Implement asynchronous module prefetching (QWEN only so far) #10594
+111
−23
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
The async offload streams reason for existence is to transfer from RAM to GPU. The post processing compute steps are a bonus on the side stream, but if the compute stream is running a long kernel, it can stall the side stream, as it wait to type-cast the bias before transferring the weight. So do a pure xfer of the weight straight up, then do everything bias, then go back to fix the weight type and do weight patches.
Implement an API that allows instrumenting a model with a prefetch queue. Units of work are on the nn.Module level.
rattus128
force-pushed
the
prs/prefetching
branch
from
e05d78f to
ec37c80
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.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.
Draft of a generic module prefetcher. Implement the core feature and give one example of how to use it with QWEN.
This is able to get very close to compute saturation whereas --async-offload as-is still has a few compute stalls.
Leaving as a draft for now, as I am still trying to find a better way.
Start comfy use QWEN to try it out. You need the following startup args:
It consumes a bit extra VRAM so you need to --reserve-vram to avoid OOMing.