Enable unit tests for AMD CI. (#6283)

test/srt/run_suite.py

@@ -85,6 +85,12 @@ suites = {
         TestFile("test_w8a8_quantization.py", 46),
         TestFile("models/lora/test_lora_cuda_graph.py", 250),
     ],
+    "per-commit-amd": [
+        TestFile("test_mla.py", 242),
+        TestFile("test_mla_deepseek_v3.py", 221),
+        TestFile("test_torch_compile.py", 76),
+        TestFile("test_torch_compile_moe.py", 172),
+    ],
     "per-commit-2-gpu": [
         TestFile("models/lora/test_lora_tp.py", 116),
         TestFile("test_data_parallelism.py", 73),
@@ -95,6 +101,9 @@ suites = {
         TestFile("test_update_weights_from_distributed.py", 103),
         TestFile("test_verl_engine.py", 64),
     ],
+    "per-commit-2-gpu-amd": [
+        TestFile("test_mla_tp.py", 170),
+    ],
     "per-commit-8-gpu": [
         # Disabled deepep tests temporarily because it takes too much time.
         # TODO: re-enable them after reducing the test time with compilation cache and smaller models.
@@ -107,6 +116,9 @@ suites = {
         TestFile("test_full_deepseek_v3.py", 250),
         TestFile("test_pp_single_node.py", 150),
     ],
+    "per-commit-8-gpu-amd": [
+        TestFile("test_full_deepseek_v3.py", 250),
+    ],
     "nightly": [
         TestFile("test_nightly_gsm8k_eval.py"),
     ],

test/srt/test_full_deepseek_v3.py

@@ -1,3 +1,4 @@
+import os
 import unittest
 from types import SimpleNamespace
 
@@ -66,7 +67,10 @@ class TestDeepseekV3(CustomTestCase):
             write_github_step_summary(
                 f"### test_bs_1_speed (deepseek-v3)\n" f"{speed=:.2f} token/s\n"
             )
-            self.assertGreater(speed, 75)
+            if os.getenv("SGLANG_AMD_CI") == "1":
+                self.assertGreater(speed, 12)
+            else:
+                self.assertGreater(speed, 75)
 
 
 class TestDeepseekV3MTP(CustomTestCase):
@@ -144,8 +148,14 @@ class TestDeepseekV3MTP(CustomTestCase):
                 f"{acc_length=:.2f}\n"
                 f"{speed=:.2f} token/s\n"
             )
-            self.assertGreater(acc_length, 2.9)
-            self.assertGreater(speed, 105)
+            if os.getenv("SGLANG_AMD_CI") == "1":
+                self.assertGreater(acc_length, 2.8)
+            else:
+                self.assertGreater(acc_length, 2.9)
+            if os.getenv("SGLANG_AMD_CI") == "1":
+                self.assertGreater(speed, 15)
+            else:
+                self.assertGreater(speed, 105)
 
 
 if __name__ == "__main__":