[CI] Update est_time for 64 tests based on actual elapsed times (#22305)

Co-authored-by: Alison Shao <alison.shao@Mac.lan>
Co-authored-by: Alison Shao <alison.shao@MacBook-Pro-D2W773R9CD.local>

test/registered/4-gpu-models/test_nvidia_nemotron_3_super_nvfp4.py

@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=600, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=290, suite="stage-c-test-4-gpu-b200")
 
 NEMOTRON_3_SUPER_NVFP4_MODEL = "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4"
 

test/registered/4-gpu-models/test_qwen35_hicache.py

@@ -18,7 +18,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=600, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=620, suite="stage-c-test-4-gpu-h100")
 
 QWEN35_27B_MODEL = "Qwen/Qwen3.5-27B"
 ACC_THRESHOLDS = {QWEN35_27B_MODEL: {"gsm8k": 0.8}}

test/registered/4-gpu-models/test_qwen3_30b.py

@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=300, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=150, suite="stage-c-test-4-gpu-h100")
 
 QWEN3_30B_MODEL_PATH = "Qwen/Qwen3-30B-A3B-FP8"
 

test/registered/4-gpu-models/test_qwen3_next_models.py

@@ -6,7 +6,7 @@ from sglang.test.kits.kl_divergence_kit import KLDivergenceMixin
 from sglang.test.kits.prefix_cache_branching_kit import PrefixCacheBranchingMixin
 from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 
-register_cuda_ci(est_time=350, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=120, suite="stage-c-test-4-gpu-h100")
 
 QWEN3_NEXT_MODEL = "Qwen/Qwen3-Next-80B-A3B-Instruct"
 

test/registered/4-gpu-models/test_qwen3_next_models_mtp.py

@@ -7,7 +7,7 @@ from sglang.test.kits.kl_divergence_kit import KLDivergenceMixin
 from sglang.test.kits.prefix_cache_branching_kit import PrefixCacheBranchingMixin
 from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 
-register_cuda_ci(est_time=500, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=430, suite="stage-c-test-4-gpu-h100")
 
 QWEN3_NEXT_MODEL = "Qwen/Qwen3-Next-80B-A3B-Instruct"
 

test/registered/8-gpu-models/test_deepseek_v3_mtp.py

@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=275, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=240, suite="stage-c-test-8-gpu-h200")
 
 FULL_DEEPSEEK_V3_MODEL_PATH = "deepseek-ai/DeepSeek-V3-0324"
 

test/registered/8-gpu-models/test_dsa_models_basic.py

@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=720, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=870, suite="stage-c-test-8-gpu-h200")
 
 DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2"
 GLM5_MODEL_PATH = "zai-org/GLM-5-FP8"

test/registered/8-gpu-models/test_dsa_models_mtp.py

@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=720, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=880, suite="stage-c-test-8-gpu-h200")
 
 FULL_DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2"
 GLM5_MODEL_PATH = "zai-org/GLM-5-FP8"

test/registered/8-gpu-models/test_mimo_models.py

@@ -5,7 +5,7 @@ from sglang.test.kits.eval_accuracy_kit import GSM8KMixin
 from sglang.test.kits.spec_decoding_kit import SpecDecodingMixin
 from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 
-register_cuda_ci(est_time=200, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=270, suite="stage-c-test-8-gpu-h200")
 
 
 class TestMiMoV2Flash(GSM8KMixin, SpecDecodingMixin, DefaultServerBase):

test/registered/8-gpu-models/test_ring_2_5_1t.py

@@ -5,7 +5,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.run_combined_tests import run_combined_tests
 from sglang.test.test_utils import ModelLaunchSettings
 
-register_cuda_ci(est_time=1800, suite="nightly-8-gpu-common", nightly=True)
+register_cuda_ci(est_time=510, suite="nightly-8-gpu-common", nightly=True)
 
 RING_2_5_1T_MODEL_PATH = "inclusionAI/Ring-2.5-1T"
 

test/registered/attention/test_fa3.py

@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
 
 # FlashAttention3 integration tests (requires SM 90+ / H100)
 # Multiple test classes: FA3, FA3+MLA, FA3+SpecDecode variants
-register_cuda_ci(est_time=300, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=390, suite="stage-b-test-1-gpu-large")
 
 GSM_DATASET_PATH = None
 

test/registered/core/test_deterministic.py

@@ -16,7 +16,7 @@ from sglang.test.test_deterministic_utils import (
 )
 from sglang.test.test_utils import is_in_amd_ci
 
-register_cuda_ci(est_time=278, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=200, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=278, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/core/test_gpt_oss_1gpu.py

@@ -3,7 +3,7 @@ import unittest
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.gpt_oss_common import BaseTestGptOss
 
-register_cuda_ci(est_time=519, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=420, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=750, suite="stage-b-test-1-gpu-small-amd-mi35x")
 
 

test/registered/core/test_gpt_oss_sm120.py

@@ -5,7 +5,7 @@ import torch
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.gpt_oss_common import BaseTestGptOss
 
-register_cuda_ci(est_time=500, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=330, suite="stage-b-test-1-gpu-small")
 
 
 @unittest.skipIf(not torch.cuda.is_available(), "CUDA is not available")

test/registered/distributed/test_disaggregation_different_tp.py

@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=600, suite="stage-c-test-8-gpu-h20")
+register_cuda_ci(est_time=330, suite="stage-c-test-8-gpu-h20")
 
 
 class TestDisaggregationMooncakePrefillLargerTP(PDDisaggregationServerBase):

test/registered/distributed/test_disaggregation_dp_attention.py

@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=580, suite="stage-c-test-8-gpu-h20")
+register_cuda_ci(est_time=400, suite="stage-c-test-8-gpu-h20")
 
 
 class TestDisaggregationDPAttention(PDDisaggregationServerBase):

test/registered/distributed/test_disaggregation_hybrid_attention.py

@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     popen_launch_pd_server,
 )
 
-register_cuda_ci(est_time=500, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=450, suite="stage-c-test-8-gpu-h200")
 
 
 @unittest.skipIf(is_in_ci(), "Temporarily disable the flaky test.")

test/registered/distributed/test_dp_attention.py

@@ -25,7 +25,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=350, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=450, suite="stage-b-test-2-gpu-large")
 
 
 class TestDPAttentionDP2TP2(

test/registered/distributed/test_dp_attention_large.py

@@ -22,7 +22,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=350, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=230, suite="stage-c-test-4-gpu-h100")
 register_amd_ci(est_time=350, suite="stage-c-test-4-gpu-amd")
 
 

test/registered/distributed/test_pp_single_node.py

@@ -31,7 +31,7 @@ from sglang.test.test_utils import (
     run_bench_one_batch_server,
 )
 
-register_cuda_ci(est_time=650, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=460, suite="stage-c-test-4-gpu-h100")
 register_amd_ci(est_time=650, suite="stage-c-test-4-gpu-amd")
 
 

test/registered/dllm/test_llada2_mini.py

@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=181, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=110, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=330, suite="stage-b-test-1-gpu-small-amd")
 
 import unittest

test/registered/ep/test_deepep_large.py

@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=563, suite="stage-c-test-deepep-8-gpu-h200")
+register_cuda_ci(est_time=490, suite="stage-c-test-deepep-8-gpu-h200")
 
 DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2"
 

test/registered/ep/test_deepep_small.py

@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=531, suite="stage-c-test-deepep-4-gpu-h100")
+register_cuda_ci(est_time=430, suite="stage-c-test-deepep-4-gpu-h100")
 
 
 class TestPureDP(CustomTestCase):

test/registered/ep/test_mooncake_ep_small.py

@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=200, suite="stage-c-test-deepep-4-gpu-h100")
+register_cuda_ci(est_time=80, suite="stage-c-test-deepep-4-gpu-h100")
 
 ib_devices = get_rdma_devices_args()
 

test/registered/hicache/test_hicache_storage_mooncake_backend.py

@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
     is_in_ci,
 )
 
-register_cuda_ci(est_time=300, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=230, suite="stage-b-test-2-gpu-large")
 
 
 class HiCacheStorageMooncakeBackendBaseMixin(HiCacheStorageBaseMixin):

test/registered/hicache/test_hicache_storage_runtime_attach_detach.py

@@ -28,7 +28,7 @@ from sglang.test.test_utils import (
 )
 from sglang.utils import wait_for_http_ready
 
-register_cuda_ci(est_time=200, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=130, suite="stage-b-test-2-gpu-large")
 
 
 class TestHiCacheStorageRuntimeAttachDetach(CustomTestCase):

test/registered/hicache/test_hicache_variants.py

@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=524, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=410, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=524, suite="stage-b-test-1-gpu-small-amd")
 """
 Consolidated HiCache variant tests.

test/registered/mla/test_mla_int8_deepseek_v3.py

@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
 )
 
 # DeepSeek-V3 INT8 quantization tests (channel and block INT8)
-register_cuda_ci(est_time=341, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=290, suite="stage-b-test-1-gpu-large")
 
 
 class TestMLADeepseekV3ChannelInt8(CustomTestCase):

test/registered/models/test_nvidia_nemotron_3_nano.py

@@ -4,7 +4,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.kits.lm_eval_kit import LMEvalMixin
 from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 
-register_cuda_ci(est_time=660, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=540, suite="stage-b-test-2-gpu-large")
 
 NEMOTRON_3_NANO_THINKING_ARGS = [
     "--trust-remote-code",

test/registered/models/test_transformers_backend_eval.py

@@ -7,7 +7,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval
 from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 
-register_cuda_ci(est_time=180, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=40, suite="stage-b-test-1-gpu-small")
 
 
 class TestTransformersBackendEval(DefaultServerBase):

test/registered/models/test_transformers_models.py

@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=450, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=260, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=320, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/moe/test_torch_compile_moe.py

@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=210, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=120, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=1400, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/openai_server/basic/test_anthropic_server.py

@@ -31,7 +31,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=120, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=40, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=140, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/openai_server/basic/test_http2_server.py

@@ -27,7 +27,7 @@ try:
 except ImportError:
     _HAS_GRANIAN = False
 
-register_cuda_ci(est_time=120, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=50, suite="stage-b-test-1-gpu-small")
 
 
 @unittest.skipUnless(_HAS_GRANIAN, "granian not installed (pip install sglang[http2])")

test/registered/openai_server/function_call/test_anthropic_tool_use.py

@@ -25,7 +25,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=120, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=50, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=140, suite="stage-b-test-1-gpu-small-amd")
 
 # System message to guide Llama3.2 to produce proper tool call format

test/registered/perf/test_bench_serving_1gpu_part1.py

@@ -19,7 +19,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=1000, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=1140, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=1100, suite="stage-b-test-1-gpu-large-amd")
 
 

test/registered/perf/test_vlm_perf_5090.py

@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=600, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=410, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=500, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/prefill_only/test_embedding_models.py

@@ -35,7 +35,7 @@ register_amd_ci(
     suite="stage-b-test-1-gpu-small-amd",
     disabled="see https://github.com/sgl-project/sglang/issues/11127",
 )
-register_cuda_ci(est_time=200, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=90, suite="stage-b-test-1-gpu-small")
 
 MODEL_TO_CONFIG = {
     "Alibaba-NLP/gte-Qwen2-1.5B-instruct": (1, 1e-5),

test/registered/prefill_only/test_encoder_embedding_models.py

@@ -29,7 +29,7 @@ from sglang.test.test_utils import CustomTestCase, get_similarities, is_in_ci
 # python -m unittest test_encoder_embedding_models.TestEncoderEmbeddingModels.test_prefill_logits
 
 
-register_cuda_ci(est_time=270, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=310, suite="stage-b-test-1-gpu-small")
 
 MODELS = [("BAAI/bge-small-en", 1, 1e-5), ("BAAI/bge-m3", 1, 1e-5)]
 

test/registered/quant/test_awq.py

@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=700, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=260, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=200, suite="stage-b-test-1-gpu-large-amd")
 
 

test/registered/quant/test_deepseek_v32_fp4_4gpu.py

@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=1200, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=530, suite="stage-c-test-4-gpu-b200")
 
 FULL_DEEPSEEK_V3_FP4_MODEL_PATH = "nvidia/DeepSeek-V3.2-NVFP4"
 SERVER_LAUNCH_TIMEOUT = 1200

test/registered/quant/test_deepseek_v32_fp4_mtp_4gpu.py

@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=1200, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=610, suite="stage-c-test-4-gpu-b200")
 
 FULL_DEEPSEEK_V3_FP4_MODEL_PATH = "nvidia/DeepSeek-V3.2-NVFP4"
 SERVER_LAUNCH_TIMEOUT = 1200

test/registered/quant/test_deepseek_v3_fp4_4gpu.py

@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=1200, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=450, suite="stage-c-test-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

@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=420, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=400, suite="stage-c-test-4-gpu-b200")
 
 MODEL_PATH = "Qwen/Qwen3-4B-Instruct-2507-FP8"
 MXFP8_MODEL_PATH = "zianglih/Qwen3-4B-Instruct-2507-MXFP8"

test/registered/quant/test_marlin_moe.py

@@ -12,7 +12,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_marlin_utils import awq_marlin_quantize, marlin_quantize
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=200, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=130, suite="stage-b-test-1-gpu-small")
 
 set_global_server_args_for_scheduler(object.__new__(ServerArgs))
 

test/registered/quant/test_nvfp4_gemm.py

@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=360, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=300, suite="stage-c-test-4-gpu-b200")
 
 MODEL_PATH = "nvidia/Llama-3.1-8B-Instruct-NVFP4"
 

test/registered/quant/test_quantization.py

@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     write_results_to_json,
 )
 
-register_cuda_ci(est_time=370, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=270, suite="stage-b-test-1-gpu-large")
 
 MODEL_SCORE_THRESHOLDS = {
     # Baselines observed with gsm8k 5-shot concatenated format via chat API,

test/registered/quant/test_torchao.py

@@ -5,7 +5,7 @@ import requests
 from sglang import Engine
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=200, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=140, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=230, suite="stage-b-test-1-gpu-small-amd")
 from sglang.lang.chat_template import get_chat_template_by_model_path
 from sglang.srt.utils import kill_process_tree

test/registered/quant/test_w4a8_deepseek_v3.py

@@ -19,7 +19,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=520, suite="stage-c-test-8-gpu-h20")
+register_cuda_ci(est_time=700, suite="stage-c-test-8-gpu-h20")
 
 
 class TestDeepseekV3W4afp8(CustomTestCase):

test/registered/rl/test_update_weights_from_disk_mxfp8.py

@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=200, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=100, suite="stage-c-test-4-gpu-b200")
 
 import unittest
 

test/registered/sampling/test_pytorch_sampling_backend.py

@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=150, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=80, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=66, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/scheduler/test_chunked_prefill.py

@@ -7,7 +7,7 @@ import unittest
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase, run_mmlu_test, run_mulit_request_test
 
-register_cuda_ci(est_time=550, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=360, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=312, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/scheduler/test_retract_decode.py

@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
 )
 from sglang.utils import is_in_ci
 
-register_cuda_ci(est_time=550, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=470, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=600, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/spec/eagle/test_deepseek_v3_fp4_mtp_small.py

@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=900, suite="stage-b-test-4-gpu-b200")
+register_cuda_ci(est_time=240, suite="stage-b-test-4-gpu-b200")
 
 FULL_DEEPSEEK_V3_FP4_MODEL_PATH = "nvidia/DeepSeek-V3-0324-FP4"
 SERVER_LAUNCH_TIMEOUT = 1200

test/registered/spec/eagle/test_eagle3_basic.py

@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     DEFAULT_TARGET_MODEL_EAGLE3,
 )
 
-register_cuda_ci(est_time=200, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=160, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=50, suite="stage-b-test-1-gpu-small")
 
 _is_hip = is_hip()

test/registered/spec/eagle/test_eagle_dp_attention.py

@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
 )
 
 # EAGLE3 with DP attention (tp=2, dp=2, requires 4 GPUs)
-register_cuda_ci(est_time=200, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=100, suite="stage-c-test-4-gpu-h100")
 register_amd_ci(est_time=200, suite="stage-c-test-4-gpu-amd")
 
 

test/registered/spec/eagle/test_eagle_infer_a.py

@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     CustomTestCase,
 )
 
-register_cuda_ci(est_time=450, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=270, suite="stage-b-test-1-gpu-large")
 
 
 class TestEAGLEEngine(CustomTestCase):

test/registered/spec/eagle/test_eagle_infer_b.py

@@ -22,7 +22,7 @@ from sglang.test.run_eval import run_eval
 from sglang.test.server_fixtures.eagle_fixture import EagleServerBase
 from sglang.test.test_utils import DEFAULT_TARGET_MODEL_EAGLE, run_logprob_check
 
-register_cuda_ci(est_time=600, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=690, suite="stage-b-test-1-gpu-large")
 
 
 class TestEAGLEServerBasic(EagleServerBase):

test/registered/vlm/test_vision_openai_server_a.py

@@ -19,7 +19,7 @@ from sglang.test.vlm_utils import (
     VideoOpenAITestMixin,
 )
 
-register_cuda_ci(est_time=957, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=700, suite="stage-b-test-1-gpu-large")
 
 
 class TestLlavaServer(ImageOpenAITestMixin):

test/registered/vlm/test_vlm_input_format.py

@@ -37,7 +37,7 @@ from sglang.srt.entrypoints.openai.protocol import ChatCompletionRequest
 from sglang.srt.parser.conversation import generate_chat_conv
 from sglang.srt.utils.hf_transformers_utils import _fix_added_tokens_encoding
 
-register_cuda_ci(est_time=447, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=620, suite="stage-b-test-1-gpu-large")
 
 IMAGE_MAN_IRONING_URL = "https://raw.githubusercontent.com/sgl-project/sgl-test-files/refs/heads/main/images/man_ironing_on_back_of_suv.png"
 IMAGE_SGL_LOGO_URL = "https://raw.githubusercontent.com/sgl-project/sgl-test-files/refs/heads/main/images/sgl_logo.png"

test/registered/vlm/test_vlm_tp4.py

@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=200, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=140, suite="stage-c-test-4-gpu-h100")
 
 QWEN35_27B_MODEL = "Qwen/Qwen3.5-27B"
 MMMU_ACCURACY_THRESHOLD = 0.65