are downloaded.
python scripts/convert_hf_checkpoint.py --model_path DAMO-NLP-SG/VideoLLaMA3-7B --save_path weights/videollama3_7b_local

[zillur01]

Hi, have you figured this out? I want to LoRA fine-tune it on a custom dataset. not sure about the stages

[kashrest]

My group fine-tuned from stage 3 for video understanding. I believe the base weights for the pre-trained model released by the original authors will be downloaded and then fine-tuned further in stage 3. You don't need to get particular weights, just start from the base model checkpoint provided through their HuggingFace page.