refactor(test): reorganize OpenAI test file structure (#7408)

test/srt/test_large_max_new_tokens.py → test/srt/openai_server/validation/test_large_max_new_tokens.py

 """
-python3 -m unittest test_large_max_new_tokens.TestLargeMaxNewTokens.test_chat_completion
+python3 -m unittest openai_server.validation.test_large_max_new_tokens.TestLargeMaxNewTokens.test_chat_completion
 """
 
 import os

test/srt/openai_server/validation/test_openai_server_ignore_eos.py

+import openai
+
+from sglang.srt.hf_transformers_utils import get_tokenizer
+from sglang.srt.utils import kill_process_tree
+from sglang.test.test_utils import (
+    DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
+    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+    DEFAULT_URL_FOR_TEST,
+    CustomTestCase,
+    popen_launch_server,
+)
+
+
+class TestOpenAIServerIgnoreEOS(CustomTestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.model = DEFAULT_SMALL_MODEL_NAME_FOR_TEST
+        cls.base_url = DEFAULT_URL_FOR_TEST
+        cls.api_key = "sk-123456"
+        cls.process = popen_launch_server(
+            cls.model,
+            cls.base_url,
+            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+            api_key=cls.api_key,
+        )
+        cls.base_url += "/v1"
+        cls.tokenizer = get_tokenizer(DEFAULT_SMALL_MODEL_NAME_FOR_TEST)
+
+    @classmethod
+    def tearDownClass(cls):
+        kill_process_tree(cls.process.pid)
+
+    def test_ignore_eos(self):
+        """
+        Test that ignore_eos=True allows generation to continue beyond EOS token
+        and reach the max_tokens limit.
+        """
+        client = openai.Client(api_key=self.api_key, base_url=self.base_url)
+
+        max_tokens = 200
+
+        response_default = client.chat.completions.create(
+            model=self.model,
+            messages=[
+                {"role": "system", "content": "You are a helpful assistant."},
+                {"role": "user", "content": "Count from 1 to 20."},
+            ],
+            temperature=0,
+            max_tokens=max_tokens,
+            extra_body={"ignore_eos": False},
+        )
+
+        response_ignore_eos = client.chat.completions.create(
+            model=self.model,
+            messages=[
+                {"role": "system", "content": "You are a helpful assistant."},
+                {"role": "user", "content": "Count from 1 to 20."},
+            ],
+            temperature=0,
+            max_tokens=max_tokens,
+            extra_body={"ignore_eos": True},
+        )
+
+        default_tokens = len(
+            self.tokenizer.encode(response_default.choices[0].message.content)
+        )
+        ignore_eos_tokens = len(
+            self.tokenizer.encode(response_ignore_eos.choices[0].message.content)
+        )
+
+        # Check if ignore_eos resulted in more tokens or exactly max_tokens
+        # The ignore_eos response should either:
+        # 1. Have more tokens than the default response (if default stopped at EOS before max_tokens)
+        # 2. Have exactly max_tokens (if it reached the max_tokens limit)
+        self.assertTrue(
+            ignore_eos_tokens > default_tokens or ignore_eos_tokens >= max_tokens,
+            f"ignore_eos did not generate more tokens: {ignore_eos_tokens} vs {default_tokens}",
+        )
+
+        self.assertEqual(
+            response_ignore_eos.choices[0].finish_reason,
+            "length",
+            f"Expected finish_reason='length' for ignore_eos=True, got {response_ignore_eos.choices[0].finish_reason}",
+        )

test/srt/run_suite.py

@@ -29,10 +29,25 @@ suites = {
         TestFile("models/test_reward_models.py", 132),
         TestFile("models/test_vlm_models.py", 437),
         TestFile("models/test_transformers_models.py", 320),
-        TestFile("openai/test_protocol.py", 10),
-        TestFile("openai/test_serving_chat.py", 10),
-        TestFile("openai/test_serving_completions.py", 10),
-        TestFile("openai/test_serving_embedding.py", 10),
+        TestFile("openai_server/basic/test_protocol.py", 10),
+        TestFile("openai_server/basic/test_serving_chat.py", 10),
+        TestFile("openai_server/basic/test_serving_completions.py", 10),
+        TestFile("openai_server/basic/test_serving_embedding.py", 10),
+        TestFile("openai_server/basic/test_openai_embedding.py", 141),
+        TestFile("openai_server/basic/test_openai_server.py", 149),
+        TestFile("openai_server/features/test_cache_report.py", 100),
+        TestFile("openai_server/features/test_enable_thinking.py", 70),
+        TestFile("openai_server/features/test_json_constrained.py", 98),
+        TestFile("openai_server/features/test_json_mode.py", 90),
+        TestFile("openai_server/features/test_openai_server_ebnf.py", 95),
+        TestFile("openai_server/features/test_openai_server_hidden_states.py", 240),
+        TestFile("openai_server/features/test_reasoning_content.py", 89),
+        TestFile("openai_server/function_call/test_openai_function_calling.py", 60),
+        TestFile("openai_server/function_call/test_tool_choice.py", 226),
+        TestFile("openai_server/validation/test_large_max_new_tokens.py", 41),
+        TestFile("openai_server/validation/test_matched_stop.py", 60),
+        TestFile("openai_server/validation/test_openai_server_ignore_eos.py", 85),
+        TestFile("openai_server/validation/test_request_length_validation.py", 31),
         TestFile("test_abort.py", 51),
         TestFile("test_block_int8.py", 22),
         TestFile("test_create_kvindices.py", 2),
@@ -40,8 +55,6 @@ suites = {
         TestFile("test_eagle_infer_a.py", 370),
         TestFile("test_eagle_infer_b.py", 270),
         TestFile("test_ebnf_constrained.py", 108),
-        TestFile("test_enable_thinking.py", 70),
-        TestFile("test_embedding_openai_server.py", 141),
         TestFile("test_eval_fp8_accuracy.py", 303),
         TestFile("test_fa3.py", 376),
         # TestFile("test_flashmla.py", 352),
@@ -54,8 +67,6 @@ suites = {
         TestFile("test_int8_kernel.py", 8),
         TestFile("test_input_embeddings.py", 38),
         TestFile("test_jinja_template_utils.py", 1),
-        TestFile("test_json_constrained.py", 98),
-        TestFile("test_large_max_new_tokens.py", 41),
         TestFile("test_metrics.py", 32),
         TestFile("test_mla.py", 167),
         TestFile("test_mla_deepseek_v3.py", 342),
@@ -64,22 +75,16 @@ suites = {
         TestFile("test_mla_fp8.py", 93),
         TestFile("test_no_chunked_prefill.py", 108),
         TestFile("test_no_overlap_scheduler.py", 234),
-        TestFile("test_openai_function_calling.py", 60),
-        TestFile("test_openai_server.py", 149),
-        TestFile("test_openai_server_hidden_states.py", 240),
         TestFile("test_penalty.py", 41),
         TestFile("test_page_size.py", 60),
         TestFile("test_pytorch_sampling_backend.py", 66),
         TestFile("test_radix_attention.py", 105),
-        TestFile("test_reasoning_content.py", 89),
         TestFile("test_regex_constrained.py", 64),
-        TestFile("test_request_length_validation.py", 31),
         TestFile("test_retract_decode.py", 54),
         TestFile("test_server_args.py", 1),
         TestFile("test_skip_tokenizer_init.py", 117),
         TestFile("test_srt_engine.py", 261),
         TestFile("test_srt_endpoint.py", 130),
-        TestFile("test_tool_choice.py", 226),
         TestFile("test_torch_compile.py", 76),
         TestFile("test_torch_compile_moe.py", 172),
         TestFile("test_torch_native_attention_backend.py", 123),
@@ -107,15 +112,32 @@ suites = {
         TestFile("test_torch_compile_moe.py", 172),
         TestFile("models/test_qwen_models.py", 82),
         TestFile("models/test_reward_models.py", 132),
+        TestFile("openai_server/basic/test_openai_embedding.py", 141),
+        TestFile("openai_server/basic/test_openai_server.py", 149),
+        TestFile("openai_server/basic/test_protocol.py", 10),
+        TestFile("openai_server/basic/test_serving_chat.py", 10),
+        TestFile("openai_server/basic/test_serving_completions.py", 10),
+        TestFile("openai_server/basic/test_serving_embedding.py", 10),
         TestFile("test_abort.py", 51),
         TestFile("test_block_int8.py", 22),
         TestFile("test_create_kvindices.py", 2),
         TestFile("test_chunked_prefill.py", 313),
-        TestFile("test_embedding_openai_server.py", 141),
         TestFile("test_eval_fp8_accuracy.py", 303),
         TestFile("test_function_call_parser.py", 10),
         TestFile("test_input_embeddings.py", 38),
-        TestFile("test_large_max_new_tokens.py", 41),
+        TestFile("openai_server/features/test_cache_report.py", 100),
+        TestFile("openai_server/features/test_enable_thinking.py", 70),
+        TestFile("openai_server/features/test_json_constrained.py", 98),
+        TestFile("openai_server/features/test_json_mode.py", 90),
+        TestFile("openai_server/features/test_openai_server_ebnf.py", 95),
+        TestFile("openai_server/features/test_openai_server_hidden_states.py", 240),
+        TestFile("openai_server/features/test_reasoning_content.py", 89),
+        TestFile("openai_server/function_call/test_openai_function_calling.py", 60),
+        TestFile("openai_server/function_call/test_tool_choice.py", 226),
+        TestFile("openai_server/validation/test_large_max_new_tokens.py", 41),
+        TestFile("openai_server/validation/test_matched_stop.py", 60),
+        TestFile("openai_server/validation/test_openai_server_ignore_eos.py", 85),
+        TestFile("openai_server/validation/test_request_length_validation.py", 31),
         TestFile("test_metrics.py", 32),
         TestFile("test_no_chunked_prefill.py", 108),
         TestFile("test_no_overlap_scheduler.py", 234),
@@ -123,9 +145,6 @@ suites = {
         TestFile("test_page_size.py", 60),
         TestFile("test_pytorch_sampling_backend.py", 66),
         TestFile("test_radix_attention.py", 105),
-        TestFile("test_reasoning_content.py", 89),
-        TestFile("test_enable_thinking.py", 70),
-        TestFile("test_request_length_validation.py", 31),
         TestFile("test_retract_decode.py", 54),
         TestFile("test_server_args.py", 1),
         TestFile("test_skip_tokenizer_init.py", 117),

test/srt/test_embedding_openai_server.py

-import unittest
-
-import openai
-
-from sglang.srt.hf_transformers_utils import get_tokenizer
-from sglang.srt.utils import kill_process_tree
-from sglang.test.test_utils import (
-    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
-    DEFAULT_URL_FOR_TEST,
-    CustomTestCase,
-    popen_launch_server,
-)
-
-
-class TestOpenAIServer(CustomTestCase):
-    @classmethod
-    def setUpClass(cls):
-        cls.model = "intfloat/e5-mistral-7b-instruct"
-        cls.base_url = DEFAULT_URL_FOR_TEST
-        cls.api_key = "sk-123456"
-        cls.process = popen_launch_server(
-            cls.model,
-            cls.base_url,
-            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
-            api_key=cls.api_key,
-        )
-        cls.base_url += "/v1"
-        cls.tokenizer = get_tokenizer(cls.model)
-
-    @classmethod
-    def tearDownClass(cls):
-        kill_process_tree(cls.process.pid)
-
-    def run_embedding(self, use_list_input, token_input):
-        client = openai.Client(api_key=self.api_key, base_url=self.base_url)
-        prompt = "The capital of France is"
-        if token_input:
-            prompt_input = self.tokenizer.encode(prompt)
-            num_prompt_tokens = len(prompt_input)
-        else:
-            prompt_input = prompt
-            num_prompt_tokens = len(self.tokenizer.encode(prompt))
-
-        if use_list_input:
-            prompt_arg = [prompt_input] * 2
-            num_prompts = len(prompt_arg)
-            num_prompt_tokens *= num_prompts
-        else:
-            prompt_arg = prompt_input
-            num_prompts = 1
-
-        response = client.embeddings.create(
-            input=prompt_arg,
-            model=self.model,
-        )
-
-        assert len(response.data) == num_prompts
-        assert isinstance(response.data, list)
-        assert response.data[0].embedding
-        assert response.data[0].index is not None
-        assert response.data[0].object == "embedding"
-        assert response.model == self.model
-        assert response.object == "list"
-        assert (
-            response.usage.prompt_tokens == num_prompt_tokens
-        ), f"{response.usage.prompt_tokens} vs {num_prompt_tokens}"
-        assert (
-            response.usage.total_tokens == num_prompt_tokens
-        ), f"{response.usage.total_tokens} vs {num_prompt_tokens}"
-
-    def run_batch(self):
-        # FIXME: not implemented
-        pass
-
-    def test_embedding(self):
-        # TODO: the fields of encoding_format, dimensions, user are skipped
-        # TODO: support use_list_input
-        for use_list_input in [False, True]:
-            for token_input in [False, True]:
-                self.run_embedding(use_list_input, token_input)
-
-    def test_batch(self):
-        self.run_batch()
-
-
-if __name__ == "__main__":
-    unittest.main()

test/srt/test_jinja_template_utils.py

 """
-Unit tests for OpenAI adapter utils.
+Unit tests for Jinja chat template utils.
 """
 
 import unittest