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

ci: B200 conditional split + LPT_SLOP removal (stage-c partition 8→3) (#25203)

Browse Source
This commit is contained in:
Alison Shao
2026-05-13 18:40:45 -07:00
committed by GitHub
parent 5fb6bde6c0
commit 1e308aec66
19 changed files with 33 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/4-gpu-models/test_gpt_oss_4gpu.py
+1 -1
View File
@@ -4,7 +4,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.gpt_oss_common import BaseTestGptOss
register_cuda_ci(est_time=392, stage="stage-c", runner_config="4-gpu-h100")
register_cuda_ci(est_time=740, stage="stage-c", runner_config="4-gpu-b200")
register_cuda_ci(est_time=350, stage="stage-c", runner_config="4-gpu-b200")
class TestGptOss4Gpu(BaseTestGptOss):
test/registered/4-gpu-models/test_nvidia_nemotron_3_super_nvfp4.py
+1 -1
View File
@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=710, stage="stage-c", runner_config="4-gpu-b200")
register_cuda_ci(est_time=540, suite="nightly-4-gpu-b200", nightly=True)
NEMOTRON_3_SUPER_NVFP4_MODEL = "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4"
test/registered/4-gpu-models/test_qwen35_fp4_mtp_v2.py
+1 -1
View File
@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=540, stage="stage-c", runner_config="4-gpu-b200")
register_cuda_ci(est_time=340, stage="stage-c", runner_config="4-gpu-b200")
QWEN35_FP4_MODEL = "nvidia/Qwen3.5-397B-A17B-NVFP4"
ACC_THRESHOLDS = {QWEN35_FP4_MODEL: {"gsm8k": 0.95}}
test/registered/4-gpu-models/test_qwen35_models.py
+1 -1
View File
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=260, stage="stage-c", runner_config="4-gpu-b200")
register_cuda_ci(est_time=370, stage="stage-c", runner_config="4-gpu-b200")
QWEN35_FP4_MODEL = "nvidia/Qwen3.5-397B-A17B-NVFP4"
ACC_THRESHOLDS = {QWEN35_FP4_MODEL: {"gsm8k": 0.95}}
test/registered/attention/test_flash_attention_4.py
+1 -1
View File
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
)
# FlashAttention4 integration test (requires SM 100+ / Blackwell B200)
register_cuda_ci(est_time=265, stage="stage-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=260, stage="stage-b", runner_config="4-gpu-b200")
@unittest.skipIf(get_device_sm() < 100, "Test requires CUDA SM 100 or higher")
test/registered/lora/test_lora_gpt_oss_20b_logprob_diff.py
+3 -3
View File
@@ -35,9 +35,9 @@ from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(
est_time=300,
stage="stage-c",
runner_config="4-gpu-b200",
est_time=90,
suite="nightly-4-gpu-b200",
nightly=True,
)
BASE_MODEL = "lmsys/gpt-oss-20b-bf16"
test/registered/lora/test_lora_nemotron_3_super_120b_a12b_logprob_diff.py
+3 -3
View File
@@ -35,9 +35,9 @@ from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(
est_time=300,
stage="stage-c",
runner_config="4-gpu-b200",
est_time=100,
suite="nightly-4-gpu-b200",
nightly=True,
)
BASE_MODEL = "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16"
test/registered/lora/test_lora_qwen3_30b_a3b_instruct_2507_logprob_diff.py
+3 -3
View File
@@ -35,9 +35,9 @@ from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(
est_time=160,
stage="stage-c",
runner_config="4-gpu-b200",
est_time=100,
suite="nightly-4-gpu-b200",
nightly=True,
)
BASE_MODEL = "Qwen/Qwen3-30B-A3B-Instruct-2507"
test/registered/lora/test_lora_qwen3_5_35b_a3b_logprob_diff.py
+1 -1
View File
@@ -35,7 +35,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(
est_time=160,
est_time=110,
stage="stage-c",
runner_config="4-gpu-b200",
)
test/registered/lora/test_lora_qwen3_vl_30b_a3b_instruct_logprob_diff.py
+1 -1
View File
@@ -35,7 +35,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(
est_time=160,
est_time=110,
stage="stage-c",
runner_config="4-gpu-b200",
)
test/registered/moe/test_cutedsl_moe.py
+1 -1
View File
@@ -16,7 +16,7 @@ except ImportError:
CuteDslMoEWrapper = None
convert_sf_to_mma_layout = None
register_cuda_ci(est_time=590, stage="stage-c", runner_config="4-gpu-b200")
register_cuda_ci(est_time=24, suite="nightly-4-gpu-b200", nightly=True)
SKIP_TEST = torch.cuda.get_device_capability() < (10, 0)
SKIP_REASON = "Nvfp4 Requires compute capability of 10 or above."
test/registered/quant/test_deepseek_v32_fp4_mtp_4gpu.py
+1 -1
View File
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
)
register_cuda_ci(
est_time=1060,
est_time=690,
stage="stage-c",
runner_config="4-gpu-b200",
)
test/registered/quant/test_deepseek_v3_fp4_4gpu.py
+1 -1
View File
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
write_github_step_summary,
)
register_cuda_ci(est_time=1190, stage="stage-c", runner_config="4-gpu-b200")
register_cuda_ci(est_time=960, stage="stage-c", runner_config="4-gpu-b200")
FULL_DEEPSEEK_V3_FP4_MODEL_PATH = "nvidia/DeepSeek-V3-0324-FP4"
SERVER_LAUNCH_TIMEOUT = 1200
test/registered/quant/test_fp8_blockwise_gemm.py
+1 -1
View File
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
try_cached_model,
)
register_cuda_ci(est_time=630, stage="stage-c", runner_config="4-gpu-b200")
register_cuda_ci(est_time=430, suite="nightly-4-gpu-b200", nightly=True)
MODEL_PATH = "Qwen/Qwen3-4B-Instruct-2507-FP8"
MXFP8_MODEL_PATH = "zianglih/Qwen3-4B-Instruct-2507-MXFP8"
test/registered/quant/test_nvfp4_gemm.py
+1 -1
View File
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
try_cached_model,
)
register_cuda_ci(est_time=420, stage="stage-c", runner_config="4-gpu-b200")
register_cuda_ci(est_time=350, stage="stage-c", runner_config="4-gpu-b200")
MODEL_PATH = "nvidia/Llama-3.1-8B-Instruct-NVFP4"
test/registered/rl/test_update_weights_from_disk_blackwell.py
+1 -1
View File
@@ -1,6 +1,6 @@
from sglang.test.ci.ci_register import register_cuda_ci
register_cuda_ci(est_time=400, stage="stage-c", runner_config="4-gpu-b200")
register_cuda_ci(est_time=320, suite="nightly-4-gpu-b200", nightly=True)
import unittest
test/registered/spec/eagle/test_deepseek_v3_fp4_mtp_small.py
+1 -1
View File
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
write_github_step_summary,
)
register_cuda_ci(est_time=420, stage="stage-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=340, stage="stage-b", runner_config="4-gpu-b200")
FULL_DEEPSEEK_V3_FP4_MODEL_PATH = "nvidia/DeepSeek-V3-0324-FP4"
SERVER_LAUNCH_TIMEOUT = 1200
test/registered/spec/eagle/test_eagle_infer_beta_dp_attention.py
+1 -1
View File
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
)
# EAGLE with DP attention on B200 (tp=2, dp=2, requires 4 B200 GPUs)
register_cuda_ci(est_time=123, stage="stage-c", runner_config="4-gpu-b200")
register_cuda_ci(est_time=90, stage="stage-c", runner_config="4-gpu-b200")
def test_gsm8k(base_url: str, model: str):