[CI] Compile with torch 2.3.0.dev20240207

.github/workflows/publish.yml

@@ -44,7 +44,7 @@ jobs:
           # manylinux docker image, but I haven't figured out how to install CUDA on manylinux.
           os: [ubuntu-20.04]
           python-version: ['3.7', '3.8', '3.9', '3.10', '3.11']
-          torch-version: ['1.12.1', '1.13.1', '2.0.1', '2.1.2', '2.2.0', '2.3.0.dev20240105']
+          torch-version: ['1.12.1', '1.13.1', '2.0.1', '2.1.2', '2.2.0', '2.3.0.dev20240207']
           cuda-version: ['11.8.0', '12.2.2']
           # We need separate wheels that either uses C++11 ABI (-D_GLIBCXX_USE_CXX11_ABI) or not.
           # Pytorch wheels currently don't use it, but nvcr images have Pytorch compiled with C++11 ABI.
@@ -63,7 +63,7 @@ jobs:
               python-version: '3.7'
             - torch-version: '2.2.0'
               python-version: '3.7'
-            - torch-version: '2.3.0.dev20240105'
+            - torch-version: '2.3.0.dev20240207'
               python-version: '3.7'
             # Pytorch <= 2.0 only supports CUDA <= 11.8
             - torch-version: '1.12.1'

setup.py

@@ -200,6 +200,11 @@ if not SKIP_CUDA_BUILD:
                         # "--ptxas-options=-v",
                         # "--ptxas-options=-O2",
                         # "-lineinfo",
+                        # "-DFLASHATTENTION_DISABLE_BACKWARD",
+                        # "-DFLASHATTENTION_DISABLE_DROPOUT",
+                        # "-DFLASHATTENTION_DISABLE_ALIBI",
+                        # "-DFLASHATTENTION_DISABLE_UNEVEN_K",
+                        # "-DFLASHATTENTION_DISABLE_LOCAL",
                     ]
                     + generator_flag
                     + cc_flag