Remove support for PyTorch 2.1.

Ongoing work to integrate with torch.compile is incompatible with
PyTorch being older than 2.2.

ghstack-source-id: 0752f4dac317ef82538c0b8c988f1c95e28a37cb
Pull Request resolved: fairinternal/xformers#1086

__original_commit__ = fairinternal/xformers@fce69ab

.github/workflows/win-build.yml

@@ -49,7 +49,7 @@ jobs:
 
       - name: Install build dependencies
         run: |
-          $PY -m pip install wheel setuptools ninja torch==2.1.0 -r requirements.txt --extra-index-url https://download.pytorch.org/whl/cu121
+          $PY -m pip install wheel setuptools ninja torch==2.2.0 -r requirements.txt --extra-index-url https://download.pytorch.org/whl/cu121
           git config --global --add safe.directory "*"
           $PY -c "import torch; print('torch', torch.__version__)"
           $PY -c "import torch; print('torch.cuda', torch.version.cuda)"

CHANGELOG.md

@@ -8,6 +8,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ### Added
 ### Improved
 ### Removed
+- Removed support for PyTorch version older than 2.2.0
 
 ## [0.0.26] - 2024-04-29
 ### Added
@@ -18,7 +19,6 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - merge_attentions no longer needs inputs to be stacked.
 - fMHA: triton_splitk now supports additive bias
 - fMHA: benchmark cleanup
-### Removed
 
 ## [0.0.25.post1] - 2024-03-29
 Pre-built binary wheels require PyTorch 2.2.2

requirements.txt

@@ -1,4 +1,4 @@
 # Example requirement, can be anything that pip knows
 # install with `pip install -r requirements.txt`, and make sure that CI does the same
-torch >= 2.1
+torch >= 2.2
 numpy