chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

[AMD] CI - migrate perf test and fix stage-b-test-1-gpu-amd (#17340)

Browse Source 
Co-authored-by: Bingxu Chen <bingxche@amd.com>
Co-authored-by: bingxche <Bingxu.Chen@amd.com>
Co-authored-by: michaelzhang-ai <michaelzhang.ai@users.noreply.github.com>
This commit is contained in:
YC Tseng
2026-01-22 18:45:05 -08:00
committed by GitHub
co-authored by Bingxu Chen bingxche michaelzhang-ai
parent 0fec8820d1
commit 04a10c9bc2
14 changed files with 68 additions and 174 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/eval/test_eval_accuracy_large.py
+2 -1
View File
@@ -7,7 +7,7 @@ import unittest
from types import SimpleNamespace
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_MODEL_NAME_FOR_TEST,
@@ -20,6 +20,7 @@ from sglang.test.test_utils import (
)
register_cuda_ci(est_time=300, suite="stage-b-test-small-1-gpu-accuracy")
register_amd_ci(est_time=300, suite="stage-b-test-small-1-gpu-accuracy-amd")
class TestEvalAccuracyLarge(CustomTestCase):
test/registered/eval/test_moe_eval_accuracy_large.py
+2 -1
View File
@@ -7,7 +7,7 @@ import unittest
from types import SimpleNamespace
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_MOE_MODEL_NAME_FOR_TEST,
@@ -20,6 +20,7 @@ from sglang.test.test_utils import (
)
register_cuda_ci(est_time=500, suite="stage-b-test-large-2-gpu-accuracy")
register_amd_ci(est_time=500, suite="stage-b-test-large-2-gpu-accuracy-amd")
class TestMoEEvalAccuracyLarge(CustomTestCase):
test/registered/models/test_compressed_tensors_models.py
+6 -2
View File
@@ -3,7 +3,7 @@
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.ci.ci_register import register_amd_ci, register_cuda_ci
from sglang.test.few_shot_gsm8k import run_eval
from sglang.test.test_utils import (
@@ -45,7 +45,11 @@ class TestCompressedTensorsLlama3FP8(CustomTestCase):
)
metrics = run_eval(args)
print(f"{metrics=}")
self.assertGreaterEqual(metrics["accuracy"], 0.45)
if is_hip():
# Lower threshold for AMD because FP8 dtype differs (fp8_fnuz)
self.assertGreaterEqual(metrics["accuracy"], 0.40)
else:
self.assertGreaterEqual(metrics["accuracy"], 0.45)
if __name__ == "__main__":
test/registered/openai_server/function_call/test_tool_choice.py
+2 -1
View File
@@ -12,7 +12,7 @@ import unittest
import openai
from sglang.srt.utils import kill_process_tree
from sglang.srt.utils import is_hip, kill_process_tree
from sglang.srt.utils.hf_transformers_utils import get_tokenizer
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
from sglang.test.test_utils import (
@@ -855,6 +855,7 @@ class TestToolChoiceMistral(TestToolChoiceLlama32):
# cls.tokenizer = get_tokenizer(cls.model)
@unittest.skipIf(is_hip(), "Disabled for AMD")
class TestToolChoiceLfm2(TestToolChoiceLlama32):
"""Test tool_choice functionality with LiquidAI LFM2 model"""
test/registered/perf/test_bench_one_batch_1gpu.py
+2 -1
View File
@@ -1,6 +1,6 @@
import unittest
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.test_utils import (
DEFAULT_MODEL_NAME_FOR_TEST,
DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
@@ -12,6 +12,7 @@ from sglang.test.test_utils import (
)
register_cuda_ci(est_time=120, suite="stage-b-test-large-1-gpu-performance")
register_amd_ci(est_time=120, suite="stage-b-test-large-1-gpu-performance-amd")
class TestBenchOneBatch1GPU(CustomTestCase):
test/registered/perf/test_bench_one_batch_2gpu.py
+2 -1
View File
@@ -1,6 +1,6 @@
import unittest
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.test_utils import (
DEFAULT_MODEL_NAME_FOR_TEST,
DEFAULT_MOE_MODEL_NAME_FOR_TEST,
@@ -12,6 +12,7 @@ from sglang.test.test_utils import (
)
register_cuda_ci(est_time=180, suite="stage-b-test-large-2-gpu-performance")
register_amd_ci(est_time=630, suite="stage-b-test-large-2-gpu-performance-amd")
class TestBenchOneBatch2GPU(CustomTestCase):
test/registered/perf/test_bench_serving_1gpu_large.py
+4 -1
View File
@@ -4,7 +4,8 @@ Performance tests for single GPU that need H200 (80GB) - FP8 and EAGLE tests.
import unittest
from sglang.test.ci.ci_register import register_cuda_ci
from sglang.srt.utils import is_hip
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
from sglang.test.test_utils import (
DEFAULT_DRAFT_MODEL_EAGLE,
DEFAULT_MODEL_NAME_FOR_TEST_FP8,
@@ -17,6 +18,7 @@ from sglang.test.test_utils import (
)
register_cuda_ci(est_time=300, suite="stage-b-test-large-1-gpu-performance")
register_amd_ci(est_time=300, suite="stage-b-test-large-1-gpu-performance-amd")
class TestBenchServing1GPULarge(CustomTestCase):
@@ -38,6 +40,7 @@ class TestBenchServing1GPULarge(CustomTestCase):
else:
self.assertGreater(res["output_throughput"], 4300)
@unittest.skipIf(is_hip(), "Skip Eagle test for ROCm")
def test_online_latency_eagle(self):
res = run_bench_serving(
model=DEFAULT_TARGET_MODEL_EAGLE,
test/registered/perf/test_bench_serving_1gpu_part1.py
+2 -1
View File
@@ -9,7 +9,7 @@ import unittest
import requests
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.test_utils import (
DEFAULT_MODEL_NAME_FOR_TEST,
CustomTestCase,
@@ -20,6 +20,7 @@ from sglang.test.test_utils import (
)
register_cuda_ci(est_time=1000, suite="stage-b-test-large-1-gpu-performance")
register_amd_ci(est_time=1100, suite="stage-b-test-large-1-gpu-performance-amd")
class TestBenchServing1GPUPart1(CustomTestCase):
test/registered/perf/test_bench_serving_1gpu_part2.py
+2 -1
View File
@@ -5,7 +5,7 @@ Works on 5090 (32GB).
import unittest
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.test_utils import (
DEFAULT_SMALL_EMBEDDING_MODEL_NAME_FOR_TEST,
DEFAULT_SMALL_MODEL_NAME_FOR_TEST_SCORE,
@@ -20,6 +20,7 @@ from sglang.test.test_utils import (
)
register_cuda_ci(est_time=900, suite="stage-b-test-large-1-gpu-performance")
register_amd_ci(est_time=900, suite="stage-b-test-large-1-gpu-performance-amd")
class TestBenchServing1GPUPart2(CustomTestCase):
test/registered/perf/test_bench_serving_2gpu.py
+2 -1
View File
@@ -4,7 +4,7 @@ Performance tests for 2-GPU that need large GPUs (H200 80GB) - MoE and Pipeline
import unittest
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.test_utils import (
DEFAULT_MOE_MODEL_NAME_FOR_TEST,
CustomTestCase,
@@ -15,6 +15,7 @@ from sglang.test.test_utils import (
)
register_cuda_ci(est_time=600, suite="stage-b-test-large-2-gpu-performance")
register_amd_ci(est_time=600, suite="stage-b-test-large-2-gpu-performance-amd")
class TestBenchServing2GPU(CustomTestCase):
test/registered/perf/test_vlm_perf_5090.py
+2 -1
View File
@@ -4,7 +4,7 @@ VLM Performance tests that work on 5090 (32GB) - VLM offline throughput and onli
import unittest
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.test_utils import (
DEFAULT_SMALL_VLM_MODEL_NAME_FOR_TEST,
CustomTestCase,
@@ -14,6 +14,7 @@ from sglang.test.test_utils import (
)
register_cuda_ci(est_time=600, suite="stage-b-test-small-1-gpu-performance")
register_amd_ci(est_time=500, suite="stage-b-test-small-1-gpu-performance-amd")
class TestVLMPerf5090(CustomTestCase):
test/run_suite.py
+5
View File
@@ -23,6 +23,11 @@ PER_COMMIT_SUITES = {
"stage-b-test-small-1-gpu-amd",
"stage-b-test-small-1-gpu-amd-mi35x",
"stage-b-test-large-2-gpu-amd",
"stage-b-test-small-1-gpu-performance-amd",
"stage-b-test-large-1-gpu-performance-amd",
"stage-b-test-large-2-gpu-performance-amd",
"stage-b-test-small-1-gpu-accuracy-amd",
"stage-b-test-large-2-gpu-accuracy-amd",
"stage-c-test-large-8-gpu-amd-mi35x",
],
HWBackend.CUDA: [