[doc][c10d] fixup fsdp tutorial #1297
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
✅ Deploy Preview for pytorch-examples-preview canceled.
|
|
looks like running python example failed? |
c-p-i-o
force-pushed
the
chirag/fix-fsdp-tutorial
branch
from
2cfc3b8
to
0834097
Compare
Unrelated to my change - but I fixed it anyway. Needed to update to a newer Python version in CI. See the additional diff I made to |
c-p-i-o
force-pushed
the
chirag/fix-fsdp-tutorial
branch
2 times, most recently
from
752efec
to
0834097
Compare
fegin
reviewed
Nov 1, 2024
| @@ -65,7 +65,7 @@ def load_model_sharded(model, rank, cfg, verbose=True): | |||
| if rank == 0: | |||
| ck = checkpoint.keys() | |||
| print(f" checkpoint key len = {len(ck)} and \n keys = {ck}") | |||
|
|
|||
| dist_cp.load_state_dict( | |||
There was a problem hiding this comment.
The DCP usage is pretty outdated. Should we also update them?
There was a problem hiding this comment.
I will update this in a subsequent change - if that's ok with you?
This change is already too large as I am fixing up the python tests that broke.
The breakage is unrelated to this change.
c-p-i-o
force-pushed
the
chirag/fix-fsdp-tutorial
branch
from
e0fba21
to
7dcd080
Compare
|
@fduwjj - CI is green now.
|
c-p-i-o
force-pushed
the
chirag/fix-fsdp-tutorial
branch
from
7dcd080
to
162b0dc
Compare
|
This change will be rebased on #1299 to fix the failing Python Examples. |
Summary: Fix up the FSDP tutorial to get it functional again. 1. Add missing import for load_dataset. 2. Use `checkpoint` instead of `_shard.checkpoint` to get rid of a warning. 3. Add nlp to requirements.txt 4. Get rid of `load_metric` as this function does not exist in new `datasets` module. 5. Add `legacy=False` to get rid of tokenizer warnings. Test Plan: Ran the tutorial as follows and ensured that it ran successfully: ``` torchrun --nnodes=1 --nproc_per_node=2 T5_training.py W1031 09:46:49.166000 2847649 torch/distributed/run.py:793] W1031 09:46:49.166000 2847649 torch/distributed/run.py:793] ***************************************** W1031 09:46:49.166000 2847649 torch/distributed/run.py:793] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. W1031 09:46:49.166000 2847649 torch/distributed/run.py:793] ***************************************** dict_keys(['train', 'validation', 'test']) Size of train dataset: (157252, 3) Size of Validation dataset: (5599, 3) dict_keys(['train', 'validation', 'test']) Size of train dataset: (157252, 3) Size of Validation dataset: (5599, 3) bFloat16 enabled for mixed precision - using bfSixteen policy ```
c-p-i-o
force-pushed
the
chirag/fix-fsdp-tutorial
branch
from
162b0dc
to
cb00288
Compare
fduwjj
approved these changes
Nov 8, 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.
Summary:
Fix up the FSDP tutorial to get it functional again.
checkpointinstead of_shard.checkpointto get rid of a warning.load_metricas this function does not exist in newdatasetsmodule.legacy=Falseto get rid of tokenizer warnings.Test Plan:
Ran the tutorial as follows and ensured that it ran successfully: