Fix PyTorch CUDACachingAllocator::snapshot() API compatibility for conda builds #65
+14
−1
Conversation
This file contains hidden or 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
…nda builds
Summary:
## Problem
The torchcomms build in the xlformers_llama4_flagship_conda environment is failing with:
```
error: too many arguments to function 'c10::cuda::CUDACachingAllocator::SnapshotInfo c10::cuda::CUDACachingAllocator::snapshot()'
```
D86907322 introduced a new method *registerMemPreHook()* that calls ```c10::cuda::CUDACachingAllocator::snapshot({device, 0})``` with arguments.
This is caused by a **PyTorch API breaking change** in `c10::cuda::CUDACachingAllocator::snapshot()`:
- **Old API** (internal builds): `snapshot({device, pool_id})`
- **New API** (conda builds): `snapshot()` - no arguments
## Root Cause
Torchcomms code was written for the older PyTorch API where `snapshot()` accepted arguments, but conda environments use a newer PyTorch version with the updated API.
## Solution
Added conditional compilation to handle both API versions:
1. **Defined `TORCHCOMMS_CONDA_BUILD` macro** in CMakeLists.txt files for conda builds
2. **Updated affected C++ files** to use the correct API based on build type:
- `comms/torchcomms/ncclx/TorchCommNCCLXCCA.cpp`
- `comms/torchcomms/nccl/TorchCommNCCLCCA.cpp`
This follows the same pattern used in `ProcessGroupNCCLX.cpp` with the `NCCLX_CONDA_BUILD` macro.
Differential Revision: D87413411
meta-cla
bot
added
the
CLA Signed
|
@MittalMakwana has exported this pull request. If you are a Meta employee, you can view the originating Diff in D87413411. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.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:
Problem
The torchcomms build in the xlformers_llama4_flagship_conda environment is failing with:
D86907322 introduced a new method registerMemPreHook() that calls
c10::cuda::CUDACachingAllocator::snapshot({device, 0})with arguments.This is caused by a PyTorch API breaking change in
c10::cuda::CUDACachingAllocator::snapshot():snapshot({device, pool_id})snapshot()- no argumentsRoot Cause
Torchcomms code was written for the older PyTorch API where
snapshot()accepted arguments, but conda environments use a newer PyTorch version with the updated API.Solution
Added conditional compilation to handle both API versions:
TORCHCOMMS_CONDA_BUILDmacro in CMakeLists.txt files for conda buildscomms/torchcomms/ncclx/TorchCommNCCLXCCA.cppcomms/torchcomms/nccl/TorchCommNCCLCCA.cppThis follows the same pattern used in
ProcessGroupNCCLX.cppwith theNCCLX_CONDA_BUILDmacro.Differential Revision: D87413411