DAMO-NLP-SG / DAMO-NLP-SG/VideoLLaMA3
Training error on stage 1 when following 'bash scripts/train/stage1_2b.sh': AttributeError: 'Linear' object has no attribute 'cal_proj_size'
- Dominant language
- Jupyter Notebook
- Stars
- 1.2k
- Forks
- 89
- PR merge metrics
- No merged PRs in 30d
Description
I face an error when i'm trying to train videollama3 locally, which is:

and it originates from this code part :

I try to figure it out and find that the 'config' doesn't get the argument of 'mm_project_type', which should be 'mlp2x_gelu', so the model directly build the mm_projector as defualt which is 'linear', and the linear mm_project has no 'cal_proj_size' which is only built when type is mlp2x_gelu, so that the error occurs.

I'm not sure if there is anything i missunderstand or miss, so I ask for a favour
Thanks a lot!
Contributor guide
No contributing guide indexed for this repository
Research direction
Start with scripts/train/stage1_2b.sh and trace how mm_project_type reaches the model's projector configuration. Compare the default Linear path with the mlp2x_gelu path and reproduce the reported cal_proj_size error. Done means stage 1 starts without the AttributeError and uses the intended projector configuration.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- bash
- Domain
- ai, machine-learning
- Issue type
- Bug
- Difficulty
- 3/5
- Estimated time
- 1-2 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100