Disable logging during precompilation #539
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
michaelbenayoun
force-pushed
the
disable_logging_during_precompilation
branch
from
ae57c3f
to
7095426
Compare
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
JingyaHuang
approved these changes
Mar 29, 2024
|
The inf2 builders have changed, and for the TGI workflow that uses the Makefile, it is missing the gawk package that was previously there. |
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Outdated suggestions cannot be applied.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.
What does this PR do?
neuron_parallel_compilereads the logs to extract the HLO graphs to compile. For some reason, strange characters are logged which makesneuron_parallel_compilefail when it tries to read the file, resulting in only a small portion of the total number of graphs being compiled. This in turn results in compilation during actual training which makes things prohibitively long. To avoid all of this we simply do not log.dp_rank=tp_rank=0 and pp_rank = pp_size - 1under PP setting so that Wandb and other callback can actually track the loss. While this is not ideal, and will be improved in the coming days, it works for now.