[AMD CI] Add moonshotai/Kimi-K2-Instruct-0905 testcases (#17656)

Author: sogalin

.github/workflows/pr-test-amd.yml

@@ -657,7 +657,7 @@ jobs:
       fail-fast: false
       matrix:
         runner: [linux-mi35x-gpu-8]
-        part: [0, 1]
+        part: [0, 1, 2]
     runs-on: ${{matrix.runner}}
     steps:
       - name: Checkout code
@@ -679,7 +679,7 @@ jobs:
       - name: Run test
         timeout-minutes: 60
         run: |
-          bash scripts/ci/amd/amd_ci_exec.sh -w "/sglang-checkout/test" python3 run_suite.py --hw amd --suite stage-c-test-large-8-gpu-amd-mi35x --auto-partition-id ${{ matrix.part }} --auto-partition-size 2 --timeout-per-file 3600
+          bash scripts/ci/amd/amd_ci_exec.sh -w "/sglang-checkout/test" python3 run_suite.py --hw amd --suite stage-c-test-large-8-gpu-amd-mi35x --auto-partition-id ${{ matrix.part }} --auto-partition-size 3 --timeout-per-file 3600
 
   stage-b-test-small-1-gpu-performance-amd:
     needs: [check-changes, call-gate, stage-a-test-1-amd]

test/registered/amd/test_kimi_k2_instruct.py

@@ -0,0 +1,95 @@
+import os
+import unittest
+from types import SimpleNamespace
+
+import requests
+
+from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_amd_ci
+from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
+from sglang.test.send_one import BenchArgs, send_one_prompt
+from sglang.test.test_utils import (
+    DEFAULT_URL_FOR_TEST,
+    CustomTestCase,
+    is_in_ci,
+    popen_launch_server,
+    write_github_step_summary,
+)
+
+register_amd_ci(est_time=3600, suite="stage-c-test-large-8-gpu-amd-mi35x")
+
+KIMI_K2_MODEL_PATH = "moonshotai/Kimi-K2-Instruct-0905"
+SERVER_LAUNCH_TIMEOUT = 3600
+
+
+class TestKimiK2Instruct0905(CustomTestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.model = KIMI_K2_MODEL_PATH
+        cls.base_url = DEFAULT_URL_FOR_TEST
+        other_args = [
+            "--tp",
+            "8",
+            "--decode-attention-backend",
+            "triton",
+            "--prefill-attention-backend",
+            "aiter",
+            "--trust-remote-code",
+            "--model-loader-extra-config",
+            '{"enable_multithread_load": true}',
+        ]
+        env = os.environ.copy()
+        env["SGLANG_USE_AITER"] = "1"
+        env["SGLANG_ROCM_FUSED_DECODE_MLA"] = "0"
+        cls.process = popen_launch_server(
+            cls.model,
+            cls.base_url,
+            timeout=SERVER_LAUNCH_TIMEOUT,
+            other_args=other_args,
+            env=env,
+        )
+
+    @classmethod
+    def tearDownClass(cls):
+        kill_process_tree(cls.process.pid)
+
+    def test_a_gsm8k(
+        self,
+    ):  # Append an "a" to make this test run first (alphabetically) to warm up the server
+        requests.get(self.base_url + "/flush_cache")
+
+        args = SimpleNamespace(
+            num_shots=8,
+            data_path=None,
+            num_questions=1319,
+            parallel=1319,
+            max_new_tokens=512,
+            host="http://127.0.0.1",
+            port=int(self.base_url.split(":")[-1]),
+        )
+        metrics = run_eval_few_shot_gsm8k(args)
+        print(f"{metrics=}")
+
+        if is_in_ci():
+            write_github_step_summary(
+                f"### test_gsm8k (Kimi-K2-Instruct-0905)\n"
+                f'{metrics["accuracy"]=:.3f}\n'
+            )
+            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)
+        _, speed = send_one_prompt(args)
+
+        print(f"{speed=:.2f}")
+
+        if is_in_ci():
+            write_github_step_summary(
+                f"### test_bs_1_speed (Kimi-K2-Instruct-0905)\n"
+                f"{speed=:.2f} token/s\n"
+            )
+            self.assertGreater(speed, 45)
+
+
+if __name__ == "__main__":
+    unittest.main()