Migrate 4-GPU/8-GPU workflow jobs to stage-c and add CI registry decorators (#17299)

.github/workflows/pr-test.yml

@@ -1298,12 +1298,12 @@ jobs:
             --total-partitions 2 \
             $CONTINUE_ON_ERROR_FLAG
 
-  unit-test-backend-4-gpu:
+  stage-c-test-4-gpu-h100:
     needs: [check-changes, call-gate, wait-for-stage-b]
     if: |
       always() &&
       (
-        (inputs.target_stage == 'unit-test-backend-4-gpu') ||
+        (inputs.target_stage == 'stage-c-test-4-gpu-h100') ||
         (
           !inputs.target_stage &&
           ((github.event_name == 'schedule' || inputs.test_parallel_dispatch == true) || (!failure() && !cancelled())) &&
@@ -1340,23 +1340,19 @@ jobs:
       - name: Run test
         timeout-minutes: 20
         run: |
-          cd test/srt
-          RETRY_FLAG=""
-          if [[ "${{ needs.check-changes.outputs.enable_retry }}" == "true" ]]; then
-            RETRY_FLAG="--enable-retry"
-          fi
+          cd test
           CONTINUE_ON_ERROR_FLAG=""
           if [[ "${{ needs.check-changes.outputs.continue_on_error }}" == "true" ]]; then
             CONTINUE_ON_ERROR_FLAG="--continue-on-error"
           fi
-          python3 run_suite.py --suite per-commit-4-gpu --auto-partition-id ${{ matrix.part }} --auto-partition-size 3 $RETRY_FLAG $CONTINUE_ON_ERROR_FLAG
+          python3 run_suite.py --hw cuda --suite stage-c-test-4-gpu-h100 --auto-partition-id ${{ matrix.part }} --auto-partition-size 3 $CONTINUE_ON_ERROR_FLAG
 
-  unit-test-backend-8-gpu-h200:
+  stage-c-test-8-gpu-h200:
     needs: [check-changes, call-gate, wait-for-stage-b]
     if: |
       always() &&
       (
-        (inputs.target_stage == 'unit-test-backend-8-gpu-h200') ||
+        (inputs.target_stage == 'stage-c-test-8-gpu-h200') ||
         (
           !inputs.target_stage &&
           ((github.event_name == 'schedule' || inputs.test_parallel_dispatch == true) || (!failure() && !cancelled())) &&
@@ -1399,23 +1395,19 @@ jobs:
       - name: Run test
         timeout-minutes: 20
         run: |
-          cd test/srt
-          RETRY_FLAG=""
-          if [[ "${{ needs.check-changes.outputs.enable_retry }}" == "true" ]]; then
-            RETRY_FLAG="--enable-retry"
-          fi
+          cd test
           CONTINUE_ON_ERROR_FLAG=""
           if [[ "${{ needs.check-changes.outputs.continue_on_error }}" == "true" ]]; then
             CONTINUE_ON_ERROR_FLAG="--continue-on-error"
           fi
-          python3 run_suite.py --suite per-commit-8-gpu-h200 --auto-partition-id ${{ matrix.part }} --auto-partition-size 4 $RETRY_FLAG $CONTINUE_ON_ERROR_FLAG
+          python3 run_suite.py --hw cuda --suite stage-c-test-8-gpu-h200 --auto-partition-id ${{ matrix.part }} --auto-partition-size 4 $CONTINUE_ON_ERROR_FLAG
 
-  unit-test-backend-8-gpu-h20:
+  stage-c-test-8-gpu-h20:
     needs: [check-changes, call-gate, wait-for-stage-b]
     if: |
       always() &&
       (
-        (inputs.target_stage == 'unit-test-backend-8-gpu-h20') ||
+        (inputs.target_stage == 'stage-c-test-8-gpu-h20') ||
         (
           !inputs.target_stage &&
           ((github.event_name == 'schedule' || inputs.test_parallel_dispatch == true) || (!failure() && !cancelled())) &&
@@ -1453,23 +1445,19 @@ jobs:
       - name: Run test
         timeout-minutes: 20
         run: |
-          cd test/srt
-          RETRY_FLAG=""
-          if [[ "${{ needs.check-changes.outputs.enable_retry }}" == "true" ]]; then
-            RETRY_FLAG="--enable-retry"
-          fi
+          cd test
           CONTINUE_ON_ERROR_FLAG=""
           if [[ "${{ needs.check-changes.outputs.continue_on_error }}" == "true" ]]; then
             CONTINUE_ON_ERROR_FLAG="--continue-on-error"
           fi
-          python3 run_suite.py --suite per-commit-8-gpu-h20 --auto-partition-id ${{ matrix.part }} --auto-partition-size 2 $RETRY_FLAG $CONTINUE_ON_ERROR_FLAG
+          python3 run_suite.py --hw cuda --suite stage-c-test-8-gpu-h20 --auto-partition-id ${{ matrix.part }} --auto-partition-size 2 $CONTINUE_ON_ERROR_FLAG
 
-  unit-test-deepep-4-gpu:
+  stage-c-test-deepep-4-gpu:
     needs: [check-changes, call-gate, wait-for-stage-b]
     if: |
       always() &&
       (
-        (inputs.target_stage == 'unit-test-deepep-4-gpu') ||
+        (inputs.target_stage == 'stage-c-test-deepep-4-gpu') ||
         (
           !inputs.target_stage &&
           ((github.event_name == 'schedule' || inputs.test_parallel_dispatch == true) || (!failure() && !cancelled())) &&
@@ -1502,23 +1490,19 @@ jobs:
       - name: Run test
         timeout-minutes: 20
         run: |
-          cd test/srt
-          RETRY_FLAG=""
-          if [[ "${{ needs.check-changes.outputs.enable_retry }}" == "true" ]]; then
-            RETRY_FLAG="--enable-retry"
-          fi
+          cd test
           CONTINUE_ON_ERROR_FLAG=""
           if [[ "${{ needs.check-changes.outputs.continue_on_error }}" == "true" ]]; then
             CONTINUE_ON_ERROR_FLAG="--continue-on-error"
           fi
-          python3 run_suite.py --suite per-commit-4-gpu-deepep $RETRY_FLAG $CONTINUE_ON_ERROR_FLAG
+          python3 run_suite.py --hw cuda --suite stage-c-test-deepep-4-gpu $CONTINUE_ON_ERROR_FLAG
 
-  unit-test-deepep-8-gpu:
+  stage-c-test-deepep-8-gpu-h200:
     needs: [check-changes, call-gate, wait-for-stage-b]
     if: |
       always() &&
       (
-        (inputs.target_stage == 'unit-test-deepep-8-gpu') ||
+        (inputs.target_stage == 'stage-c-test-deepep-8-gpu-h200') ||
         (
           !inputs.target_stage &&
           ((github.event_name == 'schedule' || inputs.test_parallel_dispatch == true) || (!failure() && !cancelled())) &&
@@ -1551,23 +1535,19 @@ jobs:
       - name: Run test
         timeout-minutes: 45
         run: |
-          cd test/srt
-          RETRY_FLAG=""
-          if [[ "${{ needs.check-changes.outputs.enable_retry }}" == "true" ]]; then
-            RETRY_FLAG="--enable-retry"
-          fi
+          cd test
           CONTINUE_ON_ERROR_FLAG=""
           if [[ "${{ needs.check-changes.outputs.continue_on_error }}" == "true" ]]; then
             CONTINUE_ON_ERROR_FLAG="--continue-on-error"
           fi
-          python3 run_suite.py --suite per-commit-8-gpu-h200-deepep $RETRY_FLAG $CONTINUE_ON_ERROR_FLAG
+          python3 run_suite.py --hw cuda --suite stage-c-test-deepep-8-gpu-h200 $CONTINUE_ON_ERROR_FLAG
 
-  unit-test-backend-4-gpu-b200:
+  stage-c-test-4-gpu-b200:
     needs: [check-changes, call-gate, wait-for-stage-b]
     if: |
       always() &&
       (
-        (inputs.target_stage == 'unit-test-backend-4-gpu-b200') ||
+        (inputs.target_stage == 'stage-c-test-4-gpu-b200') ||
         (
           !inputs.target_stage &&
           ((github.event_name == 'schedule' || inputs.test_parallel_dispatch == true) || (!failure() && !cancelled())) &&
@@ -1605,23 +1585,19 @@ jobs:
       - name: Run test
         timeout-minutes: 30
         run: |
-          cd test/srt
-          RETRY_FLAG=""
-          if [[ "${{ needs.check-changes.outputs.enable_retry }}" == "true" ]]; then
-            RETRY_FLAG="--enable-retry"
-          fi
+          cd test
           CONTINUE_ON_ERROR_FLAG=""
           if [[ "${{ needs.check-changes.outputs.continue_on_error }}" == "true" ]]; then
             CONTINUE_ON_ERROR_FLAG="--continue-on-error"
           fi
-          IS_BLACKWELL=1 python3 run_suite.py --suite per-commit-4-gpu-b200 --auto-partition-id ${{ matrix.part }} --auto-partition-size 3 --timeout-per-file 1800 $RETRY_FLAG $CONTINUE_ON_ERROR_FLAG
+          IS_BLACKWELL=1 python3 run_suite.py --hw cuda --suite stage-c-test-4-gpu-b200 --auto-partition-id ${{ matrix.part }} --auto-partition-size 3 --timeout-per-file 1800 $CONTINUE_ON_ERROR_FLAG
 
-  unit-test-backend-4-gpu-gb200:
+  stage-c-test-4-gpu-gb200:
     needs: [check-changes, call-gate, wait-for-stage-b, sgl-kernel-build-wheels-arm]
     if: |
       always() &&
       (
-        (inputs.target_stage == 'unit-test-backend-4-gpu-gb200') ||
+        (inputs.target_stage == 'stage-c-test-4-gpu-gb200') ||
         (
           !inputs.target_stage &&
           ((github.event_name == 'schedule' || inputs.test_parallel_dispatch == true) || (!failure() && !cancelled())) &&
@@ -1656,16 +1632,12 @@ jobs:
       - name: Run test
         timeout-minutes: 45
         run: |
-          cd test/srt
-          RETRY_FLAG=""
-          if [[ "${{ needs.check-changes.outputs.enable_retry }}" == "true" ]]; then
-            RETRY_FLAG="--enable-retry"
-          fi
+          cd test
           CONTINUE_ON_ERROR_FLAG=""
           if [[ "${{ needs.check-changes.outputs.continue_on_error }}" == "true" ]]; then
             CONTINUE_ON_ERROR_FLAG="--continue-on-error"
           fi
-          python3 run_suite.py --suite per-commit-4-gpu-gb200 --auto-partition-id 0 --auto-partition-size 1 --timeout-per-file 3600 $RETRY_FLAG $CONTINUE_ON_ERROR_FLAG
+          python3 run_suite.py --hw cuda --suite stage-c-test-4-gpu-gb200 --timeout-per-file 3600 $CONTINUE_ON_ERROR_FLAG
 
   pr-test-finish:
     needs:
@@ -1694,13 +1666,13 @@ jobs:
         stage-b-test-large-2-gpu,
         stage-c-test-large-4-gpu,
         stage-b-test-4-gpu-b200,
-        unit-test-backend-4-gpu,
-        unit-test-backend-8-gpu-h20,
-        unit-test-backend-8-gpu-h200,
-        unit-test-deepep-4-gpu,
-        unit-test-deepep-8-gpu,
-        unit-test-backend-4-gpu-b200,
-        unit-test-backend-4-gpu-gb200,
+        stage-c-test-4-gpu-h100,
+        stage-c-test-8-gpu-h20,
+        stage-c-test-8-gpu-h200,
+        stage-c-test-deepep-4-gpu,
+        stage-c-test-deepep-8-gpu-h200,
+        stage-c-test-4-gpu-b200,
+        stage-c-test-4-gpu-gb200,
       ]
     if: always()
     runs-on: ubuntu-latest

scripts/ci/utils/slash_command_handler.py

@@ -258,6 +258,13 @@ def handle_rerun_stage(
         "stage-b-test-large-2-gpu",
         "stage-c-test-large-4-gpu",
         "stage-c-test-large-4-gpu-b200",
+        "stage-c-test-4-gpu-h100",
+        "stage-c-test-8-gpu-h200",
+        "stage-c-test-8-gpu-h20",
+        "stage-c-test-4-gpu-b200",
+        "stage-c-test-4-gpu-gb200",
+        "stage-c-test-deepep-4-gpu",
+        "stage-c-test-deepep-8-gpu-h200",
         "multimodal-gen-test-1-gpu",
         "multimodal-gen-test-2-gpu",
         "quantization-test",

test/registered/4-gpu-models/test_deepseek_v3_cutedsl_4gpu.py

@@ -3,6 +3,7 @@ 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.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_DEEPSEEK_NVFP4_MODEL_FOR_TEST,
@@ -13,6 +14,8 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
+register_cuda_ci(est_time=1800, suite="stage-c-test-4-gpu-gb200")
+
 
 class TestDeepseekR1Nvfp4CuteDSLDeepEP(CustomTestCase):
     @classmethod

test/registered/4-gpu-models/test_gpt_oss_4gpu.py

@@ -1,7 +1,11 @@
 import unittest
 
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.gpt_oss_common import BaseTestGptOss
 
+register_cuda_ci(est_time=300, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=300, suite="stage-c-test-4-gpu-b200")
+
 
 class TestGptOss4Gpu(BaseTestGptOss):
     def test_bf16_120b(self):

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

@@ -4,6 +4,7 @@ from types import SimpleNamespace
 import requests
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval
 from sglang.test.kl_test_utils import (
     test_input_output_logprobs_match_decode_cache_hit_helper,
@@ -16,6 +17,8 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_cuda_ci(est_time=350, suite="stage-c-test-4-gpu-h100")
+
 QWEN3_NEXT_MODEL = "Qwen/Qwen3-Next-80B-A3B-Instruct"
 
 ACC_THRESHOLDS = {

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

@@ -4,6 +4,7 @@ from types import SimpleNamespace
 import requests
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval
 from sglang.test.kl_test_utils import (
     test_input_output_logprobs_match_decode_cache_hit_helper,
@@ -16,6 +17,8 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_cuda_ci(est_time=500, suite="stage-c-test-4-gpu-h100")
+
 QWEN3_NEXT_MODEL = "Qwen/Qwen3-Next-80B-A3B-Instruct"
 
 ACC_THRESHOLDS = {

test/registered/8-gpu-models/test_deepseek_v32_basic.py

@@ -2,6 +2,7 @@ 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.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.send_one import BenchArgs, send_one_prompt
 from sglang.test.test_utils import (
@@ -13,6 +14,8 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
+register_cuda_ci(est_time=360, suite="stage-c-test-8-gpu-h200")
+
 DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2-Exp"
 
 

test/registered/8-gpu-models/test_deepseek_v32_mtp.py

@@ -4,6 +4,7 @@ from types import SimpleNamespace
 import requests
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.send_one import BenchArgs, send_one_prompt
 from sglang.test.test_utils import (
@@ -15,6 +16,8 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
+register_cuda_ci(est_time=360, suite="stage-c-test-8-gpu-h200")
+
 FULL_DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2-Exp"
 
 

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

@@ -2,6 +2,7 @@ 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.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.send_one import BenchArgs, send_one_prompt
 from sglang.test.test_utils import (
@@ -14,6 +15,8 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
+register_cuda_ci(est_time=275, suite="stage-c-test-8-gpu-h200")
+
 FULL_DEEPSEEK_V3_MODEL_PATH = "deepseek-ai/DeepSeek-V3-0324"
 
 

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

@@ -4,6 +4,7 @@ from types import SimpleNamespace
 import requests
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.send_one import BenchArgs, send_one_prompt
 from sglang.test.test_utils import (
@@ -16,6 +17,8 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
+register_cuda_ci(est_time=275, suite="stage-c-test-8-gpu-h200")
+
 FULL_DEEPSEEK_V3_MODEL_PATH = "deepseek-ai/DeepSeek-V3-0324"
 
 

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

@@ -1,9 +1,12 @@
 import unittest
 
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.kits.gsm8k_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")
+
 
 class TestMiMoV2Flash(GSM8KMixin, SpecDecodingMixin, DefaultServerBase):
     gsm8k_accuracy_thres = 0.75

test/registered/amd/test_moriep_small.py

@@ -3,6 +3,7 @@ import unittest
 from types import SimpleNamespace
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_amd_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_DEEPEP_MODEL_NAME_FOR_TEST,
@@ -13,6 +14,8 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_amd_ci(est_time=1200, suite="stage-c-test-large-8-gpu-amd")
+
 
 class TestPureDP(CustomTestCase):
 

test/registered/distributed/test_disaggregation_aarch64.py

@@ -2,6 +2,7 @@ import os
 import unittest
 from types import SimpleNamespace
 
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.server_fixtures.disaggregation_fixture import (
     PDDisaggregationServerBase,
@@ -12,6 +13,8 @@ from sglang.test.test_utils import (
     popen_launch_pd_server,
 )
 
+register_cuda_ci(est_time=300, suite="stage-c-test-4-gpu-gb200")
+
 
 class TestDisaggregationMooncakeAARCH64Accuracy(PDDisaggregationServerBase):
     @classmethod

test/registered/distributed/test_disaggregation_different_tp.py

@@ -2,6 +2,7 @@ import unittest
 from types import SimpleNamespace
 
 from sglang.srt.environ import envs
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.server_fixtures.disaggregation_fixture import (
     PDDisaggregationServerBase,
@@ -14,6 +15,8 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
+register_cuda_ci(est_time=600, suite="stage-c-test-8-gpu-h20")
+
 
 class TestDisaggregationMooncakePrefillLargerTP(PDDisaggregationServerBase):
     @classmethod

test/registered/distributed/test_disaggregation_dp_attention.py

@@ -2,6 +2,7 @@ import unittest
 from types import SimpleNamespace
 
 from sglang.srt.environ import envs
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.server_fixtures.disaggregation_fixture import (
     PDDisaggregationServerBase,
@@ -13,6 +14,8 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
+register_cuda_ci(est_time=155, suite="stage-c-test-8-gpu-h20")
+
 
 class TestDisaggregationDPAttention(PDDisaggregationServerBase):
     PREFILL_DP_SIZE = 4

test/registered/distributed/test_disaggregation_hybrid_attention.py

@@ -1,6 +1,7 @@
 import unittest
 from types import SimpleNamespace
 
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.server_fixtures.disaggregation_fixture import (
     PDDisaggregationServerBase,
@@ -11,6 +12,10 @@ from sglang.test.test_utils import (
     popen_launch_pd_server,
 )
 
+register_cuda_ci(
+    est_time=400, suite="stage-c-test-8-gpu-h200", disabled="TCP fallback flaky"
+)
+
 
 @unittest.skipIf(is_in_ci(), "Temporarily disable the flaky test.")
 class TestDisaggregationHybridAttentionMamba(PDDisaggregationServerBase):

test/registered/distributed/test_disaggregation_pp.py

@@ -2,6 +2,7 @@ import time
 import unittest
 from types import SimpleNamespace
 
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval
 from sglang.test.server_fixtures.disaggregation_fixture import (
     PDDisaggregationServerBase,
@@ -13,6 +14,8 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
+register_cuda_ci(est_time=180, suite="stage-c-test-8-gpu-h20")
+
 
 class TestDisaggregationPrefillPPAccuracy(PDDisaggregationServerBase):
     @classmethod

test/registered/distributed/test_epd_disaggregation.py

@@ -3,6 +3,7 @@ import threading
 import unittest
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.kits.mmmu_vlm_kit import _run_lmms_eval_with_retry
 from sglang.test.server_fixtures.disaggregation_fixture import (
     PDDisaggregationServerBase,
@@ -14,6 +15,8 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_cuda_ci(est_time=150, suite="stage-c-test-4-gpu-h100")
+
 
 @unittest.skipIf(is_in_ci(), "Skipping in CI to reduce multi-GPU runtime")
 class TestEPDDisaggregationOneEncoder(PDDisaggregationServerBase):

test/registered/distributed/test_pp_single_node.py

@@ -15,6 +15,7 @@ import requests
 from sglang.bench_one_batch_server import BenchArgs as OneBatchBenchArgs
 from sglang.srt.server_args import ServerArgs
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.run_eval import run_eval
 from sglang.test.test_utils import (
@@ -30,6 +31,8 @@ from sglang.test.test_utils import (
     run_bench_one_batch_server,
 )
 
+register_cuda_ci(est_time=500, suite="stage-c-test-4-gpu-h100")
+
 
 class TestPPAccuracy(unittest.TestCase):
     @classmethod

test/registered/ep/test_deepep_large.py

@@ -4,6 +4,7 @@ from types import SimpleNamespace
 import requests
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.send_one import BenchArgs, send_one_prompt
 from sglang.test.test_utils import (
@@ -14,6 +15,8 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_cuda_ci(est_time=563, suite="stage-c-test-deepep-8-gpu-h200")
+
 DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2-Exp"
 
 

test/registered/ep/test_deepep_small.py

@@ -5,6 +5,7 @@ from types import SimpleNamespace
 import requests
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_MODEL_NAME_FOR_TEST_MLA,
@@ -15,6 +16,8 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_cuda_ci(est_time=531, suite="stage-c-test-deepep-4-gpu")
+
 
 class TestPureDP(CustomTestCase):
     @classmethod

test/registered/ep/test_mooncake_ep_small.py

@@ -3,6 +3,7 @@ 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.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.server_fixtures.disaggregation_fixture import get_rdma_devices_args
 from sglang.test.test_utils import (
@@ -14,6 +15,8 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_cuda_ci(est_time=660, suite="stage-c-test-deepep-4-gpu")
+
 ib_devices = get_rdma_devices_args()
 
 

test/registered/layers/test_fla_layernorm_guard.py

@@ -11,6 +11,13 @@ from sglang.srt.layers.attention.fla.layernorm_gated import (
     _layer_norm_fwd as layer_norm_fwd,
 )
 from sglang.srt.layers.attention.fla.layernorm_gated import layernorm_fn, rms_norm_ref
+from sglang.test.ci.ci_register import register_cuda_ci
+
+register_cuda_ci(
+    est_time=60,
+    suite="stage-b-test-large-2-gpu",
+    disabled="Temporarily disabled",
+)
 
 # Optional dependency in sglang repo; skip collection cleanly if absent.
 custom_all_reduce_utils = pytest.importorskip(

test/registered/models/test_dummy_grok_models.py

@@ -1,7 +1,14 @@
 import unittest
 
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase, is_in_ci, run_bench_one_batch
 
+register_cuda_ci(
+    est_time=120,
+    suite="stage-b-test-large-2-gpu",
+    disabled="Temporarily disabled",
+)
+
 
 class TestDummyGrok1(CustomTestCase):
 

test/registered/models/test_ministral3_models.py

@@ -1,10 +1,17 @@
 import unittest
 
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.kits.gsm8k_accuracy_kit import GSM8KMixin
 from sglang.test.kits.mmmu_vlm_kit import MMMUMixin
 from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 from sglang.test.server_fixtures.mmmu_fixture import MMMUServerBase
 
+register_cuda_ci(
+    est_time=200,
+    suite="stage-b-test-small-1-gpu",
+    disabled="Temporarily disabled",
+)
+
 MODEL = "mistralai/Ministral-3-3B-Instruct-2512"
 
 

test/registered/models/test_qwen3_next_models_pcg.py

@@ -10,6 +10,7 @@ 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.few_shot_gsm8k import run_eval
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -18,6 +19,12 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_cuda_ci(
+    est_time=400,
+    suite="stage-c-test-4-gpu-h100",
+    disabled="Intermittent failures, see #17039",
+)
+
 QWEN3_NEXT_MODEL = "Qwen/Qwen3-Next-80B-A3B-Instruct"
 
 ACC_THRESHOLDS = {

test/registered/profiling/test_profile_v2.py

@@ -8,6 +8,7 @@ import requests
 
 from sglang.srt.environ import envs
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import (
     DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -16,6 +17,12 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_cuda_ci(
+    est_time=120,
+    suite="stage-b-test-small-1-gpu",
+    disabled="Temporarily disabled",
+)
+
 
 class TestStartProfile(CustomTestCase):
 

test/registered/quant/test_deepseek_v32_fp4_4gpu.py

@@ -2,6 +2,7 @@ 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.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_URL_FOR_TEST,
@@ -11,6 +12,8 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
+register_cuda_ci(est_time=600, 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

@@ -3,6 +3,7 @@ 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.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.send_one import BenchArgs, send_one_prompt
 from sglang.test.test_utils import (
@@ -13,6 +14,8 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
+register_cuda_ci(est_time=1500, 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

@@ -3,6 +3,7 @@ from types import SimpleNamespace
 from urllib.parse import urlparse
 
 from sglang.srt.utils import get_device_sm, kill_process_tree
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -11,6 +12,8 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
+register_cuda_ci(est_time=280, suite="stage-c-test-4-gpu-b200")
+
 MODEL_PATH = "Qwen/Qwen3-4B-Instruct-2507-FP8"
 
 

test/registered/quant/test_nvfp4_gemm.py

@@ -3,6 +3,7 @@ from types import SimpleNamespace
 from urllib.parse import urlparse
 
 from sglang.srt.utils import get_device_sm, kill_process_tree
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -11,6 +12,8 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
+register_cuda_ci(est_time=360, suite="stage-c-test-4-gpu-b200")
+
 MODEL_PATH = "nvidia/Llama-3.1-8B-Instruct-NVFP4"
 
 

test/registered/quant/test_w4a8_deepseek_v3.py

@@ -5,6 +5,7 @@ from types import SimpleNamespace
 import requests
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_DEEPSEEK_W4AFP8_MODEL_FOR_TEST,
@@ -18,6 +19,8 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
+register_cuda_ci(est_time=520, suite="stage-c-test-8-gpu-h20")
+
 
 class TestDeepseekV3W4afp8(CustomTestCase):
     @classmethod

test/registered/rl/test_multi_instance_release_memory_occupation.py

@@ -12,6 +12,7 @@ from torch.distributed.device_mesh import init_device_mesh
 from transformers import AutoModelForCausalLM
 
 from sglang.srt.entrypoints.engine import Engine as SglangEngine
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import (
     DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
     DEFAULT_SMALL_MODEL_NAME_FOR_TEST_BASE,
@@ -19,6 +20,8 @@ from sglang.test.test_utils import (
     find_available_port,
 )
 
+register_cuda_ci(est_time=64, suite="stage-c-test-4-gpu-h100")
+
 TEST_SUITE = dict(
     model_path=DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
     mem_fraction_static=0.83,

test/registered/rl/test_release_memory_occupation.py

@@ -23,6 +23,8 @@ ensuring correct weight updates and text generation.
 3. Tensor Parallel Tests: Tests memory release and resume operations with different tensor parallel
 configurations (tp=1, tp=2) to ensure proper memory management in distributed settings. For different
 data parallel size, we test it in verl.
+
+NOTE: This test is temporarily disabled.
 """
 
 import os
@@ -38,6 +40,7 @@ from sglang.srt.constants import (
     GPU_MEMORY_TYPE_KV_CACHE,
     GPU_MEMORY_TYPE_WEIGHTS,
 )
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import (
     DEFAULT_HYBRID_MAMBA_MODEL_NAME_FOR_TEST,
     DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
@@ -47,6 +50,12 @@ from sglang.test.test_utils import (
     CustomTestCase,
 )
 
+register_cuda_ci(
+    est_time=200,
+    suite="stage-c-test-4-gpu-h100",
+    disabled="Temporarily disabled - needs investigation",
+)
+
 # (temporarily) set to true to observe memory usage in nvidia-smi more clearly
 _DEBUG_EXTRA = False
 

test/registered/scheduler/test_prefill_delayer.py

@@ -16,6 +16,7 @@ import torch.multiprocessing as mp
 from sglang.bench_serving import run_benchmark
 from sglang.srt.managers.prefill_delayer import PrefillDelayer
 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_MLA_MODEL_NAME_FOR_TEST,
@@ -26,6 +27,12 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_cuda_ci(
+    est_time=300,
+    suite="stage-c-test-8-gpu-h200",
+    disabled="Temporarily disabled",
+)
+
 WORLD_SIZE = os.environ.get("SGLANG_TEST_WORLD_SIZE", "8")
 
 # ============================ Unit Tests ============================

test/run_suite.py

@@ -32,8 +32,14 @@ PER_COMMIT_SUITES = {
         "stage-b-test-large-1-gpu",
         "stage-b-test-large-2-gpu",
         "stage-c-test-large-4-gpu",
-        "stage-b-test-4-gpu-b200",
-        "stage-c-test-large-4-gpu-b200",
+        "stage-c-test-4-gpu-h100",
+        "stage-c-test-4-gpu-b200",
+        "stage-c-test-4-gpu-gb200",
+        "stage-c-test-deepep-4-gpu",
+        "stage-c-test-8-gpu-h20",
+        "stage-c-test-8-gpu-h200",
+        "stage-c-test-8-gpu-b200",
+        "stage-c-test-deepep-8-gpu-h200",
     ],
     HWBackend.NPU: [],
 }

test/srt/models/test_kimi_k2_models.py

@@ -1,70 +0,0 @@
-import unittest
-from types import SimpleNamespace
-
-import requests
-
-from sglang.srt.utils import kill_process_tree
-from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
-from sglang.test.test_utils import (
-    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
-    DEFAULT_URL_FOR_TEST,
-    CustomTestCase,
-    is_in_ci,
-    popen_launch_server,
-    write_github_step_summary,
-)
-
-
-class TestKimiK2Thinking(CustomTestCase):
-    @classmethod
-    def setUpClass(cls):
-        cls.model = "moonshotai/Kimi-K2-Thinking"
-        cls.base_url = DEFAULT_URL_FOR_TEST
-        other_args = [
-            "--tp",
-            "8",
-            "--trust-remote-code",
-            "--tool-call-parser",
-            "kimi_k2",
-            "--reasoning-parser",
-            "kimi_k2",
-            "--model-loader-extra-config",
-            '{"enable_multithread_load": true, "num_threads": 64}',
-        ]
-        cls.process = popen_launch_server(
-            cls.model,
-            cls.base_url,
-            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
-            other_args=other_args,
-        )
-
-    @classmethod
-    def tearDownClass(cls):
-        kill_process_tree(cls.process.pid)
-
-    def test_a_gsm8k(
-        self,
-    ):
-        requests.get(self.base_url + "/flush_cache")
-
-        args = SimpleNamespace(
-            num_shots=5,
-            data_path=None,
-            num_questions=200,
-            max_new_tokens=512,
-            parallel=128,
-            host="http://127.0.0.1",
-            port=int(self.base_url.split(":")[-1]),
-        )
-        metrics = run_eval_few_shot_gsm8k(args)
-        print(f"{metrics=}")
-
-        if is_in_ci():
-            write_github_step_summary(
-                f"### test_gsm8k (Kimi-K2-Thinking)\n" f'{metrics["accuracy"]=:.3f}\n'
-            )
-            self.assertGreater(metrics["accuracy"], 0.95)
-
-
-if __name__ == "__main__":
-    unittest.main()

test/srt/run_suite.py

@@ -7,65 +7,13 @@ import tabulate
 from sglang.test.ci.ci_utils import TestFile, run_unittest_files
 
 # NOTE: please sort the test cases alphabetically by the test file name
+# NOTE: per-commit-4-gpu, per-commit-8-gpu-h200, per-commit-8-gpu-h20, per-commit-4-gpu-b200,
+# per-commit-4-gpu-gb200, per-commit-4-gpu-deepep, and per-commit-8-gpu-h200-deepep suites
+# have been migrated to stage-c suites in test/registered/ using the CI registry system.
 suites = {
-    "per-commit-4-gpu": [
-        TestFile("models/test_qwen3_next_models.py", 350),
-        TestFile("models/test_qwen3_next_models_mtp.py", 500),
-        TestFile("test_gpt_oss_4gpu.py", 300),
-        TestFile("test_multi_instance_release_memory_occupation.py", 64),
-        TestFile("test_pp_single_node.py", 500),
-        TestFile("test_epd_disaggregation.py", 150),
-    ],
-    "per-commit-8-gpu-h200": [
-        TestFile("test_deepseek_v3_basic.py", 275),
-        TestFile("test_deepseek_v3_mtp.py", 275),
-        TestFile("test_disaggregation_hybrid_attention.py", 400),
-        TestFile("models/test_kimi_k2_models.py", 200),
-        TestFile("test_deepseek_v32_basic.py", 360),
-        TestFile("test_deepseek_v32_mtp.py", 360),
-        TestFile("models/test_mimo_models.py", 200),
-    ],
-    "per-commit-8-gpu-h20": [
-        TestFile("quant/test_w4a8_deepseek_v3.py", 520),
-        TestFile("test_disaggregation_different_tp.py", 600),
-        TestFile("test_disaggregation_pp.py", 180),
-        TestFile("test_disaggregation_dp_attention.py", 155),
-    ],
-    "per-commit-4-gpu-b200": [
-        TestFile("test_deepseek_v3_fp4_4gpu.py", 1500),
-        TestFile("test_fp8_blockwise_gemm.py", 280),
-        TestFile("test_gpt_oss_4gpu.py", 300),
-        TestFile("test_nvfp4_gemm.py", 360),
-        TestFile("test_deepseek_v32_fp4_4gpu.py", 600),
-    ],
-    # "per-commit-8-gpu-b200": [
-    #     TestFile("test_mistral_large3_basic.py", 275),  # Moved to nightly - large model
-    # ],
-    "per-commit-4-gpu-gb200": [
-        TestFile("test_deepseek_v3_cutedsl_4gpu.py", 1800),
-        TestFile("test_disaggregation_aarch64.py", 300),
-    ],
-    "per-commit-4-gpu-deepep": [
-        TestFile("ep/test_deepep_small.py", 531),
-        TestFile("ep/test_mooncake_ep_small.py", 660),
-    ],
-    "per-commit-8-gpu-h200-deepep": [
-        TestFile("ep/test_deepep_large.py", 563),
-    ],
     # quantization_test suite migrated to test/registered/quant/
-    "__not_in_ci__": [
-        TestFile("test_release_memory_occupation.py", 200),  # Temporarily disabled
-        TestFile("models/test_dummy_grok_models.py"),
-        TestFile("test_profile_v2.py"),
-        TestFile("models/test_ministral3_models.py"),
-        TestFile("test_mistral_large3_basic.py"),
-        TestFile("test_prefill_delayer.py"),
-        TestFile("test_fla_layernorm_guard.py"),
-        TestFile(
-            "models/test_qwen3_next_models_pcg.py"
-        ),  # Disabled: intermittent failures, see #17039
-        TestFile("ep/test_moriep_small.py"),
-    ],
+    # All CUDA tests migrated to test/registered/
+    "__not_in_ci__": [],
 }
 
 # Add AMD tests
@@ -89,9 +37,8 @@ suite_amd = {
         # TestFile("test_wave_attention_backend.py", 150), # Disabled temporarily, see https://github.com/sgl-project/sglang/issues/11127
         # The time estimation for `test_int4fp8_moe.py` assumes `mistralai/Mixtral-8x7B-Instruct-v0.1` is already cached (running on 1xMI300X).
     ],
-    "per-commit-4-gpu-amd": [
-        TestFile("test_pp_single_node.py", 150),
-    ],
+    # per-commit-4-gpu-amd migrated to test/registered/distributed/ using the CI registry system
+    "per-commit-4-gpu-amd": [],
     # NOTE: AMD nightly suites (nightly-amd, nightly-amd-vlm, nightly-amd-8-gpu)
     # have been migrated to test/registered/amd/nightly/ and are now managed
     # by test/run_suite.py using the registry system.

test/srt/test_mistral_large3_basic.py

@@ -1,89 +0,0 @@
-import os
-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.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
-from sglang.test.send_one import BenchArgs, send_one_prompt
-from sglang.test.test_utils import (
-    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
-    DEFAULT_URL_FOR_TEST,
-    CustomTestCase,
-    is_in_ci,
-    popen_launch_server,
-    write_github_step_summary,
-)
-
-register_cuda_ci(est_time=600, suite="nightly-8-gpu-b200", nightly=True)
-MISTRAL_LARGE3_MODEL_PATH = "mistralai/Mistral-Large-3-675B-Instruct-2512"
-
-
-class TestMistralLarge3Basic(CustomTestCase):
-    @classmethod
-    def setUpClass(cls):
-        # Set environment variable to disable JIT DeepGemm
-        os.environ["SGLANG_ENABLE_JIT_DEEPGEMM"] = "0"
-
-        cls.model = MISTRAL_LARGE3_MODEL_PATH
-        cls.base_url = DEFAULT_URL_FOR_TEST
-        other_args = [
-            "--tp",
-            "8",
-            "--attention-backend",
-            "trtllm_mla",
-            "--model-loader-extra-config",
-            '{"enable_multithread_load": true}',
-            "--chat-template",
-            "mistral",
-        ]
-        cls.process = popen_launch_server(
-            cls.model,
-            cls.base_url,
-            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH * 5,
-            other_args=other_args,
-        )
-
-    @classmethod
-    def tearDownClass(cls):
-        kill_process_tree(cls.process.pid)
-        # Clean up environment variable
-        if "SGLANG_ENABLE_JIT_DEEPGEMM" in os.environ:
-            del os.environ["SGLANG_ENABLE_JIT_DEEPGEMM"]
-
-    def test_a_gsm8k(
-        self,
-    ):  # Append an "a" to make this test run first (alphabetically) to warm up the server
-        args = SimpleNamespace(
-            num_shots=8,
-            data_path=None,
-            num_questions=1400,
-            parallel=1400,
-            max_new_tokens=512,
-            host="http://127.0.0.1",
-            port=int(self.base_url.split(":")[-1]),
-        )
-        metrics = run_eval_few_shot_gsm8k(args)
-        print(f"{metrics=}")
-
-        if is_in_ci():
-            write_github_step_summary(
-                f"### test_gsm8k (mistral-large-3)\n" f'{metrics["accuracy"]=:.3f}\n'
-            )
-            self.assertGreater(metrics["accuracy"], 0.90)
-
-    def test_bs_1_speed(self):
-        args = BenchArgs(port=int(self.base_url.split(":")[-1]), max_new_tokens=2048)
-        acc_length, speed = send_one_prompt(args)
-
-        print(f"{speed=:.2f}")
-
-        if is_in_ci():
-            write_github_step_summary(
-                f"### test_bs_1_speed (mistral-large-3)\n" f"{speed=:.2f} token/s\n"
-            )
-            self.assertGreater(speed, 50)
-
-
-if __name__ == "__main__":
-    unittest.main()