[AMD] Support Hierarchical Caching on AMD GPUs (#8236)

test/srt/hicache/test_hicache.py

@@ -1,7 +1,7 @@
 import unittest
 from types import SimpleNamespace
 
-from sglang.srt.utils import kill_process_tree
+from sglang.srt.utils import is_hip, kill_process_tree
 from sglang.test.run_eval import run_eval
 from sglang.test.test_utils import (
     DEFAULT_MODEL_NAME_FOR_TEST,
@@ -11,6 +11,8 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+_is_hip = is_hip()
+
 
 class TestHiCache(CustomTestCase):
     @classmethod
@@ -26,7 +28,7 @@ class TestHiCache(CustomTestCase):
                 "--mem-fraction-static",
                 0.7,
                 "--hicache-size",
-                100,
+                100 if not _is_hip else 200,
             ],
         )
 

test/srt/hicache/test_hicache_mla.py

@@ -1,7 +1,7 @@
 import unittest
 from types import SimpleNamespace
 
-from sglang.srt.utils import kill_process_tree
+from sglang.srt.utils import is_hip, kill_process_tree
 from sglang.test.run_eval import run_eval
 from sglang.test.test_utils import (
     DEFAULT_MLA_MODEL_NAME_FOR_TEST,
@@ -11,6 +11,12 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+_is_hip = is_hip()
+if _is_hip:
+    hicache_args = ["--hicache-size", 200]
+else:
+    hicache_args = ["--hicache-ratio", 2]
+
 
 class TestHierarchicalMLA(CustomTestCase):
     @classmethod
@@ -24,9 +30,8 @@ class TestHierarchicalMLA(CustomTestCase):
             other_args=[
                 "--trust-remote-code",
                 "--enable-hierarchical-cache",
-                "--hicache-ratio",
-                2,
-            ],
+            ]
+            + hicache_args,
         )
 
     @classmethod

test/srt/hicache/test_hicache_storage.py

@@ -1,7 +1,7 @@
 import unittest
 from types import SimpleNamespace
 
-from sglang.srt.utils import kill_process_tree
+from sglang.srt.utils import is_hip, kill_process_tree
 from sglang.test.run_eval import run_eval
 from sglang.test.test_utils import (
     DEFAULT_MODEL_NAME_FOR_TEST,
@@ -11,6 +11,8 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+_is_hip = is_hip()
+
 
 class TestHiCache(CustomTestCase):
     @classmethod
@@ -26,7 +28,7 @@ class TestHiCache(CustomTestCase):
                 "--mem-fraction-static",
                 0.7,
                 "--hicache-size",
-                100,
+                100 if not _is_hip else 200,
                 "--page-size",
                 "64",
                 "--hicache-storage-backend",

test/srt/run_suite.py

@@ -162,6 +162,9 @@ suites = {
 # Add AMD tests
 suite_amd = {
     "per-commit-amd": [
+        TestFile("hicache/test_hicache.py", 116),
+        TestFile("hicache/test_hicache_mla.py", 127),
+        TestFile("hicache/test_hicache_storage.py", 127),
         TestFile("lora/test_lora.py", 200),
         TestFile("lora/test_lora_eviction.py", 200),
         TestFile("lora/test_lora_backend.py", 99),