[Feature] Load weights from distributed #470
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
25 tasks
Collaborator
|
How is the performance so far? |
Collaborator
Author
|
Need to offload encoder for CIs to pass first |
Edenzzzz
commented
Jun 28, 2025
Collaborator
Author
Edenzzzz
left a comment
There was a problem hiding this comment.
Let's merge this to avoid more conflicts from get_local_torch_device
Collaborator
|
why the transformer test failed? |
BrianChen1129
approved these changes
Jun 28, 2025
Collaborator
Author
|
#557 (comment) it failed a while ago |
Collaborator
okay, then just merge |
SolitaryThinker
added a commit
that referenced
this pull request
Jun 29, 2025
This reverts commit c5155b2.
qimcis
pushed a commit
to qimcis/FastVideo
that referenced
this pull request
Oct 30, 2025
2 tasks
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The read speed of nvme SSDs is only a few GB/s, far slower than the 800GB/s on NVLink. This PR supports loading weights on one rank and broadcasting them to the node.
After this PR, text encoders will be offloaded (w/ layer-wise prefetch) to cpu by default using FSDP, because encoder tests will simply OOM on A40 without this.
Results on H100 and 14B model (
Wan-AI/Wan2.1-T2V-14B-Diffusers)python examples/inference/basic/basic.pyTODO