[AMD] Enable Piecewise CUDA Graph for AMD GPUs (#22299)

test/registered/amd/test_deepseek_r1_mxfp4_8gpu.py

@@ -40,6 +40,11 @@ class TestDeepseekR1MXFP4(CustomTestCase):
             "131072",
             "--model-loader-extra-config",
             '{"enable_multithread_load": true}',
+            "--enforce-piecewise-cuda-graph",
+            "--piecewise-cuda-graph-compiler",
+            "eager",
+            "--piecewise-cuda-graph-max-tokens",
+            "8192",
         ]
         cls.process = popen_launch_server(
             cls.model,
@@ -73,7 +78,7 @@ class TestDeepseekR1MXFP4(CustomTestCase):
             write_github_step_summary(
                 f"### test_gsm8k (deepseek-r1-mxfp4)\n" f'{metrics["accuracy"]=:.3f}\n'
             )
-            self.assertGreater(metrics["accuracy"], 0.94)
+        self.assertGreater(metrics["accuracy"], 0.94)
 
     def test_bs_1_speed(self):
         args = BenchArgs(port=int(self.base_url.split(":")[-1]), max_new_tokens=2048)
@@ -85,7 +90,7 @@ class TestDeepseekR1MXFP4(CustomTestCase):
             write_github_step_summary(
                 f"### test_bs_1_speed (deepseek-r1-mxfp4)\n" f"{speed=:.2f} token/s\n"
             )
-            self.assertGreater(speed, 75)
+        self.assertGreater(speed, 75)
 
 
 class TestDeepseekR1MXFP4MTP(CustomTestCase):

test/registered/piecewise_cuda_graph/test_piecewise_cuda_graph_support_1_gpu.py

@@ -5,7 +5,7 @@ import torch
 from sglang import Engine
 from sglang.lang.chat_template import get_chat_template_by_model_path
 from sglang.srt.utils import kill_process_tree
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.run_eval import run_eval
 from sglang.test.test_utils import (
     DEFAULT_IMAGE_URL,
@@ -18,6 +18,7 @@ from sglang.test.test_utils import (
 
 # CI Registration
 register_cuda_ci(est_time=180, stage="base-b", runner_config="1-gpu-large")
+register_amd_ci(est_time=180, suite="stage-b-test-1-gpu-large-amd")
 
 
 class TestPiecewiseCudaGraphQwen25VL(CustomTestCase):