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

test/srt/openai_server/basic/test_openai_embedding.py

@@ -0,0 +1,97 @@
+import unittest
+
+import openai
+
+from sglang.srt.utils import kill_process_tree
+from sglang.test.test_utils import (
+    DEFAULT_SMALL_EMBEDDING_MODEL_NAME_FOR_TEST,
+    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+    DEFAULT_URL_FOR_TEST,
+    CustomTestCase,
+    popen_launch_server,
+)
+
+
+class TestOpenAIEmbedding(CustomTestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.model = DEFAULT_SMALL_EMBEDDING_MODEL_NAME_FOR_TEST
+        cls.base_url = DEFAULT_URL_FOR_TEST
+        cls.api_key = "sk-123456"
+
+        # Configure embedding-specific args
+        other_args = ["--is-embedding", "--enable-metrics"]
+        cls.process = popen_launch_server(
+            cls.model,
+            cls.base_url,
+            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+            api_key=cls.api_key,
+            other_args=other_args,
+        )
+        cls.base_url += "/v1"
+
+    @classmethod
+    def tearDownClass(cls):
+        kill_process_tree(cls.process.pid)
+
+    def test_embedding_single(self):
+        """Test single embedding request"""
+        client = openai.Client(api_key=self.api_key, base_url=self.base_url)
+        response = client.embeddings.create(model=self.model, input="Hello world")
+        self.assertEqual(len(response.data), 1)
+        self.assertTrue(len(response.data[0].embedding) > 0)
+
+    def test_embedding_batch(self):
+        """Test batch embedding request"""
+        client = openai.Client(api_key=self.api_key, base_url=self.base_url)
+        response = client.embeddings.create(
+            model=self.model, input=["Hello world", "Test text"]
+        )
+        self.assertEqual(len(response.data), 2)
+        self.assertTrue(len(response.data[0].embedding) > 0)
+        self.assertTrue(len(response.data[1].embedding) > 0)
+
+    def test_embedding_single_batch_str(self):
+        """Test embedding with a List[str] and length equals to 1"""
+        client = openai.Client(api_key=self.api_key, base_url=self.base_url)
+        response = client.embeddings.create(model=self.model, input=["Hello world"])
+        self.assertEqual(len(response.data), 1)
+        self.assertTrue(len(response.data[0].embedding) > 0)
+
+    def test_embedding_single_int_list(self):
+        """Test embedding with a List[int] or List[List[int]]]"""
+        client = openai.Client(api_key=self.api_key, base_url=self.base_url)
+        response = client.embeddings.create(
+            model=self.model,
+            input=[[15339, 314, 703, 284, 612, 262, 10658, 10188, 286, 2061]],
+        )
+        self.assertEqual(len(response.data), 1)
+        self.assertTrue(len(response.data[0].embedding) > 0)
+
+        client = openai.Client(api_key=self.api_key, base_url=self.base_url)
+        response = client.embeddings.create(
+            model=self.model,
+            input=[15339, 314, 703, 284, 612, 262, 10658, 10188, 286, 2061],
+        )
+        self.assertEqual(len(response.data), 1)
+        self.assertTrue(len(response.data[0].embedding) > 0)
+
+    def test_empty_string_embedding(self):
+        """Test embedding an empty string."""
+
+        client = openai.Client(api_key=self.api_key, base_url=self.base_url)
+
+        # Text embedding example with empty string
+        text = ""
+        # Expect a BadRequestError for empty input
+        with self.assertRaises(openai.BadRequestError) as cm:
+            client.embeddings.create(
+                model=self.model,
+                input=text,
+            )
+        # check the status code
+        self.assertEqual(cm.exception.status_code, 400)
+
+
+if __name__ == "__main__":
+    unittest.main()

test/srt/openai_server/basic/test_openai_server.py

@@ -0,0 +1,743 @@
+"""
+python3 -m unittest openai_server.basic.test_openai_server.TestOpenAIServer.test_completion
+python3 -m unittest openai_server.basic.test_openai_server.TestOpenAIServer.test_completion_stream
+python3 -m unittest openai_server.basic.test_openai_server.TestOpenAIServer.test_chat_completion
+python3 -m unittest openai_server.basic.test_openai_server.TestOpenAIServer.test_chat_completion_stream
+"""
+
+import json
+import re
+import unittest
+
+import numpy as np
+import openai
+import requests
+
+from sglang.srt.hf_transformers_utils import get_tokenizer
+from sglang.srt.utils import kill_process_tree
+from sglang.test.runners import TEST_RERANK_QUERY_DOCS
+from sglang.test.test_utils import (
+    DEFAULT_SMALL_CROSS_ENCODER_MODEL_NAME_FOR_TEST,
+    DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
+    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+    DEFAULT_URL_FOR_TEST,
+    CustomTestCase,
+    popen_launch_server,
+)
+
+
+class TestOpenAIServer(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 run_completion(
+        self, echo, logprobs, use_list_input, parallel_sample_num, 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, prompt_input]
+            num_choices = len(prompt_arg)
+            num_prompt_tokens *= 2
+        else:
+            prompt_arg = prompt_input
+            num_choices = 1
+
+        response = client.completions.create(
+            model=self.model,
+            prompt=prompt_arg,
+            temperature=0,
+            max_tokens=32,
+            echo=echo,
+            logprobs=logprobs,
+            n=parallel_sample_num,
+        )
+
+        assert len(response.choices) == num_choices * parallel_sample_num
+
+        if echo:
+            text = response.choices[0].text
+            assert text.startswith(prompt)
+
+        if logprobs:
+            assert response.choices[0].logprobs
+            assert isinstance(response.choices[0].logprobs.tokens[0], str)
+            assert isinstance(response.choices[0].logprobs.top_logprobs[1], dict)
+            ret_num_top_logprobs = len(response.choices[0].logprobs.top_logprobs[1])
+
+            # FIXME: Sometimes, some top_logprobs are missing in the return value. The reason is that some output id maps to the same output token and duplicate in the map
+            # assert ret_num_top_logprobs == logprobs, f"{ret_num_top_logprobs} vs {logprobs}"
+            assert ret_num_top_logprobs > 0
+
+            # when echo=True and request.logprobs>0, logprob_start_len is 0, so the first token's logprob would be None.
+            if not echo:
+                assert response.choices[0].logprobs.token_logprobs[0]
+
+        assert response.id
+        assert response.created
+        assert (
+            response.usage.prompt_tokens == num_prompt_tokens
+        ), f"{response.usage.prompt_tokens} vs {num_prompt_tokens}"
+        assert response.usage.completion_tokens > 0
+        assert response.usage.total_tokens > 0
+
+    def run_completion_stream(
+        self, echo, logprobs, use_list_input, parallel_sample_num, 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, prompt_input]
+            num_choices = len(prompt_arg)
+            num_prompt_tokens *= 2
+        else:
+            prompt_arg = prompt_input
+            num_choices = 1
+
+        generator = client.completions.create(
+            model=self.model,
+            prompt=prompt_arg,
+            temperature=0,
+            max_tokens=32,
+            echo=echo,
+            logprobs=logprobs,
+            stream=True,
+            stream_options={"include_usage": True},
+            n=parallel_sample_num,
+        )
+
+        is_firsts = {}
+        for response in generator:
+            usage = response.usage
+            if usage is not None:
+                assert usage.prompt_tokens > 0, f"usage.prompt_tokens was zero"
+                assert usage.completion_tokens > 0, f"usage.completion_tokens was zero"
+                assert usage.total_tokens > 0, f"usage.total_tokens was zero"
+                continue
+
+            index = response.choices[0].index
+            is_first = is_firsts.get(index, True)
+
+            if logprobs:
+                assert response.choices[0].logprobs, f"no logprobs in response"
+                assert isinstance(
+                    response.choices[0].logprobs.tokens[0], str
+                ), f"{response.choices[0].logprobs.tokens[0]} is not a string"
+                if not (is_first and echo):
+                    assert isinstance(
+                        response.choices[0].logprobs.top_logprobs[0], dict
+                    ), f"top_logprobs was not a dictionary"
+                    ret_num_top_logprobs = len(
+                        response.choices[0].logprobs.top_logprobs[0]
+                    )
+                    # FIXME: Sometimes, some top_logprobs are missing in the return value. The reason is that some output id maps to the same output token and duplicate in the map
+                    # assert ret_num_top_logprobs == logprobs, f"{ret_num_top_logprobs} vs {logprobs}"
+                    assert ret_num_top_logprobs > 0, f"ret_num_top_logprobs was 0"
+
+            if is_first:
+                if echo:
+                    assert response.choices[0].text.startswith(
+                        prompt
+                    ), f"{response.choices[0].text} and all args {echo} {logprobs} {token_input} {is_first}"
+                is_firsts[index] = False
+            assert response.id, f"no id in response"
+            assert response.created, f"no created in response"
+
+        for index in [i for i in range(parallel_sample_num * num_choices)]:
+            assert not is_firsts.get(
+                index, True
+            ), f"index {index} is not found in the response"
+
+    def run_chat_completion(self, logprobs, parallel_sample_num):
+        client = openai.Client(api_key=self.api_key, base_url=self.base_url)
+        response = client.chat.completions.create(
+            model=self.model,
+            messages=[
+                {"role": "system", "content": "You are a helpful AI assistant"},
+                {
+                    "role": "user",
+                    "content": "What is the capital of France? Answer in a few words.",
+                },
+            ],
+            temperature=0,
+            logprobs=logprobs is not None and logprobs > 0,
+            top_logprobs=logprobs,
+            n=parallel_sample_num,
+        )
+
+        if logprobs:
+            assert isinstance(
+                response.choices[0].logprobs.content[0].top_logprobs[0].token, str
+            )
+
+            ret_num_top_logprobs = len(
+                response.choices[0].logprobs.content[0].top_logprobs
+            )
+            assert (
+                ret_num_top_logprobs == logprobs
+            ), f"{ret_num_top_logprobs} vs {logprobs}"
+
+        assert len(response.choices) == parallel_sample_num
+        assert response.choices[0].message.role == "assistant"
+        assert isinstance(response.choices[0].message.content, str)
+        assert response.id
+        assert response.created
+        assert response.usage.prompt_tokens > 0
+        assert response.usage.completion_tokens > 0
+        assert response.usage.total_tokens > 0
+
+    def run_chat_completion_stream(self, logprobs, parallel_sample_num=1):
+        client = openai.Client(api_key=self.api_key, base_url=self.base_url)
+        generator = client.chat.completions.create(
+            model=self.model,
+            messages=[
+                {"role": "system", "content": "You are a helpful AI assistant"},
+                {"role": "user", "content": "What is the capital of France?"},
+            ],
+            temperature=0,
+            logprobs=logprobs is not None and logprobs > 0,
+            top_logprobs=logprobs,
+            stream=True,
+            stream_options={"include_usage": True},
+            n=parallel_sample_num,
+        )
+
+        is_firsts = {}
+        is_finished = {}
+        for response in generator:
+            usage = response.usage
+            if usage is not None:
+                assert usage.prompt_tokens > 0, f"usage.prompt_tokens was zero"
+                assert usage.completion_tokens > 0, f"usage.completion_tokens was zero"
+                assert usage.total_tokens > 0, f"usage.total_tokens was zero"
+                continue
+
+            index = response.choices[0].index
+            finish_reason = response.choices[0].finish_reason
+            if finish_reason is not None:
+                is_finished[index] = True
+
+            data = response.choices[0].delta
+
+            if is_firsts.get(index, True):
+                assert (
+                    data.role == "assistant"
+                ), f"data.role was not 'assistant' for first chunk"
+                is_firsts[index] = False
+                continue
+
+            if logprobs and not is_finished.get(index, False):
+                assert response.choices[0].logprobs, f"logprobs was not returned"
+                assert isinstance(
+                    response.choices[0].logprobs.content[0].top_logprobs[0].token, str
+                ), f"top_logprobs token was not a string"
+                assert isinstance(
+                    response.choices[0].logprobs.content[0].top_logprobs, list
+                ), f"top_logprobs was not a list"
+                ret_num_top_logprobs = len(
+                    response.choices[0].logprobs.content[0].top_logprobs
+                )
+                assert (
+                    ret_num_top_logprobs == logprobs
+                ), f"{ret_num_top_logprobs} vs {logprobs}"
+
+            assert (
+                isinstance(data.content, str)
+                or isinstance(data.reasoning_content, str)
+                or (isinstance(data.tool_calls, list) and len(data.tool_calls) > 0)
+                or response.choices[0].finish_reason
+            )
+            assert response.id
+            assert response.created
+
+        for index in [i for i in range(parallel_sample_num)]:
+            assert not is_firsts.get(
+                index, True
+            ), f"index {index} is not found in the response"
+
+    def test_completion(self):
+        for echo in [False, True]:
+            for logprobs in [None, 5]:
+                for use_list_input in [True, False]:
+                    for parallel_sample_num in [1, 2]:
+                        for token_input in [False, True]:
+                            self.run_completion(
+                                echo,
+                                logprobs,
+                                use_list_input,
+                                parallel_sample_num,
+                                token_input,
+                            )
+
+    def test_completion_stream(self):
+        # parallel sampling and list input are not supported in streaming mode
+        for echo in [False, True]:
+            for logprobs in [None, 5]:
+                for use_list_input in [True, False]:
+                    for parallel_sample_num in [1, 2]:
+                        for token_input in [False, True]:
+                            self.run_completion_stream(
+                                echo,
+                                logprobs,
+                                use_list_input,
+                                parallel_sample_num,
+                                token_input,
+                            )
+
+    def test_chat_completion(self):
+        for logprobs in [None, 5]:
+            for parallel_sample_num in [1, 2]:
+                self.run_chat_completion(logprobs, parallel_sample_num)
+
+    def test_chat_completion_stream(self):
+        for logprobs in [None, 5]:
+            for parallel_sample_num in [1, 2]:
+                self.run_chat_completion_stream(logprobs, parallel_sample_num)
+
+    def test_regex(self):
+        client = openai.Client(api_key=self.api_key, base_url=self.base_url)
+
+        regex = (
+            r"""\{\n"""
+            + r"""   "name": "[\w]+",\n"""
+            + r"""   "population": [\d]+\n"""
+            + r"""\}"""
+        )
+
+        response = client.chat.completions.create(
+            model=self.model,
+            messages=[
+                {"role": "system", "content": "You are a helpful AI assistant"},
+                {"role": "user", "content": "Introduce the capital of France."},
+            ],
+            temperature=0,
+            max_tokens=128,
+            extra_body={"regex": regex},
+        )
+        text = response.choices[0].message.content
+
+        try:
+            js_obj = json.loads(text)
+        except (TypeError, json.decoder.JSONDecodeError):
+            print("JSONDecodeError", text)
+            raise
+        assert isinstance(js_obj["name"], str)
+        assert isinstance(js_obj["population"], int)
+
+    def test_penalty(self):
+        client = openai.Client(api_key=self.api_key, base_url=self.base_url)
+
+        response = client.chat.completions.create(
+            model=self.model,
+            messages=[
+                {"role": "system", "content": "You are a helpful AI assistant"},
+                {"role": "user", "content": "Introduce the capital of France."},
+            ],
+            temperature=0,
+            max_tokens=32,
+            frequency_penalty=1.0,
+        )
+        text = response.choices[0].message.content
+        assert isinstance(text, str)
+
+    def test_response_prefill(self):
+        client = openai.Client(api_key=self.api_key, base_url=self.base_url)
+
+        response = client.chat.completions.create(
+            model="meta-llama/Llama-3.1-8B-Instruct",
+            messages=[
+                {"role": "system", "content": "You are a helpful AI assistant"},
+                {
+                    "role": "user",
+                    "content": """
+Extract the name, size, price, and color from this product description as a JSON object:
+
+<description>
+The SmartHome Mini is a compact smart home assistant available in black or white for only $49.99. At just 5 inches wide, it lets you control lights, thermostats, and other connected devices via voice or app—no matter where you place it in your home. This affordable little hub brings convenient hands-free control to your smart devices.
+</description>
+""",
+                },
+                {
+                    "role": "assistant",
+                    "content": "{\n",
+                },
+            ],
+            temperature=0,
+            extra_body={"continue_final_message": True},
+        )
+
+        assert (
+            response.choices[0]
+            .message.content.strip()
+            .startswith('"name": "SmartHome Mini",')
+        )
+
+    def test_model_list(self):
+        client = openai.Client(api_key=self.api_key, base_url=self.base_url)
+        models = list(client.models.list())
+        assert len(models) == 1
+        assert isinstance(getattr(models[0], "max_model_len", None), int)
+
+    def test_retrieve_model(self):
+        client = openai.Client(api_key=self.api_key, base_url=self.base_url)
+
+        # Test retrieving an existing model
+        retrieved_model = client.models.retrieve(self.model)
+        self.assertEqual(retrieved_model.id, self.model)
+        self.assertEqual(retrieved_model.root, self.model)
+
+        # Test retrieving a non-existent model
+        with self.assertRaises(openai.NotFoundError):
+            client.models.retrieve("non-existent-model")
+
+
+# -------------------------------------------------------------------------
+#    EBNF Test Class: TestOpenAIServerEBNF
+#    Launches the server with xgrammar, has only EBNF tests
+# -------------------------------------------------------------------------
+class TestOpenAIServerEBNF(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"
+
+        # passing xgrammar specifically
+        other_args = ["--grammar-backend", "xgrammar"]
+        cls.process = popen_launch_server(
+            cls.model,
+            cls.base_url,
+            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+            api_key=cls.api_key,
+            other_args=other_args,
+        )
+        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_ebnf(self):
+        """
+        Ensure we can pass `ebnf` to the local openai server
+        and that it enforces the grammar.
+        """
+        client = openai.Client(api_key=self.api_key, base_url=self.base_url)
+        ebnf_grammar = r"""
+        root ::= "Hello" | "Hi" | "Hey"
+        """
+        pattern = re.compile(r"^(Hello|Hi|Hey)[.!?]*\s*$")
+
+        response = client.chat.completions.create(
+            model=self.model,
+            messages=[
+                {"role": "system", "content": "You are a helpful EBNF test bot."},
+                {"role": "user", "content": "Say a greeting (Hello, Hi, or Hey)."},
+            ],
+            temperature=0,
+            max_tokens=32,
+            extra_body={"ebnf": ebnf_grammar},
+        )
+        text = response.choices[0].message.content.strip()
+        self.assertTrue(len(text) > 0, "Got empty text from EBNF generation")
+        self.assertRegex(text, pattern, f"Text '{text}' doesn't match EBNF choices")
+
+    def test_ebnf_strict_json(self):
+        """
+        A stricter EBNF that produces exactly {"name":"Alice"} format
+        with no trailing punctuation or extra fields.
+        """
+        client = openai.Client(api_key=self.api_key, base_url=self.base_url)
+        ebnf_grammar = r"""
+        root    ::= "{" pair "}"
+        pair    ::= "\"name\"" ":" string
+        string  ::= "\"" [A-Za-z]+ "\""
+        """
+        pattern = re.compile(r'^\{"name":"[A-Za-z]+"\}$')
+
+        response = client.chat.completions.create(
+            model=self.model,
+            messages=[
+                {"role": "system", "content": "EBNF mini-JSON generator."},
+                {
+                    "role": "user",
+                    "content": "Generate single key JSON with only letters.",
+                },
+            ],
+            temperature=0,
+            max_tokens=64,
+            extra_body={"ebnf": ebnf_grammar},
+        )
+        text = response.choices[0].message.content.strip()
+        self.assertTrue(len(text) > 0, "Got empty text from EBNF strict JSON test")
+        self.assertRegex(
+            text, pattern, f"Text '{text}' not matching the EBNF strict JSON shape"
+        )
+
+
+class TestOpenAIV1Rerank(CustomTestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.model = DEFAULT_SMALL_CROSS_ENCODER_MODEL_NAME_FOR_TEST
+        cls.base_url = DEFAULT_URL_FOR_TEST
+        cls.api_key = "sk-123456"
+        cls.score_tolerance = 1e-2
+
+        # Configure embedding-specific args
+        other_args = [
+            "--is-embedding",
+            "--enable-metrics",
+            "--disable-radix-cache",
+            "--chunked-prefill-size",
+            "-1",
+            "--attention-backend",
+            "torch_native",
+        ]
+        cls.process = popen_launch_server(
+            cls.model,
+            cls.base_url,
+            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+            api_key=cls.api_key,
+            other_args=other_args,
+        )
+        cls.base_url += "/v1/rerank"
+
+    @classmethod
+    def tearDownClass(cls):
+        kill_process_tree(cls.process.pid)
+
+    def run_rerank(self, query, docs):
+        response = requests.post(
+            self.base_url,
+            headers={
+                "Authorization": f"Bearer {self.api_key}",
+                "Content-Type": "application/json",
+            },
+            json={"query": query, "documents": docs},
+        )
+
+        return response.json()
+
+    def test_rerank_single(self):
+        """Test single rerank request"""
+        query = TEST_RERANK_QUERY_DOCS[0]["query"]
+        docs = TEST_RERANK_QUERY_DOCS[0]["documents"]
+
+        response = self.run_rerank(query, docs)
+
+        self.assertEqual(len(response), 1)
+        self.assertTrue(isinstance(response[0]["score"], float))
+        self.assertTrue(isinstance(response[0]["document"], str))
+        self.assertTrue(isinstance(response[0]["index"], int))
+
+    def test_rerank_batch(self):
+        """Test batch rerank request"""
+        query = TEST_RERANK_QUERY_DOCS[1]["query"]
+        docs = TEST_RERANK_QUERY_DOCS[1]["documents"]
+
+        response = self.run_rerank(query, docs)
+
+        self.assertEqual(len(response), 2)
+        self.assertTrue(isinstance(response[0]["score"], float))
+        self.assertTrue(isinstance(response[1]["score"], float))
+        self.assertTrue(isinstance(response[0]["document"], str))
+        self.assertTrue(isinstance(response[1]["document"], str))
+        self.assertTrue(isinstance(response[0]["index"], int))
+        self.assertTrue(isinstance(response[1]["index"], int))
+
+
+class TestOpenAIV1Score(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/score"
+        cls.tokenizer = get_tokenizer(DEFAULT_SMALL_MODEL_NAME_FOR_TEST)
+
+    @classmethod
+    def tearDownClass(cls):
+        kill_process_tree(cls.process.pid)
+
+    def run_score(
+        self, query, items, label_token_ids, apply_softmax=False, item_first=False
+    ):
+        response = requests.post(
+            self.base_url,
+            headers={
+                "Authorization": f"Bearer {self.api_key}",
+                "Content-Type": "application/json",
+            },
+            json={
+                "model": self.model,
+                "query": query,
+                "items": items,
+                "label_token_ids": label_token_ids,
+                "apply_softmax": apply_softmax,
+                "item_first": item_first,
+            },
+        )
+        return response.json()
+
+    def test_score_text_input(self):
+        """Test scoring with text input"""
+        query = "The capital of France is"
+        items = ["Paris", "London", "Berlin"]
+
+        # Get valid token IDs from the tokenizer
+        label_token_ids = []
+        for item in items:
+            token_ids = self.tokenizer.encode(item, add_special_tokens=False)
+            if not token_ids:
+                self.fail(f"Failed to encode item: {item}")
+            label_token_ids.append(token_ids[0])
+
+        response = self.run_score(query, items, label_token_ids, apply_softmax=True)
+
+        # Handle error responses
+        if response.get("type") == "BadRequestError":
+            self.fail(f"Score request failed with error: {response['message']}")
+
+        # Verify response structure
+        self.assertIn("scores", response, "Response should have a 'scores' field")
+        self.assertIsInstance(response["scores"], list, "scores should be a list")
+        self.assertEqual(
+            len(response["scores"]),
+            len(items),
+            "Number of scores should match number of items",
+        )
+
+        # Each score should be a list of floats in the order of label_token_ids
+        for i, score_list in enumerate(response["scores"]):
+            self.assertIsInstance(score_list, list, f"Score {i} should be a list")
+            self.assertEqual(
+                len(score_list),
+                len(label_token_ids),
+                f"Score {i} length should match label_token_ids",
+            )
+            self.assertTrue(
+                all(isinstance(v, float) for v in score_list),
+                f"Score {i} values should be floats",
+            )
+            self.assertAlmostEqual(
+                sum(score_list),
+                1.0,
+                places=6,
+                msg=f"Score {i} probabilities should sum to 1",
+            )
+
+    def test_score_token_input(self):
+        """Test scoring with token IDs input"""
+        query = "The capital of France is"
+        items = ["Paris", "London", "Berlin"]
+
+        # Get valid token IDs
+        query_ids = self.tokenizer.encode(query, add_special_tokens=False)
+        item_ids = [
+            self.tokenizer.encode(item, add_special_tokens=False) for item in items
+        ]
+        label_token_ids = [
+            ids[0] for ids in item_ids if ids
+        ]  # Get first token ID of each item
+
+        response = self.run_score(
+            query_ids, item_ids, label_token_ids, apply_softmax=True
+        )
+
+        # Handle error responses
+        if response.get("type") == "BadRequestError":
+            self.fail(f"Score request failed with error: {response['message']}")
+
+        # Verify response structure
+        self.assertIn("scores", response, "Response should have a 'scores' field")
+        self.assertIsInstance(response["scores"], list, "scores should be a list")
+        self.assertEqual(
+            len(response["scores"]),
+            len(items),
+            "Number of scores should match number of items",
+        )
+
+        # Each score should be a list of floats in the order of label_token_ids
+        for i, score_list in enumerate(response["scores"]):
+            self.assertIsInstance(score_list, list, f"Score {i} should be a list")
+            self.assertEqual(
+                len(score_list),
+                len(label_token_ids),
+                f"Score {i} length should match label_token_ids",
+            )
+            self.assertTrue(
+                all(isinstance(v, float) for v in score_list),
+                f"Score {i} values should be floats",
+            )
+            self.assertAlmostEqual(
+                sum(score_list),
+                1.0,
+                places=6,
+                msg=f"Score {i} probabilities should sum to 1",
+            )
+
+    def test_score_error_handling(self):
+        """Test error handling for invalid inputs"""
+        query = "The capital of France is"
+        items = ["Paris", "London", "Berlin"]
+
+        # Test with invalid token ID
+        response = requests.post(
+            self.base_url,
+            headers={
+                "Authorization": f"Bearer {self.api_key}",
+                "Content-Type": "application/json",
+            },
+            json={
+                "model": self.model,
+                "query": query,
+                "items": items,
+                "label_token_ids": [999999],  # Invalid token ID
+                "apply_softmax": True,
+            },
+        )
+        self.assertEqual(response.status_code, 400)
+        error_response = response.json()
+        self.assertEqual(error_response["type"], "BadRequestError")
+        self.assertIn("Token ID 999999 is out of vocabulary", error_response["message"])
+
+
+if __name__ == "__main__":
+    unittest.main()

test/srt/openai_server/basic/test_protocol.py

@@ -0,0 +1,279 @@
+# Copyright 2023-2024 SGLang Team
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+"""Tests for OpenAI API protocol models"""
+
+import json
+import time
+import unittest
+from typing import Dict, List, Optional
+
+from pydantic import ValidationError
+
+from sglang.srt.entrypoints.openai.protocol import (
+    BatchRequest,
+    BatchResponse,
+    ChatCompletionMessageContentImagePart,
+    ChatCompletionMessageContentTextPart,
+    ChatCompletionRequest,
+    ChatCompletionResponse,
+    ChatCompletionResponseChoice,
+    ChatCompletionResponseStreamChoice,
+    ChatCompletionStreamResponse,
+    ChatCompletionTokenLogprob,
+    ChatMessage,
+    ChoiceLogprobs,
+    CompletionRequest,
+    CompletionResponse,
+    CompletionResponseChoice,
+    DeltaMessage,
+    EmbeddingObject,
+    EmbeddingRequest,
+    EmbeddingResponse,
+    ErrorResponse,
+    FileDeleteResponse,
+    FileRequest,
+    FileResponse,
+    Function,
+    FunctionResponse,
+    JsonSchemaResponseFormat,
+    LogProbs,
+    ModelCard,
+    ModelList,
+    MultimodalEmbeddingInput,
+    ResponseFormat,
+    ScoringRequest,
+    ScoringResponse,
+    StreamOptions,
+    StructuralTagResponseFormat,
+    Tool,
+    ToolCall,
+    ToolChoice,
+    TopLogprob,
+    UsageInfo,
+)
+
+
+class TestModelCard(unittest.TestCase):
+    """Test ModelCard protocol model"""
+
+    def test_model_card_serialization(self):
+        """Test model card JSON serialization"""
+        card = ModelCard(id="test-model", max_model_len=4096)
+        data = card.model_dump()
+        self.assertEqual(data["id"], "test-model")
+        self.assertEqual(data["object"], "model")
+        self.assertEqual(data["max_model_len"], 4096)
+
+
+class TestModelList(unittest.TestCase):
+    """Test ModelList protocol model"""
+
+    def test_empty_model_list(self):
+        """Test empty model list creation"""
+        model_list = ModelList()
+        self.assertEqual(model_list.object, "list")
+        self.assertEqual(len(model_list.data), 0)
+
+    def test_model_list_with_cards(self):
+        """Test model list with model cards"""
+        cards = [
+            ModelCard(id="model-1"),
+            ModelCard(id="model-2", max_model_len=2048),
+        ]
+        model_list = ModelList(data=cards)
+        self.assertEqual(len(model_list.data), 2)
+        self.assertEqual(model_list.data[0].id, "model-1")
+        self.assertEqual(model_list.data[1].id, "model-2")
+
+
+class TestCompletionRequest(unittest.TestCase):
+    """Test CompletionRequest protocol model"""
+
+    def test_basic_completion_request(self):
+        """Test basic completion request"""
+        request = CompletionRequest(model="test-model", prompt="Hello world")
+        self.assertEqual(request.model, "test-model")
+        self.assertEqual(request.prompt, "Hello world")
+        self.assertEqual(request.max_tokens, 16)  # default
+        self.assertEqual(request.temperature, 1.0)  # default
+        self.assertEqual(request.n, 1)  # default
+        self.assertFalse(request.stream)  # default
+        self.assertFalse(request.echo)  # default
+
+    def test_completion_request_sglang_extensions(self):
+        """Test completion request with SGLang-specific extensions"""
+        request = CompletionRequest(
+            model="test-model",
+            prompt="Hello",
+            top_k=50,
+            min_p=0.1,
+            repetition_penalty=1.1,
+            regex=r"\d+",
+            json_schema='{"type": "object"}',
+            lora_path="/path/to/lora",
+        )
+        self.assertEqual(request.top_k, 50)
+        self.assertEqual(request.min_p, 0.1)
+        self.assertEqual(request.repetition_penalty, 1.1)
+        self.assertEqual(request.regex, r"\d+")
+        self.assertEqual(request.json_schema, '{"type": "object"}')
+        self.assertEqual(request.lora_path, "/path/to/lora")
+
+    def test_completion_request_validation_errors(self):
+        """Test completion request validation errors"""
+        with self.assertRaises(ValidationError):
+            CompletionRequest()  # missing required fields
+
+        with self.assertRaises(ValidationError):
+            CompletionRequest(model="test-model")  # missing prompt
+
+
+class TestChatCompletionRequest(unittest.TestCase):
+    """Test ChatCompletionRequest protocol model"""
+
+    def test_basic_chat_completion_request(self):
+        """Test basic chat completion request"""
+        messages = [{"role": "user", "content": "Hello"}]
+        request = ChatCompletionRequest(model="test-model", messages=messages)
+        self.assertEqual(request.model, "test-model")
+        self.assertEqual(len(request.messages), 1)
+        self.assertEqual(request.messages[0].role, "user")
+        self.assertEqual(request.messages[0].content, "Hello")
+        self.assertEqual(request.temperature, 0.7)  # default
+        self.assertFalse(request.stream)  # default
+        self.assertEqual(request.tool_choice, "none")  # default when no tools
+
+    def test_chat_completion_tool_choice_validation(self):
+        """Test tool choice validation logic"""
+        messages = [{"role": "user", "content": "Hello"}]
+
+        # No tools, tool_choice should default to "none"
+        request1 = ChatCompletionRequest(model="test-model", messages=messages)
+        self.assertEqual(request1.tool_choice, "none")
+
+        # With tools, tool_choice should default to "auto"
+        tools = [
+            {
+                "type": "function",
+                "function": {"name": "test_func", "description": "Test function"},
+            }
+        ]
+        request2 = ChatCompletionRequest(
+            model="test-model", messages=messages, tools=tools
+        )
+        self.assertEqual(request2.tool_choice, "auto")
+
+    def test_chat_completion_sglang_extensions(self):
+        """Test chat completion with SGLang extensions"""
+        messages = [{"role": "user", "content": "Hello"}]
+        request = ChatCompletionRequest(
+            model="test-model",
+            messages=messages,
+            top_k=40,
+            min_p=0.05,
+            separate_reasoning=False,
+            stream_reasoning=False,
+            chat_template_kwargs={"custom_param": "value"},
+        )
+        self.assertEqual(request.top_k, 40)
+        self.assertEqual(request.min_p, 0.05)
+        self.assertFalse(request.separate_reasoning)
+        self.assertFalse(request.stream_reasoning)
+        self.assertEqual(request.chat_template_kwargs, {"custom_param": "value"})
+
+
+class TestModelSerialization(unittest.TestCase):
+    """Test model serialization with hidden states"""
+
+    def test_hidden_states_excluded_when_none(self):
+        """Test that None hidden_states are excluded with exclude_none=True"""
+        choice = ChatCompletionResponseChoice(
+            index=0,
+            message=ChatMessage(role="assistant", content="Hello"),
+            finish_reason="stop",
+            hidden_states=None,
+        )
+
+        response = ChatCompletionResponse(
+            id="test-id",
+            model="test-model",
+            choices=[choice],
+            usage=UsageInfo(prompt_tokens=5, completion_tokens=1, total_tokens=6),
+        )
+
+        # Test exclude_none serialization (should exclude None hidden_states)
+        data = response.model_dump(exclude_none=True)
+        self.assertNotIn("hidden_states", data["choices"][0])
+
+    def test_hidden_states_included_when_not_none(self):
+        """Test that non-None hidden_states are included"""
+        choice = ChatCompletionResponseChoice(
+            index=0,
+            message=ChatMessage(role="assistant", content="Hello"),
+            finish_reason="stop",
+            hidden_states=[0.1, 0.2, 0.3],
+        )
+
+        response = ChatCompletionResponse(
+            id="test-id",
+            model="test-model",
+            choices=[choice],
+            usage=UsageInfo(prompt_tokens=5, completion_tokens=1, total_tokens=6),
+        )
+
+        # Test exclude_none serialization (should include non-None hidden_states)
+        data = response.model_dump(exclude_none=True)
+        self.assertIn("hidden_states", data["choices"][0])
+        self.assertEqual(data["choices"][0]["hidden_states"], [0.1, 0.2, 0.3])
+
+
+class TestValidationEdgeCases(unittest.TestCase):
+    """Test edge cases and validation scenarios"""
+
+    def test_invalid_tool_choice_type(self):
+        """Test invalid tool choice type"""
+        messages = [{"role": "user", "content": "Hello"}]
+        with self.assertRaises(ValidationError):
+            ChatCompletionRequest(
+                model="test-model", messages=messages, tool_choice=123
+            )
+
+    def test_negative_token_limits(self):
+        """Test negative token limits"""
+        with self.assertRaises(ValidationError):
+            CompletionRequest(model="test-model", prompt="Hello", max_tokens=-1)
+
+    def test_model_serialization_roundtrip(self):
+        """Test that models can be serialized and deserialized"""
+        original_request = ChatCompletionRequest(
+            model="test-model",
+            messages=[{"role": "user", "content": "Hello"}],
+            temperature=0.7,
+            max_tokens=100,
+        )
+
+        # Serialize to dict
+        data = original_request.model_dump()
+
+        # Deserialize back
+        restored_request = ChatCompletionRequest(**data)
+
+        self.assertEqual(restored_request.model, original_request.model)
+        self.assertEqual(restored_request.temperature, original_request.temperature)
+        self.assertEqual(restored_request.max_tokens, original_request.max_tokens)
+        self.assertEqual(len(restored_request.messages), len(original_request.messages))
+
+
+if __name__ == "__main__":
+    unittest.main(verbosity=2)

test/srt/openai_server/basic/test_serving_chat.py

@@ -0,0 +1,146 @@
+"""
+Unit-tests for OpenAIServingChat — rewritten to use only the std-lib 'unittest'.
+Run with either:
+    python tests/test_serving_chat_unit.py -v
+or
+    python -m unittest discover -s tests -p "test_*unit.py" -v
+"""
+
+import unittest
+import uuid
+from typing import Optional
+from unittest.mock import Mock, patch
+
+from fastapi import Request
+
+from sglang.srt.entrypoints.openai.protocol import ChatCompletionRequest
+from sglang.srt.entrypoints.openai.serving_chat import OpenAIServingChat
+from sglang.srt.managers.io_struct import GenerateReqInput
+
+
+class _MockTokenizerManager:
+    """Minimal mock that satisfies OpenAIServingChat."""
+
+    def __init__(self):
+        self.model_config = Mock(is_multimodal=False)
+        self.server_args = Mock(
+            enable_cache_report=False,
+            tool_call_parser="hermes",
+            reasoning_parser=None,
+        )
+        self.chat_template_name: Optional[str] = "llama-3"
+
+        # tokenizer stub
+        self.tokenizer = Mock()
+        self.tokenizer.encode.return_value = [1, 2, 3, 4, 5]
+        self.tokenizer.decode.return_value = "Test response"
+        self.tokenizer.chat_template = None
+        self.tokenizer.bos_token_id = 1
+
+        # async generator stub for generate_request
+        async def _mock_generate():
+            yield {
+                "text": "Test response",
+                "meta_info": {
+                    "id": f"chatcmpl-{uuid.uuid4()}",
+                    "prompt_tokens": 10,
+                    "completion_tokens": 5,
+                    "cached_tokens": 0,
+                    "finish_reason": {"type": "stop", "matched": None},
+                    "output_token_logprobs": [(0.1, 1, "Test"), (0.2, 2, "response")],
+                    "output_top_logprobs": None,
+                },
+                "index": 0,
+            }
+
+        self.generate_request = Mock(return_value=_mock_generate())
+        self.create_abort_task = Mock()
+
+
+class _MockTemplateManager:
+    """Minimal mock for TemplateManager."""
+
+    def __init__(self):
+        self.chat_template_name: Optional[str] = "llama-3"
+        self.jinja_template_content_format: Optional[str] = None
+        self.completion_template_name: Optional[str] = None
+
+
+class ServingChatTestCase(unittest.TestCase):
+    # ------------- common fixtures -------------
+    def setUp(self):
+        self.tm = _MockTokenizerManager()
+        self.template_manager = _MockTemplateManager()
+        self.chat = OpenAIServingChat(self.tm, self.template_manager)
+
+        # frequently reused requests
+        self.basic_req = ChatCompletionRequest(
+            model="x",
+            messages=[{"role": "user", "content": "Hi?"}],
+            temperature=0.7,
+            max_tokens=100,
+            stream=False,
+        )
+        self.stream_req = ChatCompletionRequest(
+            model="x",
+            messages=[{"role": "user", "content": "Hi?"}],
+            temperature=0.7,
+            max_tokens=100,
+            stream=True,
+        )
+
+        self.fastapi_request = Mock(spec=Request)
+        self.fastapi_request.headers = {}
+
+    # ------------- conversion tests -------------
+    def test_convert_to_internal_request_single(self):
+        with patch(
+            "sglang.srt.entrypoints.openai.serving_chat.generate_chat_conv"
+        ) as conv_mock, patch.object(self.chat, "_process_messages") as proc_mock:
+            conv_ins = Mock()
+            conv_ins.get_prompt.return_value = "Test prompt"
+            conv_ins.image_data = conv_ins.audio_data = None
+            conv_ins.modalities = []
+            conv_ins.stop_str = ["</s>"]
+            conv_mock.return_value = conv_ins
+
+            proc_mock.return_value = (
+                "Test prompt",
+                [1, 2, 3],
+                None,
+                None,
+                [],
+                ["</s>"],
+                None,
+            )
+
+            adapted, processed = self.chat._convert_to_internal_request(self.basic_req)
+            self.assertIsInstance(adapted, GenerateReqInput)
+            self.assertFalse(adapted.stream)
+            self.assertEqual(processed, self.basic_req)
+
+    # ------------- sampling-params -------------
+    def test_sampling_param_build(self):
+        req = ChatCompletionRequest(
+            model="x",
+            messages=[{"role": "user", "content": "Hi"}],
+            temperature=0.8,
+            max_tokens=150,
+            min_tokens=5,
+            top_p=0.9,
+            stop=["</s>"],
+        )
+        with patch.object(
+            self.chat,
+            "_process_messages",
+            return_value=("Prompt", [1], None, None, [], ["</s>"], None),
+        ):
+            params = self.chat._build_sampling_params(req, ["</s>"], None)
+            self.assertEqual(params["temperature"], 0.8)
+            self.assertEqual(params["max_new_tokens"], 150)
+            self.assertEqual(params["min_new_tokens"], 5)
+            self.assertEqual(params["stop"], ["</s>"])
+
+
+if __name__ == "__main__":
+    unittest.main(verbosity=2)

test/srt/openai_server/basic/test_serving_completions.py

@@ -0,0 +1,100 @@
+"""
+Unit-tests for the refactored completions-serving handler (no pytest).
+Run with:
+    python -m unittest tests.test_serving_completions_unit -v
+"""
+
+import unittest
+from typing import Optional
+from unittest.mock import AsyncMock, Mock, patch
+
+from sglang.srt.entrypoints.openai.protocol import CompletionRequest
+from sglang.srt.entrypoints.openai.serving_completions import OpenAIServingCompletion
+from sglang.srt.managers.tokenizer_manager import TokenizerManager
+
+
+class _MockTemplateManager:
+    """Minimal mock for TemplateManager."""
+
+    def __init__(self):
+        self.chat_template_name: Optional[str] = None
+        self.jinja_template_content_format: Optional[str] = None
+        self.completion_template_name: Optional[str] = (
+            None  # Set to None to avoid template processing
+        )
+
+
+class ServingCompletionTestCase(unittest.TestCase):
+    """Bundle all prompt/echo tests in one TestCase."""
+
+    # ---------- shared test fixtures ----------
+    def setUp(self):
+        # build the mock TokenizerManager once for every test
+        tm = Mock(spec=TokenizerManager)
+
+        tm.tokenizer = Mock()
+        tm.tokenizer.encode.return_value = [1, 2, 3, 4]
+        tm.tokenizer.decode.return_value = "decoded text"
+        tm.tokenizer.bos_token_id = 1
+
+        tm.model_config = Mock(is_multimodal=False)
+        tm.server_args = Mock(enable_cache_report=False)
+
+        tm.generate_request = AsyncMock()
+        tm.create_abort_task = Mock()
+
+        self.template_manager = _MockTemplateManager()
+        self.sc = OpenAIServingCompletion(tm, self.template_manager)
+
+    # ---------- prompt-handling ----------
+    def test_single_string_prompt(self):
+        req = CompletionRequest(model="x", prompt="Hello world", max_tokens=100)
+        internal, _ = self.sc._convert_to_internal_request(req)
+        self.assertEqual(internal.text, "Hello world")
+
+    def test_single_token_ids_prompt(self):
+        req = CompletionRequest(model="x", prompt=[1, 2, 3, 4], max_tokens=100)
+        internal, _ = self.sc._convert_to_internal_request(req)
+        self.assertEqual(internal.input_ids, [1, 2, 3, 4])
+
+    # ---------- echo-handling ----------
+    def test_echo_with_string_prompt_streaming(self):
+        req = CompletionRequest(model="x", prompt="Hello", max_tokens=1, echo=True)
+        self.assertEqual(self.sc._get_echo_text(req, 0), "Hello")
+
+    def test_echo_with_list_of_strings_streaming(self):
+        req = CompletionRequest(
+            model="x", prompt=["A", "B"], max_tokens=1, echo=True, n=1
+        )
+        self.assertEqual(self.sc._get_echo_text(req, 0), "A")
+        self.assertEqual(self.sc._get_echo_text(req, 1), "B")
+
+    def test_echo_with_token_ids_streaming(self):
+        req = CompletionRequest(model="x", prompt=[1, 2, 3], max_tokens=1, echo=True)
+        self.sc.tokenizer_manager.tokenizer.decode.return_value = "decoded_prompt"
+        self.assertEqual(self.sc._get_echo_text(req, 0), "decoded_prompt")
+
+    def test_echo_with_multiple_token_ids_streaming(self):
+        req = CompletionRequest(
+            model="x", prompt=[[1, 2], [3, 4]], max_tokens=1, echo=True, n=1
+        )
+        self.sc.tokenizer_manager.tokenizer.decode.return_value = "decoded"
+        self.assertEqual(self.sc._get_echo_text(req, 0), "decoded")
+
+    def test_prepare_echo_prompts_non_streaming(self):
+        # single string
+        req = CompletionRequest(model="x", prompt="Hi", echo=True)
+        self.assertEqual(self.sc._prepare_echo_prompts(req), ["Hi"])
+
+        # list of strings
+        req = CompletionRequest(model="x", prompt=["Hi", "Yo"], echo=True)
+        self.assertEqual(self.sc._prepare_echo_prompts(req), ["Hi", "Yo"])
+
+        # token IDs
+        req = CompletionRequest(model="x", prompt=[1, 2, 3], echo=True)
+        self.sc.tokenizer_manager.tokenizer.decode.return_value = "decoded"
+        self.assertEqual(self.sc._prepare_echo_prompts(req), ["decoded"])
+
+
+if __name__ == "__main__":
+    unittest.main(verbosity=2)

test/srt/openai_server/basic/test_serving_embedding.py

@@ -0,0 +1,145 @@
+"""
+Unit tests for the OpenAIServingEmbedding class from serving_embedding.py.
+"""
+
+import unittest
+import uuid
+from unittest.mock import Mock
+
+from fastapi import Request
+
+from sglang.srt.entrypoints.openai.protocol import (
+    EmbeddingRequest,
+    EmbeddingResponse,
+    MultimodalEmbeddingInput,
+)
+from sglang.srt.entrypoints.openai.serving_embedding import OpenAIServingEmbedding
+from sglang.srt.managers.io_struct import EmbeddingReqInput
+
+
+# Mock TokenizerManager for embedding tests
+class _MockTokenizerManager:
+    def __init__(self):
+        self.model_config = Mock()
+        self.model_config.is_multimodal = False
+        self.server_args = Mock()
+        self.server_args.enable_cache_report = False
+        self.model_path = "test-model"
+
+        # Mock tokenizer
+        self.tokenizer = Mock()
+        self.tokenizer.encode = Mock(return_value=[1, 2, 3, 4, 5])
+        self.tokenizer.decode = Mock(return_value="Test embedding input")
+        self.tokenizer.chat_template = None
+        self.tokenizer.bos_token_id = 1
+
+        # Mock generate_request method for embeddings
+        async def mock_generate_embedding():
+            yield {
+                "embedding": [0.1, 0.2, 0.3, 0.4, 0.5] * 20,  # 100-dim embedding
+                "meta_info": {
+                    "id": f"embd-{uuid.uuid4()}",
+                    "prompt_tokens": 5,
+                },
+            }
+
+        self.generate_request = Mock(return_value=mock_generate_embedding())
+
+
+# Mock TemplateManager for embedding tests
+class _MockTemplateManager:
+    def __init__(self):
+        self.chat_template_name = None  # None for embeddings usually
+        self.jinja_template_content_format = None
+        self.completion_template_name = None
+
+
+class ServingEmbeddingTestCase(unittest.TestCase):
+    def setUp(self):
+        """Set up test fixtures."""
+        self.tokenizer_manager = _MockTokenizerManager()
+        self.template_manager = _MockTemplateManager()
+        self.serving_embedding = OpenAIServingEmbedding(
+            self.tokenizer_manager, self.template_manager
+        )
+
+        self.request = Mock(spec=Request)
+        self.request.headers = {}
+
+        self.basic_req = EmbeddingRequest(
+            model="test-model",
+            input="Hello, how are you?",
+            encoding_format="float",
+        )
+        self.list_req = EmbeddingRequest(
+            model="test-model",
+            input=["Hello, how are you?", "I am fine, thank you!"],
+            encoding_format="float",
+        )
+        self.multimodal_req = EmbeddingRequest(
+            model="test-model",
+            input=[
+                MultimodalEmbeddingInput(text="Hello", image="base64_image_data"),
+                MultimodalEmbeddingInput(text="World", image=None),
+            ],
+            encoding_format="float",
+        )
+        self.token_ids_req = EmbeddingRequest(
+            model="test-model",
+            input=[1, 2, 3, 4, 5],
+            encoding_format="float",
+        )
+
+    def test_convert_single_string_request(self):
+        """Test converting single string request to internal format."""
+        adapted_request, processed_request = (
+            self.serving_embedding._convert_to_internal_request(self.basic_req)
+        )
+
+        self.assertIsInstance(adapted_request, EmbeddingReqInput)
+        self.assertEqual(adapted_request.text, "Hello, how are you?")
+        # self.assertEqual(adapted_request.rid, "test-id")
+        self.assertEqual(processed_request, self.basic_req)
+
+    def test_convert_list_string_request(self):
+        """Test converting list of strings request to internal format."""
+        adapted_request, processed_request = (
+            self.serving_embedding._convert_to_internal_request(self.list_req)
+        )
+
+        self.assertIsInstance(adapted_request, EmbeddingReqInput)
+        self.assertEqual(
+            adapted_request.text, ["Hello, how are you?", "I am fine, thank you!"]
+        )
+        # self.assertEqual(adapted_request.rid, "test-id")
+        self.assertEqual(processed_request, self.list_req)
+
+    def test_convert_token_ids_request(self):
+        """Test converting token IDs request to internal format."""
+        adapted_request, processed_request = (
+            self.serving_embedding._convert_to_internal_request(self.token_ids_req)
+        )
+
+        self.assertIsInstance(adapted_request, EmbeddingReqInput)
+        self.assertEqual(adapted_request.input_ids, [1, 2, 3, 4, 5])
+        # self.assertEqual(adapted_request.rid, "test-id")
+        self.assertEqual(processed_request, self.token_ids_req)
+
+    def test_convert_multimodal_request(self):
+        """Test converting multimodal request to internal format."""
+        adapted_request, processed_request = (
+            self.serving_embedding._convert_to_internal_request(self.multimodal_req)
+        )
+
+        self.assertIsInstance(adapted_request, EmbeddingReqInput)
+        # Should extract text and images separately
+        self.assertEqual(len(adapted_request.text), 2)
+        self.assertIn("Hello", adapted_request.text)
+        self.assertIn("World", adapted_request.text)
+        self.assertEqual(adapted_request.image_data[0], "base64_image_data")
+        self.assertIsNone(adapted_request.image_data[1])
+        # self.assertEqual(adapted_request.rid, "test-id")
+
+
+if __name__ == "__main__":
+    unittest.main(verbosity=2)