- 
                Notifications
    You must be signed in to change notification settings 
- Fork 248
Bump torchtune pin to a 9-24 commit; Update Flamingo Definition #1195
Conversation
| 🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/torchchat/1195
 Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 5ce5e9d with merge base 6d2ef4a ( This comment was automatically generated by Dr. CI and updates every 15 minutes. | 
| set -x | ||
| $PIP_EXECUTABLE install git+https://github.com/pytorch/torchtune.git@18efc81dda1c537bb7c25058ff059b4623ccff58 | ||
| ) | 
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This works on Mac now right?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
yeah and mac tests passed
        
          
                torchchat/generate.py
              
                Outdated
          
        
      | or self.model.config.model_type == ModelType.Flamingo | ||
| ): | ||
| model.setup_caches(max_batch_size=1, dtype=self.dtype) | ||
| model.setup_caches(batch_size=1, dtype=self.dtype, encoder_max_seq_len=6404, decoder_max_seq_len=max_seq_length-1) | 
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Magic number 6404?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
it is the size that can hold single image input while one-gpu affordable.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I will leave a comment
Torchtune updates flamingo model definition, as well as its inference. This PR aims to update torchchat pipeline to sync with torchtune