minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[AMD] Replace MI325 with MI300 CI Runners (#31409)

Browse Source
This commit is contained in:
Bingxu Chen
2026-07-29 08:18:07 -07:00
committed by GitHub
parent e1f2f9d1fa
commit bfc450248e
9 changed files with 200 additions and 174 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/amd/accuracy/mi30x/test_minimax_m27_eval_amd.py
+1 -1
View File
@@ -167,7 +167,7 @@ class TestMiniMaxM27EvalAMD(unittest.TestCase):
def setUpClass(cls):
cls.models = MINIMAX_M27_MODELS
cls.base_url = DEFAULT_URL_FOR_TEST
cls.num_questions = int(os.environ.get("GSM8K_NUM_QUESTIONS", "200"))
cls.num_questions = int(os.environ.get("GSM8K_NUM_QUESTIONS", "1319"))
def test_minimax_m27_accuracy(self):
"""Test MiniMax-M2.7 with GSM8K completion benchmark."""
test/registered/perf/test_bench_serving_1gpu_part1.py
+9 -3
View File
@@ -117,7 +117,7 @@ class TestBenchServing1GPUPart1(CustomTestCase):
f"Output throughput: {res['output_throughput']:.2f} token/s\n"
)
if is_in_amd_ci():
self.assertGreater(res["output_throughput"], 3500)
self.assertGreater(res["output_throughput"], 2700)
else:
self.assertGreater(res["output_throughput"], 3700)
@@ -150,7 +150,10 @@ class TestBenchServing1GPUPart1(CustomTestCase):
f"median_e2e_latency_ms: {res['median_e2e_latency_ms']:.2f} ms\n"
f"median_ttft_ms: {res['median_ttft_ms']:.2f} ms\n"
)
self.assertLess(res["median_e2e_latency_ms"], 2400)
if is_in_amd_ci():
self.assertLess(res["median_e2e_latency_ms"], 3320)
else:
self.assertLess(res["median_e2e_latency_ms"], 2400)
# relax for mi300x (LoRA TTFT ~2x slower than mi325)
if is_in_amd_ci():
self.assertLess(res["median_ttft_ms"], 100)
@@ -166,7 +169,10 @@ class TestBenchServing1GPUPart1(CustomTestCase):
f"median_e2e_latency_ms: {res['median_e2e_latency_ms']:.2f} ms\n"
f"median_ttft_ms: {res['median_ttft_ms']:.2f} ms\n"
)
self.assertLess(res["median_e2e_latency_ms"], 4000)
if is_in_amd_ci():
self.assertLess(res["median_e2e_latency_ms"], 6000)
else:
self.assertLess(res["median_e2e_latency_ms"], 4000)
# relax for mi300x (LoRA TTFT ~2x slower than mi325)
if is_in_amd_ci():
self.assertLess(res["median_ttft_ms"], 130)