Fix gpt_oss_common import path and migrate core tests (#16426)

test/registered/core/test_deterministic.py

@@ -9,11 +9,14 @@ test into unit tests so that's easily reproducible in CI.
 
 import unittest
 
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_deterministic_utils import (
     COMMON_SERVER_ARGS,
     TestDeterministicBase,
 )
 
+register_cuda_ci(est_time=228, suite="stage-b-test-small-1-gpu")
+
 
 class TestFlashinferDeterministic(TestDeterministicBase):
     # Test with flashinfer attention backend

test/registered/core/test_gpt_oss_1gpu.py

@@ -1,6 +1,10 @@
 import unittest
 
-from test_gpt_oss_common import BaseTestGptOss
+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=402, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=750, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestGptOss1Gpu(BaseTestGptOss):

test/registered/core/test_hidden_states.py

@@ -4,8 +4,11 @@ import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
 import sglang as sgl
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import DEFAULT_SMALL_MODEL_NAME_FOR_TEST, CustomTestCase
 
+register_cuda_ci(est_time=55, suite="stage-b-test-small-1-gpu")
+
 
 class TestHiddenState(CustomTestCase):
     def test_return_hidden_states(self):

test/registered/core/test_input_embeddings.py

@@ -7,6 +7,7 @@ import requests
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import (
     DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -15,6 +16,9 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_cuda_ci(est_time=38, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=38, suite="stage-b-test-small-1-gpu-amd")
+
 
 class TestInputEmbeds(CustomTestCase):
     @classmethod

test/registered/core/test_io_struct.py

@@ -2,12 +2,16 @@ import copy
 import unittest
 
 from sglang.srt.managers.io_struct import GenerateReqInput
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import (
     DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
     DEFAULT_URL_FOR_TEST,
     CustomTestCase,
 )
 
+register_cuda_ci(est_time=8, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=8, suite="stage-b-test-small-1-gpu-amd")
+
 
 class TestGenerateReqInputNormalization(CustomTestCase):
     """Test the normalization of GenerateReqInput for batch processing and different input formats."""

test/registered/core/test_model_hooks.py

@@ -6,8 +6,12 @@ import torch.nn as nn
 
 from sglang.srt.model_executor.hook_manager import register_forward_hooks
 from sglang.srt.server_args import ServerArgs
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
+register_cuda_ci(est_time=6, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=10, suite="stage-b-test-small-1-gpu-amd")
+
 HOOK_CALLS = []
 
 

test/registered/core/test_page_size.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, register_cuda_ci
 from sglang.test.run_eval import run_eval
 from sglang.test.test_utils import (
     DEFAULT_MODEL_NAME_FOR_TEST,
@@ -12,6 +13,9 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_cuda_ci(est_time=60, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=60, suite="stage-b-test-small-1-gpu-amd")
+
 
 class TestPageSize(CustomTestCase):
     @classmethod

test/registered/core/test_request_queue_validation.py

@@ -4,6 +4,7 @@ import re
 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.test_utils import (
     DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -16,6 +17,9 @@ from sglang.test.test_utils import (
     send_generate_requests,
 )
 
+register_cuda_ci(est_time=47, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=70, suite="stage-b-test-small-1-gpu-amd")
+
 
 class TestMaxQueuedRequests(CustomTestCase):
     @classmethod

test/registered/core/test_score_api.py

@@ -4,8 +4,11 @@ import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
 from sglang.srt.entrypoints.engine import Engine
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import DEFAULT_SMALL_MODEL_NAME_FOR_TEST, CustomTestCase
 
+register_cuda_ci(est_time=260, suite="stage-b-test-small-1-gpu")
+
 TEST_MODEL_NAME = DEFAULT_SMALL_MODEL_NAME_FOR_TEST
 
 

test/registered/core/test_server_args.py

@@ -3,8 +3,12 @@ import unittest
 from unittest.mock import patch
 
 from sglang.srt.server_args import PortArgs, ServerArgs, prepare_server_args
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
+register_cuda_ci(est_time=9, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=1, suite="stage-b-test-small-1-gpu-amd")
+
 
 class TestPrepareServerArgs(CustomTestCase):
     def test_prepare_server_args(self):

test/registered/core/test_srt_endpoint.py

@@ -17,6 +17,7 @@ import requests
 
 from sglang.srt.sampling.custom_logit_processor import CustomLogitProcessor
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import (
     DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -26,6 +27,9 @@ from sglang.test.test_utils import (
     run_logprob_check,
 )
 
+register_cuda_ci(est_time=127, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=130, suite="stage-b-test-small-1-gpu-amd")
+
 
 class TestSRTEndpoint(CustomTestCase):
     @classmethod

test/registered/core/test_srt_engine.py

@@ -14,6 +14,7 @@ import sglang as sgl
 from sglang.bench_offline_throughput import BenchArgs, throughput_test
 from sglang.srt.server_args import ServerArgs
 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.few_shot_gsm8k_engine import run_eval
 from sglang.test.test_utils import (
     DEFAULT_SMALL_EMBEDDING_MODEL_NAME_FOR_TEST,
@@ -21,6 +22,9 @@ from sglang.test.test_utils import (
     CustomTestCase,
 )
 
+register_cuda_ci(est_time=252, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=261, suite="stage-b-test-small-1-gpu-amd")
+
 
 class TestSRTEngine(CustomTestCase):
 

test/srt/run_suite.py

@@ -9,24 +9,12 @@ from sglang.test.ci.ci_utils import TestFile, run_unittest_files
 # NOTE: please sort the test cases alphabetically by the test file name
 suites = {
     "per-commit-1-gpu": [
-        TestFile("test_deterministic.py", 228),
         TestFile("test_evs.py", 20),
         TestFile("test_external_models.py", 30),
-        TestFile("test_gpt_oss_1gpu.py", 402),
-        TestFile("test_hidden_states.py", 55),
-        TestFile("test_input_embeddings.py", 38),
-        TestFile("test_io_struct.py", 8),
         TestFile("test_jinja_template_utils.py", 7),
-        TestFile("test_model_hooks.py", 6),
         TestFile("test_modelopt_loader.py", 11),
         TestFile("test_multi_tokenizer.py", 230),
-        TestFile("test_page_size.py", 60),
-        TestFile("test_request_queue_validation.py", 47),
-        TestFile("test_score_api.py", 260),
-        TestFile("test_server_args.py", 9),
         TestFile("test_skip_tokenizer_init.py", 77),
-        TestFile("test_srt_endpoint.py", 127),
-        TestFile("test_srt_engine.py", 252),
         TestFile("test_utils_update_weights.py", 29),
         TestFile("test_video_utils.py", 5),
         TestFile("test_modelopt_export.py", 9),
@@ -98,7 +86,6 @@ suites = {
         TestFile("test_bench_one_batch.py"),
         TestFile("test_bench_serving.py"),
         TestFile("test_eval_accuracy_large.py"),
-        TestFile("test_gpt_oss_common.py"),
         TestFile("test_moe_eval_accuracy_large.py"),
         TestFile("test_profile_v2.py"),
         TestFile("models/test_ministral3_models.py"),
@@ -120,18 +107,10 @@ suite_amd = {
         # TestFile("lora/test_lora_qwen3.py", 97), # Disabled temporarily, see https://github.com/sgl-project/sglang/issues/13107
         TestFile("test_bench_typebaseddispatcher.py", 10),
         TestFile("test_external_models.py", 45),
-        TestFile("test_input_embeddings.py", 38),
-        TestFile("test_io_struct.py", 8),
         TestFile("test_jinja_template_utils.py", 1),
-        TestFile("test_model_hooks.py", 10),
         TestFile("test_multi_tokenizer.py", 345),
-        TestFile("test_page_size.py", 60),
-        TestFile("test_request_queue_validation.py", 70),
         TestFile("test_rope_rocm.py", 3),
-        TestFile("test_server_args.py", 1),
         TestFile("test_skip_tokenizer_init.py", 117),
-        TestFile("test_srt_endpoint.py", 130),
-        TestFile("test_srt_engine.py", 261),
         # TestFile("test_torch_compile_moe.py", 210), # Disabled temporarily, see https://github.com/sgl-project/sglang/issues/13107
         TestFile("test_type_based_dispatcher.py", 10),
         TestFile("test_video_utils.py", 8),
@@ -143,9 +122,7 @@ suite_amd = {
         # TestFile("test_vision_chunked_prefill.py", 175), # Disabled temporarily and track in #7701
         # TestFile("test_wave_attention_backend.py", 150), # Disabled temporarily, see https://github.com/sgl-project/sglang/issues/11127
     ],
-    "per-commit-amd-mi35x": [
-        TestFile("test_gpt_oss_1gpu.py", 750),
-    ],
+    "per-commit-amd-mi35x": [],
     "per-commit-2-gpu-amd": [
         TestFile("test_data_parallelism.py", 73),
         TestFile("test_load_weights_from_remote_instance.py", 72),

test/srt/test_gpt_oss_4gpu.py

@@ -1,6 +1,6 @@
 import unittest
 
-from test_gpt_oss_common import BaseTestGptOss
+from sglang.test.gpt_oss_common import BaseTestGptOss
 
 
 class TestGptOss4Gpu(BaseTestGptOss):

test/srt/test_gpt_oss_common.py

@@ -1,152 +0,0 @@
-import json
-import os
-from concurrent.futures import ThreadPoolExecutor
-from types import SimpleNamespace
-from typing import Dict, List, Literal, Optional
-
-import requests
-
-from sglang.srt.utils import is_hip, kill_process_tree
-from sglang.test.run_eval import run_eval
-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,
-)
-
-_base_url = DEFAULT_URL_FOR_TEST
-_is_hip = is_hip()
-
-
-class BaseTestGptOss(CustomTestCase):
-    def run_test(
-        self,
-        model_variant: Literal["20b", "120b"],
-        quantization: Literal["mxfp4", "bf16"],
-        expected_score_of_reasoning_effort: Dict[str, float],
-        other_args: Optional[List[str]] = None,
-    ):
-        if other_args is None:
-            other_args = []
-
-        model = {
-            ("20b", "bf16"): "lmsys/gpt-oss-20b-bf16",
-            ("120b", "bf16"): "lmsys/gpt-oss-120b-bf16",
-            ("20b", "mxfp4"): "openai/gpt-oss-20b",
-            ("120b", "mxfp4"): "openai/gpt-oss-120b",
-        }[(model_variant, quantization)]
-
-        if model_variant == "20b":
-            other_args += ["--cuda-graph-max-bs", "600"]
-        if _is_hip:
-            os.environ["SGLANG_USE_AITER"] = "0"
-        self._run_test_raw(
-            model=model,
-            expected_score_of_reasoning_effort=expected_score_of_reasoning_effort,
-            other_args=other_args,
-        )
-
-    def _run_test_raw(
-        self,
-        model: str,
-        expected_score_of_reasoning_effort: Dict[str, float],
-        other_args: List[str],
-    ):
-        process = popen_launch_server(
-            model,
-            _base_url,
-            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
-            other_args=other_args,
-        )
-
-        try:
-            self._check_streaming_responses_api_request(model)
-
-            # run multiple tests in parallel since we are mostly bound by the longest generate sequence
-            # instead of the number of questions
-            with ThreadPoolExecutor(max_workers=4) as executor:
-                list(
-                    executor.map(
-                        lambda d: self._run_one_eval(**d),
-                        [
-                            dict(
-                                model=model,
-                                reasoning_effort=reasoning_effort,
-                                expected_score=expected_score,
-                            )
-                            for reasoning_effort, expected_score in expected_score_of_reasoning_effort.items()
-                        ],
-                    )
-                )
-        finally:
-            kill_process_tree(process.pid)
-
-    def _check_streaming_responses_api_request(self, model):
-        # Use requests to verify /v1/responses streaming
-        url = f"{_base_url}/v1/responses"
-        payload = {
-            "model": model,
-            "input": "What is 1 + 1?",
-            "stream": True,
-            "temperature": 0,
-        }
-
-        response = requests.post(url, json=payload, stream=True)
-        if response.status_code != 200:
-            print(f"Response API failed: {response.text}")
-        response.raise_for_status()
-
-        content = ""
-        for line in response.iter_lines():
-            if line:
-                decoded_line = line.decode("utf-8")
-                if decoded_line.startswith("data: "):
-                    data_str = decoded_line[6:]
-                    if data_str.strip() == "[DONE]":
-                        break
-
-                    try:
-                        data = json.loads(data_str)
-                        if data.get("type") == "response.output_text.delta":
-                            delta = data.get("delta", "")
-                            content += delta
-                    except json.JSONDecodeError:
-                        pass
-
-        print(f"Streaming check response: {content}")
-        self.assertTrue(len(content) > 0)
-        self.assertIn("2", content)
-
-    def _run_one_eval(self, model, reasoning_effort, expected_score):
-        args = SimpleNamespace(
-            base_url=_base_url,
-            model=model,
-            eval_name="gpqa",
-            num_examples=198,
-            # use enough threads to allow parallelism
-            num_threads=198,
-            # TODO 4k is still not enough, we need e.g. 64k token, but that is super slow
-            # otherwise a lot of questions are not answered
-            max_tokens=4096,
-            # simple-evals by default use 0.5 and is better than 0.0 temperature
-            # but here for reproducibility, we use 0.1
-            temperature=0.1,
-            reasoning_effort=reasoning_effort,
-        )
-
-        setup = f"model={model} reasoning_effort={reasoning_effort} expected_score={expected_score}"
-
-        print(f"Evaluation start: {setup}")
-        metrics = run_eval(args)
-        print(f"Evaluation end: {setup} {metrics=}")
-        self.assertGreaterEqual(metrics["score"], expected_score)
-
-        if is_in_ci():
-            write_github_step_summary(
-                f"### test_gpt_oss_common\n"
-                f"Setup: {setup}\n"
-                f"Score: {metrics['score']:.2f}\n"
-            )