feat: add ARM64 self-hosted runner build support

- Update build workflow to include ARM64 self-hosted runner job.
- Modify Dockerfile and compose.yml for ARM64 runner setup.
- Add ARM64 self-hosted matrix configuration.
- Update README with ARM64 self-hosted runner instructions.

create_matrix.py

@@ -90,6 +90,34 @@ LINUX_SELF_HOSTED_MATRIX = {
     "cuda-version": ["12.8", "13.0"],
 }
 
+LINUX_ARM64_SELF_HOSTED_MATRIX = {
+    "flash-attn-version": [
+        # "2.6.3",
+        "2.7.4",
+        # "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",
+        "2.9.1",
+    ],
+    "cuda-version": [
+        "12.4",
+        # "12.6",
+        "12.8",
+        # "12.9",
+        "13.0",
+    ],
+}
+
 WINDOWS_MATRIX = {
     "flash-attn-version": [
         "2.6.3",
@@ -133,12 +161,15 @@ def main():
                 "linux": False,
                 # "linux": LINUX_MATRIX,
                 #
-                # "linux_arm64": False,
-                "linux_arm64": LINUX_ARM64_MATRIX,
+                "linux_arm64": False,
+                # "linux_arm64": LINUX_ARM64_MATRIX,
                 #
                 "linux_self_hosted": False,
                 # "linux_self_hosted": LINUX_SELF_HOSTED_MATRIX,
                 #
+                # "linux_arm64_self_hosted": False,
+                "linux_arm64_self_hosted": LINUX_ARM64_SELF_HOSTED_MATRIX,
+                #
                 "windows": False,
                 # "windows": WINDOWS_MATRIX,
                 #