[CI] slash handler: lookup runs_on from runner_configs.yml (#25394)

test/manual/4-gpu-models/test_qwen35_fp4_triton.py

@@ -1,7 +1,6 @@
 import unittest
 
 from sglang.test.accuracy_test_runner import AccuracyTestParams
-from sglang.test.ci.ci_register import register_cuda_ci
 
 # This eval harness applies the chat_template, which is critical for qwen3.5
 # to get good accuracy on gsm8k
@@ -11,8 +10,6 @@ from sglang.test.test_utils import (
     ModelLaunchSettings,
 )
 
-register_cuda_ci(est_time=720, suite="stage-c-test-4-gpu-b200")
-
 QWEN35_FP4_MODEL = "nvidia/Qwen3.5-397B-A17B-NVFP4"
 ACC_THRESHOLDS = {QWEN35_FP4_MODEL: {"gsm8k": 0.95}}
 

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

@@ -1,13 +1,10 @@
 import unittest
 
-from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.kits.eval_accuracy_kit import GSM8KMixin
 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=142, suite="stage-c-test-4-gpu-h100")
-
 QWEN3_NEXT_MODEL = "Qwen/Qwen3-Next-80B-A3B-Instruct"
 
 

test/manual/8-gpu-models/test_deepseek_v3_basic.py

@@ -2,7 +2,6 @@ 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.run_eval import run_eval
 from sglang.test.send_one import BenchArgs, send_one_prompt
 from sglang.test.test_utils import (
@@ -15,8 +14,6 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=301, suite="stage-c-test-8-gpu-h200")
-
 FULL_DEEPSEEK_V3_MODEL_PATH = "deepseek-ai/DeepSeek-V3-0324"
 
 

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

@@ -2,7 +2,6 @@ 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.run_eval import run_eval
 from sglang.test.send_one import BenchArgs, send_one_prompt
 from sglang.test.test_utils import (
@@ -14,8 +13,6 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=1047, 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/manual/attention/test_fa3.py

@@ -5,7 +5,6 @@ import requests
 
 from sglang.srt.environ import envs
 from sglang.srt.utils import get_device_sm, kill_process_tree
-from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.run_eval import run_eval
 from sglang.test.test_utils import (
     DEFAULT_DRAFT_MODEL_EAGLE3,
@@ -20,8 +19,6 @@ 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=551, suite="stage-b-test-1-gpu-large")
-
 GSM_DATASET_PATH = None
 
 # In case of some machine lack internet connection, we can set OFFLINE_MODE to True.

test/manual/attention/test_local_attn.py

@@ -5,7 +5,6 @@ from types import SimpleNamespace
 import requests
 
 from sglang.srt.utils import get_device_sm, kill_process_tree
-from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.run_eval import run_eval
 from sglang.test.test_utils import (
     DEFAULT_MODEL_NAME_FOR_TEST_LOCAL_ATTENTION,
@@ -15,10 +14,8 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+
 # Local attention with FA3 (requires SM 90+ / H100, tp=4)
-register_cuda_ci(est_time=217, suite="stage-c-test-4-gpu-h100")
-
-
 @unittest.skipIf(get_device_sm() < 90, "Test requires CUDA SM 90 or higher")
 class TestFlashAttention3LocalAttn(CustomTestCase):
     model = DEFAULT_MODEL_NAME_FOR_TEST_LOCAL_ATTENTION

test/manual/core/test_gpt_oss_1gpu.py

@@ -1,11 +1,7 @@
 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=408, suite="stage-b-test-1-gpu-large")
-register_amd_ci(est_time=750, suite="stage-b-test-1-gpu-small-amd-mi35x")
-
 
 class TestGptOss1Gpu(BaseTestGptOss):
     def test_mxfp4_20b(self):

test/manual/distributed/test_dp_attention_large.py

@@ -5,7 +5,6 @@ import requests
 
 from sglang.lang.chat_template import get_chat_template_by_model_path
 from sglang.srt.utils import kill_process_tree
-from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.kits.ebnf_constrained_kit import EBNFConstrainedMixin
 from sglang.test.kits.json_constrained_kit import JSONConstrainedMixin
 from sglang.test.kits.regex_constrained_kit import RegexConstrainedMixin
@@ -22,9 +21,6 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=245, suite="stage-c-test-4-gpu-h100")
-register_amd_ci(est_time=350, suite="stage-c-test-4-gpu-amd")
-
 
 @unittest.skipIf(
     is_in_amd_ci(),

test/manual/eval/test_eval_accuracy_large.py

@@ -6,7 +6,6 @@ python -m unittest test_eval_accuracy_large.TestEvalAccuracyLarge.test_mmlu
 import unittest
 
 from sglang.srt.utils import kill_process_tree
-from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.kits.eval_accuracy_kit import HumanEvalMixin, MGSMEnMixin, MMLUMixin
 from sglang.test.test_utils import (
     DEFAULT_MODEL_NAME_FOR_TEST,
@@ -16,9 +15,6 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=496, suite="stage-b-test-1-gpu-small")
-register_amd_ci(est_time=420, suite="stage-b-test-1-gpu-small-amd")
-
 
 class TestEvalAccuracyLarge(CustomTestCase, MMLUMixin, HumanEvalMixin, MGSMEnMixin):
     mmlu_score_threshold = 0.70

test/manual/lora/test_lora_backend.py

@@ -17,7 +17,6 @@ import os
 import unittest
 from typing import List
 
-from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.lora_utils import (
     ALL_OTHER_LORA_MODELS,
     BACKENDS,
@@ -29,13 +28,6 @@ from sglang.test.lora_utils import (
 )
 from sglang.test.test_utils import CustomTestCase, is_in_ci
 
-register_cuda_ci(est_time=224, suite="stage-b-test-1-gpu-small")
-register_amd_ci(
-    est_time=200,
-    suite="stage-b-test-1-gpu-small-amd",
-    disabled="see https://github.com/sgl-project/sglang/issues/13107",
-)
-
 
 class TestLoRABackend(CustomTestCase):
 

test/manual/mla/test_mla.py

@@ -1,7 +1,6 @@
 import unittest
 
 from sglang.srt.utils import kill_process_tree
-from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.kits.eval_accuracy_kit import MGSMEnMixin
 from sglang.test.test_utils import (
     DEFAULT_MLA_MODEL_NAME_FOR_TEST,
@@ -11,11 +10,8 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+
 # MLA attention test with MGSM evaluation
-register_cuda_ci(est_time=181, suite="stage-b-test-1-gpu-large")
-register_amd_ci(est_time=1100, suite="stage-b-test-1-gpu-small-amd")
-
-
 class TestMLA(CustomTestCase, MGSMEnMixin):
     mgsm_en_score_threshold = 0.8
 

test/manual/mla/test_mla_deepseek_v3.py

@@ -5,7 +5,6 @@ from types import SimpleNamespace
 import requests
 
 from sglang.srt.utils import is_cuda, is_hip, kill_process_tree
-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_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -15,14 +14,6 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-# DeepSeek-V3 MLA tests with torch compile, FA3, and MTP speculative decoding
-register_cuda_ci(est_time=543, suite="stage-b-test-1-gpu-large")
-register_amd_ci(
-    est_time=221,
-    suite="stage-b-test-1-gpu-small-amd",
-    disabled="see https://github.com/sgl-project/sglang/issues/12574",
-)
-
 
 class TestMLADeepseekV3(CustomTestCase):
     @classmethod

test/manual/models/test_nvidia_nemotron_nano_v2.py

@@ -1,12 +1,9 @@
 import unittest
 
 from sglang.srt.utils import is_blackwell
-from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.kits.eval_accuracy_kit import GSM8KMixin
 from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 
-register_cuda_ci(est_time=249, suite="stage-b-test-2-gpu-large")
-
 
 class TestNvidiaNemotronNanoV2BF16(GSM8KMixin, DefaultServerBase):
     model = "nvidia/NVIDIA-Nemotron-Nano-9B-v2"

test/manual/models/test_nvidia_nemotron_nano_v2_vl.py

@@ -1,6 +1,5 @@
 import unittest
 
-from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.kits.eval_accuracy_kit import GSM8KMixin
 from sglang.test.kits.mmmu_vlm_kit import MMMUMixin
 from sglang.test.server_fixtures.default_fixture import DefaultServerBase
@@ -10,8 +9,6 @@ from sglang.test.server_fixtures.mmmu_fixture import MMMUServerBase
 # GSM8k + MMMU evaluation
 
 
-register_cuda_ci(est_time=256, suite="stage-b-test-1-gpu-large")
-
 MODEL = "nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16"
 
 

test/manual/models/test_qwen_models.py

@@ -4,7 +4,6 @@ import unittest
 from types import SimpleNamespace
 
 from sglang.srt.utils import kill_process_tree
-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_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -13,9 +12,6 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=108, suite="stage-b-test-1-gpu-small")
-register_amd_ci(est_time=130, suite="stage-b-test-1-gpu-small-amd")
-
 
 class TestQwen2(CustomTestCase):
     @classmethod

test/manual/openai_server/function_call/test_tool_choice.py

@@ -14,7 +14,6 @@ import openai
 
 from sglang.srt.utils import 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 (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
     DEFAULT_URL_FOR_TEST,
@@ -22,9 +21,6 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=204, stage="stage-b", runner_config="1-gpu-small")
-register_amd_ci(est_time=258, suite="stage-b-test-1-gpu-small-amd")
-
 
 class TestToolChoiceLlama32(CustomTestCase):
 

test/manual/perf/test_bench_one_batch_1gpu.py

@@ -5,7 +5,6 @@ import unittest
 
 import numpy as np
 
-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,
@@ -16,9 +15,6 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=95, suite="stage-b-test-1-gpu-large")
-register_amd_ci(est_time=120, suite="stage-b-test-1-gpu-large-amd")
-
 
 class TestBenchOneBatch1GPU(CustomTestCase):
 

test/manual/prefill_only/test_cross_encoder_models.py

@@ -4,16 +4,12 @@ import unittest
 
 import torch
 
-from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.runners import TEST_RERANK_QUERY_DOCS, HFRunner, SRTRunner
 from sglang.test.test_utils import CustomTestCase, is_in_ci
 
 # Cross encoder model tests
 
 
-register_cuda_ci(est_time=125, suite="stage-b-test-1-gpu-small")
-register_amd_ci(est_time=150, suite="stage-b-test-1-gpu-small-amd")
-
 MODELS = [
     ("cross-encoder/ms-marco-MiniLM-L6-v2", 1, 1e-2),
     ("BAAI/bge-reranker-v2-m3", 1, 1e-2),

test/manual/prefill_only/test_encoder_embedding_models.py

@@ -6,7 +6,6 @@ import unittest
 import torch
 from transformers import AutoConfig, AutoTokenizer
 
-from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.runners import DEFAULT_PROMPTS, HFRunner, SRTRunner
 from sglang.test.test_utils import CustomTestCase, get_similarities, is_in_ci
 
@@ -29,8 +28,6 @@ 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=444, suite="stage-b-test-1-gpu-small")
-
 MODELS = [("BAAI/bge-small-en", 1, 1e-5), ("BAAI/bge-m3", 1, 1e-5)]
 
 ATTENTION_BACKEND = ["torch_native", "triton", "flashinfer"]

test/manual/quant/test_autoround.py

@@ -7,7 +7,6 @@ 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.run_eval import run_eval
 from sglang.test.test_utils import (
     DEFAULT_AUTOROUND_MODEL_NAME_FOR_TEST,
@@ -17,8 +16,6 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=99, suite="stage-b-test-1-gpu-large")
-
 
 class TestAutoRound(CustomTestCase):
     @classmethod

test/manual/quant/test_deepseek_v32_fp4_4gpu.py

@@ -2,7 +2,6 @@ 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.run_eval import run_eval
 from sglang.test.send_one import BenchArgs, send_one_prompt
 from sglang.test.test_utils import (
@@ -13,8 +12,6 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=874, suite="stage-c-test-4-gpu-b200")
-
 FULL_DEEPSEEK_V3_FP4_MODEL_PATH = "nvidia/DeepSeek-V3.2-NVFP4"
 SERVER_LAUNCH_TIMEOUT = 1200
 

test/manual/quant/test_eval_fp8_accuracy.py

@@ -2,7 +2,6 @@ import unittest
 from types import SimpleNamespace
 
 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.run_eval import run_eval
 from sglang.test.test_utils import (
     DEFAULT_MODEL_NAME_FOR_ACCURACY_TEST_FP8,
@@ -14,9 +13,6 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=351, suite="stage-b-test-1-gpu-large")
-register_amd_ci(est_time=600, suite="stage-b-test-1-gpu-small-amd")
-
 
 class TestEvalFP8Accuracy(CustomTestCase):
     @classmethod

test/manual/quant/test_quantization.py

@@ -4,7 +4,6 @@ import warnings
 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.run_eval import run_eval
 from sglang.test.test_utils import (
     DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_QUANT_TP1,
@@ -16,8 +15,6 @@ from sglang.test.test_utils import (
     write_results_to_json,
 )
 
-register_cuda_ci(est_time=460, suite="stage-b-test-1-gpu-large")
-
 MODEL_SCORE_THRESHOLDS = {
     # Baselines observed with gsm8k 5-shot concatenated format via chat API,
     # which scores lower than reported benchmarks using proper CoT format.

test/manual/scheduler/test_no_chunked_prefill.py

@@ -1,6 +1,5 @@
 import unittest
 
-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,
@@ -8,9 +7,6 @@ from sglang.test.test_utils import (
     run_mmlu_test,
 )
 
-register_cuda_ci(est_time=131, suite="stage-b-test-1-gpu-large")
-register_amd_ci(est_time=108, suite="stage-b-test-1-gpu-small-amd")
-
 
 class TestNoChunkedPrefill(CustomTestCase):
 

test/manual/scheduler/test_no_overlap_scheduler.py

@@ -6,12 +6,8 @@ python3 test_overlap_schedule.py
 
 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
 
-register_cuda_ci(est_time=267, suite="stage-b-test-1-gpu-large")
-register_amd_ci(est_time=275, suite="stage-b-test-1-gpu-small-amd")
-
 
 class TestOverlapSchedule(CustomTestCase):
     def test_no_radix_attention_chunked_prefill(self):

test/manual/spec/eagle/test_eagle3_basic.py

@@ -4,7 +4,6 @@ from types import SimpleNamespace
 import requests
 
 from sglang.srt.utils import is_hip
-from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.run_eval import run_eval
 from sglang.test.server_fixtures.eagle_fixture import EagleServerBase
 from sglang.test.test_utils import (
@@ -12,9 +11,6 @@ from sglang.test.test_utils import (
     DEFAULT_TARGET_MODEL_EAGLE3,
 )
 
-register_cuda_ci(est_time=88, suite="stage-b-test-1-gpu-small")
-register_amd_ci(est_time=50, suite="stage-b-test-1-gpu-small")
-
 _is_hip = is_hip()