OSSForks/pytorch
1
0
Fork 0
mirror of https://github.com/zebrajr/pytorch.git synced 2025-12-06 12:20:52 +01:00
Code Issues Actions 38 Packages Projects Releases Wiki Activity

[TorchRec][PT2] disable contextlib in PT2 train pipeline (#147254)

Browse Source 
[TorchRec][PT2] disable contextlib in PT2 train pipeline (#147254)

Summary:

# context
* more details in the [post](https://fb.workplace.com/groups/1075192433118967/permalink/1587079018596970/)
* disable contextlib with PT2

Test Plan:
* run command
```
TORCH_SHOW_CPP_STACKTRACES=1 TORCHDYNAMO_EXTENDED_DEBUG_CPP=1 TORCH_LOGS="+dynamo,+graph_code,output_code,dynamic,aot,guards,verbose_guards,recompiles,graph_breaks" TORCH_TRACE=/var/tmp/tt buck2 run fbcode//mode/opt fbcode//aps_models/ads/icvr:icvr_launcher_live -- mode=fmc/local_ig_fm_ultra_mini training.pipeline_type=pt2 data_loader.dataset.table_ds=[2024-12-02] 2>&1 | tee -a output.log
```
* old tlparse
https://manifold.edge.x2p.facebook.net/v0/read/tree/logs/.tmpYYAS3o/index.html?bucketName=tlparse_reports&apiKey=tlparse_reports-key&withPayload=1&timeoutMsec=100
* new tlparse
https://manifold.edge.x2p.facebook.net/v0/read/tree/logs/.tmpUJhCGZ/index.html?bucketName=tlparse_reports&apiKey=tlparse_reports-key&withPayload=1&timeoutMsec=100

Reviewed By: Microve

Differential Revision: D68480678
This commit is contained in:
Huanyu He 2025-02-22 09:57:55 -08:00 committed by GitHub
parent fa8e3a28a7
commit 85ea679834
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
torch/nn/parallel/distributed.py
View File

@ -1444,11 +1444,11 @@ class DistributedDataParallel(Module, Joinable):
"""`TorchDynamo` requires DDP's status and module for cooperative optimization."""
return cls._active_ddp_module
@torch._disable_dynamo(recursive=True)
# note, this ctxmgr function is marked 'skip' in torchdynamo, so dynamo only kicks in
# for the 'module_to_run' underneath
# see torch._dynamo/eval_frame.py TorchPatcher.patch for more details
@contextmanager
@torch._disable_dynamo(recursive=False)
def _inside_ddp_forward(self):
DistributedDataParallel._active_ddp_module = self
try: