chore: Update docs and regenerate packages

README.md

@@ -42,6 +42,10 @@ pip install ./flash_attn-2.6.3+cu124torch2.5-cp312-cp312-linux_x86_64.whl
 
 ## Packages
 
+> [!NOTE]
+> Since v0.5.0, wheels are built with a local version label indicating the CUDA and PyTorch versions.  
+> Example: `pip list` -> `flash_attn==2.8.3 -> flash_attn==2.8.3+cu130torch2.9`
+
 See [./docs/packages.md](./docs/packages.md) for the full list of available packages.
 
 ## History

create_matrix.py

@@ -22,21 +22,20 @@ EXCLUDE = [
 ]
 
 LINUX_MATRIX = {
-    "flash-attn-version": [
+    "flash-attn-version": ["2.6.3", "2.7.4", "2.8.3"],
-        # "2.6.3", "2.7.4.post1"
+    "python-version": ["3.10", "3.11", "3.12", "3.13"],
-        "2.8.3"
-    ],
-    "python-version": [
-        # "3.10", "3.11", "3.12",
-        "3.13"
-    ],
     "torch-version": [
-        # "2.5.1", "2.6.0", "2.7.1", "2.8.0",
+        "2.5.1",
+        "2.6.0",
+        "2.7.1",
+        "2.8.0",
         "2.9.1",
     ],
     "cuda-version": [
-        # "12.4.1", "12.6.3", "12.8.1", "12.9.1",
+        "12.4.1",
-        # "12.8.1",
+        "12.6.3",
+        "12.8.1",
+        "12.9.1",
         "13.0.2",
     ],
 }
@@ -50,20 +49,29 @@ LINUX_SELF_HOSTED_MATRIX = {
 
 WINDOWS_MATRIX = {
     "flash-attn-version": [
-        # "2.7.4.post1",
+        "2.6.3",
-        "2.8.3"
+        "2.7.4",
+        "2.8.3",
     ],
     "python-version": [
-        # "3.10", "3.11", "3.12",
+        "3.10",
-        "3.13"
+        "3.11",
+        "3.12",
+        "3.13",
     ],
     "torch-version": [
-        # "2.5.1", "2.6.0", "2.7.1", "2.8.0",
+        "2.5.1",
-        "2.9.1"
+        "2.6.0",
+        "2.7.1",
+        "2.8.0",
+        "2.9.1",
     ],
     "cuda-version": [
-        # "12.4.1", "12.6.3", "12.8.1", "12.9.1",
+        "12.4.1",
-        "13.0.1"
+        "12.6.3",
+        "12.8.1",
+        "12.9.1",
+        "13.0.2",
     ],
 }
 

create_packages.py

@@ -15,6 +15,11 @@ import pandas as pd
 
 from common import normalize_platform_name, parse_wheel_filename
 
+ADD_NOTE = """> [!NOTE]
+> Since v0.5.0, wheels are built with a local version label indicating the CUDA and PyTorch versions.  
+> Example: `pip list` -> `flash_attn==2.8.3 -> flash_attn==2.8.3+cu130torch2.9`
+"""
+
 
 def parse_numeric_version(text: str) -> tuple:
     """Extract numeric version tuple for sorting."""
@@ -496,7 +501,7 @@ def main() -> None:
     output_path.parent.mkdir(parents=True, exist_ok=True)
 
     # Generate markdown with "# Packages" header for standalone file
-    standalone_markdown = f"# Packages\n\n{markdown}"
+    standalone_markdown = f"# Packages\n\n{ADD_NOTE}\n{markdown}"
 
     with output_path.open("w", encoding="utf-8") as f:
         f.write(standalone_markdown)

docs/packages.md

@@ -1,5 +1,9 @@
 # Packages
 
+> [!NOTE]
+> Since v0.5.0, wheels are built with a local version label indicating the CUDA and PyTorch versions.  
+> Example: `pip list` -> `flash_attn==2.8.3 -> flash_attn==2.8.3+cu130torch2.9`
+
 ## Table of Contents
 
 - [Linux x86_64](#linux-x86_64)
@@ -106,7 +110,7 @@
 | 3.13 | 2.8 | 12.9 | [Download1(v0.4.12)](https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.4.12/flash_attn-2.8.3%2Bcu129torch2.8-cp313-cp313-linux_x86_64.whl) |
 | 3.13 | 2.9 | 12.6 | [Download1(v0.4.17)](https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.4.17/flash_attn-2.8.3%2Bcu126torch2.9-cp313-cp313-linux_x86_64.whl), [Download2(v0.4.15)](https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.4.15/flash_attn-2.8.3%2Bcu126torch2.9-cp313-cp313-linux_x86_64.whl) |
 | 3.13 | 2.9 | 12.8 | [Download1(v0.4.17)](https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.4.17/flash_attn-2.8.3%2Bcu128torch2.9-cp313-cp313-linux_x86_64.whl), [Download2(v0.4.15)](https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.4.15/flash_attn-2.8.3%2Bcu128torch2.9-cp313-cp313-linux_x86_64.whl) |
-| 3.13 | 2.9 | 13.0 | [Download1(v0.4.18)](https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.4.18/flash_attn-2.8.3%2Bcu130torch2.9-cp313-cp313-linux_x86_64.whl) |
+| 3.13 | 2.9 | 13.0 | [Download1(v0.5.2)](https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.5.2/flash_attn-2.8.3%2Bcu130torch2.9-cp313-cp313-linux_x86_64.whl), [Download2(v0.4.18)](https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.4.18/flash_attn-2.8.3%2Bcu130torch2.9-cp313-cp313-linux_x86_64.whl) |
 
 </details>