Add Multi-tiered Checkpointing Support to Pathways - #3952
Conversation
Codecov Report❌ Patch coverage is
📢 Thoughts on this report? Let us know! |
|
This PR has been automatically marked as stale because it has not had recent activity. It will be closed soon if no further activity occurs. Thank you for your contributions. |
31e7104 to
3651ea8
Compare
3651ea8 to
5513d51
Compare
25934d5 to
bbfd1a7
Compare
|
🤖 Hi @abhinavclemson, I've received your request, and I'm working on it now! You can track my progress in the logs for more details. |
|
🤖 I'm sorry @abhinavclemson, but I was unable to process your request. Please see the logs for more details. |
dfd1730 to
948c186
Compare
|
🤖 Hi @abhinavclemson, I've received your request, and I'm working on it now! You can track my progress in the logs for more details. |
|
🤖 I'm sorry @abhinavclemson, but I was unable to process your request. Please see the logs for more details. |
948c186 to
3071a95
Compare
| def test_engram_integration(self): | ||
| """AOT test for Engram implementation""" | ||
| compiled_trainstep_file = "/tmp/test_engram_integration" | ||
| transformers.AutoConfig.register("deepseek_v32", DeepseekV32Config) |
There was a problem hiding this comment.
FYI Removing because the transformers upgrade added the deepseek v3.2 config: huggingface/transformers#41251
Post training also already has an upgrade to transformers as well.
3071a95 to
73cd6ad
Compare
Description
Adds Multi-tiered checkpointing support to MaxText. This will also work with Elastic Training Pause & Resume as well as Elastic Resizing.
This feature is expected to roughly increase the goodput by ~6.5% and is only dependent on Orbax.
Also pins the minimum Orbax version (0.12.1) to accommodate the necessary changes. Regenerated the TPU requirements using instructions from here.
Tests
Tested internally against candidate Orbax.
Checklist
Before submitting this PR, please make sure (put X in square brackets):
gemini-reviewlabel.