ggml : limit n_threads to the max n_tasks #5238
Merged
Conversation
This file contains 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
|
It fixes prompt processing speed, but not token generation. I'll do more tresting later today. |
|
This effectively sets the number of threads to 1 with full offload, so if the regression was caused by the change to |
|
No I'm not completely sure, I'll try testing again soon. |
|
I can also reproduce the performance regression with CUDA, it's worse on windows but it is also noticeable under WSL, and this does fix it for me. |
ggerganov
approved these changes
Jan 31, 2024
|
Ok, it's working. I messed up my first test, but this does fix the issue. |
jordankanter
pushed a commit
to jordankanter/llama.cpp
that referenced
this pull request
Feb 3, 2024
hodlen
pushed a commit
to hodlen/llama.cpp
that referenced
this pull request
Apr 1, 2024
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Outdated suggestions cannot be applied.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.
This avoids using a larger number of threads that the ops in the graph can actually use.
With full offload, the CPU graph only contains a get_rows that can only use one thread, so this should fix the performance regression when full offloading due to the overhead of launching additional threads. @stduhpf can you check if this fixes the issue?