thu-ml / thu-ml/TurboDiffusion

Sparse Linear Layer weights are 0 for 14B 480p model

Open
#109 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
3.7k
Forks
277
Avg merge
2h 57m
Merged PRs (30d)
2

Description

I downloaded them from Huggingface and upon inspection I noticed that layers with "local_attn" in their names are just 0 for 14B 480p model. This is not the case for 1.3B model. I was wondering if this is intended, or should I load it differently? I used regular torch.load to obtain state dictionary.

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by comparing the Hugging Face 14B 480p checkpoint with the 1.3B model, focusing on state-dictionary layers whose names contain "local_attn" and the regular torch.load path. Done means determining whether the zero weights are intentional or whether the checkpoint requires a different loading procedure.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.