[Frontend] Support GPT-4V Chat Completions API

docs/source/models/supported_models.rst

@@ -89,8 +89,8 @@ Alongside each architecture, we include some popular models that use it.
     - ✅︎
   * - :code:`LlavaForConditionalGeneration`
     - LLaVA-1.5
-    - :code:`llava-hf/llava-1.5-7b-hf`\*, :code:`llava-hf/llava-1.5-13b-hf`\*, etc.
-    -
+    - :code:`llava-hf/llava-1.5-7b-hf`, :code:`llava-hf/llava-1.5-13b-hf`, etc.
+    - 
   * - :code:`MiniCPMForCausalLM`
     - MiniCPM
     - :code:`openbmb/MiniCPM-2B-sft-bf16`, :code:`openbmb/MiniCPM-2B-dpo-bf16`, etc.

docs/source/models/vlm.rst

@@ -54,3 +54,44 @@ For now, we only support a single image per text prompt. To pass an image to the
         print(generated_text)
 
 A code example can be found in `examples/llava_example.py <https://github.com/vllm-project/vllm/blob/main/examples/llava_example.py>`_.
+
+OpenAI-Compatible Server
+------------------------
+
+We support image inputs to the OpenAI Chat API, as described in `GPT-4 with Vision <https://platform.openai.com/docs/guides/vision>`_.
+
+Here is a simple example using the :code:`openai` package:
+
+.. code-block:: python
+
+    from openai import OpenAI
+
+    openai_api_key = "EMPTY"
+    openai_api_base = "http://localhost:8000/v1"
+
+    client = OpenAI(
+        api_key=openai_api_key,
+        base_url=openai_api_base,
+    )
+
+    # Note that this model expects the image to come before the main text
+    chat_response = client.chat.completions.create(
+        model="llava-hf/llava-1.5-7b-hf",
+        messages=[{
+            "role": "user",
+            "content": [
+                {
+                    "type": "image_url",
+                    "image_url": {
+                        "url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg",
+                    },
+                },
+                {"type": "text", "text": "What's in this image?"},
+            ],
+        }],
+    )
+    print("Chat response:", chat_response)
+
+.. note::
+
+    For now, we only support a single image per API call. Also, the ``detail`` parameter is ignored since it may not be applicable to other models.

examples/template_llava.jinja

@@ -0,0 +1,11 @@
+{%- for message in messages -%}
+    {{ message['role'].upper() + ': ' + message['content'] }}
+    {%- if (loop.last and add_generation_prompt) or not loop.last -%}
+        {{- '\n' -}}
+    {%- endif -%}
+{%- endfor -%}
+
+
+{%- if add_generation_prompt and messages[-1]['role'] != 'assistant' -%}
+    {{- 'ASSISTANT:' -}}
+{% endif %}

tests/entrypoints/test_openai_server.py

@@ -558,50 +558,52 @@ async def test_chat_streaming(server, client: openai.AsyncOpenAI,
 )
 async def test_batch_completions(server, client: openai.AsyncOpenAI,
                                  model_name: str):
-    # test simple list
-    batch = await client.completions.create(
-        model=model_name,
-        prompt=["Hello, my name is", "Hello, my name is"],
-        max_tokens=5,
-        temperature=0.0,
-    )
-    assert len(batch.choices) == 2
-    assert batch.choices[0].text == batch.choices[1].text
-
-    # test n = 2
-    batch = await client.completions.create(
-        model=model_name,
-        prompt=["Hello, my name is", "Hello, my name is"],
-        n=2,
-        max_tokens=5,
-        temperature=0.0,
-        extra_body=dict(
-            # NOTE: this has to be true for n > 1 in vLLM, but not necessary
-            # for official client.
-            use_beam_search=True),
-    )
-    assert len(batch.choices) == 4
-    assert batch.choices[0].text != batch.choices[
-        1].text, "beam search should be different"
-    assert batch.choices[0].text == batch.choices[
-        2].text, "two copies of the same prompt should be the same"
-    assert batch.choices[1].text == batch.choices[
-        3].text, "two copies of the same prompt should be the same"
-
-    # test streaming
-    batch = await client.completions.create(
-        model=model_name,
-        prompt=["Hello, my name is", "Hello, my name is"],
-        max_tokens=5,
-        temperature=0.0,
-        stream=True,
-    )
-    texts = [""] * 2
-    async for chunk in batch:
-        assert len(chunk.choices) == 1
-        choice = chunk.choices[0]
-        texts[choice.index] += choice.text
-    assert texts[0] == texts[1]
+    # test using text and token IDs
+    for prompts in (["Hello, my name is"] * 2, [[0, 0, 0, 0, 0]] * 2):
+        # test simple list
+        batch = await client.completions.create(
+            model=model_name,
+            prompt=prompts,
+            max_tokens=5,
+            temperature=0.0,
+        )
+        assert len(batch.choices) == 2
+        assert batch.choices[0].text == batch.choices[1].text
+
+        # test n = 2
+        batch = await client.completions.create(
+            model=model_name,
+            prompt=prompts,
+            n=2,
+            max_tokens=5,
+            temperature=0.0,
+            extra_body=dict(
+                # NOTE: this has to be true for n > 1 in vLLM, but not necessary
+                # for official client.
+                use_beam_search=True),
+        )
+        assert len(batch.choices) == 4
+        assert batch.choices[0].text != batch.choices[
+            1].text, "beam search should be different"
+        assert batch.choices[0].text == batch.choices[
+            2].text, "two copies of the same prompt should be the same"
+        assert batch.choices[1].text == batch.choices[
+            3].text, "two copies of the same prompt should be the same"
+
+        # test streaming
+        batch = await client.completions.create(
+            model=model_name,
+            prompt=prompts,
+            max_tokens=5,
+            temperature=0.0,
+            stream=True,
+        )
+        texts = [""] * 2
+        async for chunk in batch:
+            assert len(chunk.choices) == 1
+            choice = chunk.choices[0]
+            texts[choice.index] += choice.text
+        assert texts[0] == texts[1]
 
 
 @pytest.mark.asyncio
@@ -1047,7 +1049,7 @@ async def test_echo_logprob_completion(server, client: openai.AsyncOpenAI,
         prompt_text = tokenizer.decode(prompt) if isinstance(prompt,
                                                              list) else prompt
         assert (completion.choices[0].text is not None
-                and re.search(r"^" + prompt_text, completion.choices[0].text))
+                and completion.choices[0].text.startswith(prompt_text))
         logprobs = completion.choices[0].logprobs
         assert logprobs is not None
         assert len(logprobs.text_offset) > 5

tests/entrypoints/test_openai_server_vision.py

@@ -0,0 +1,176 @@
+from pathlib import Path
+
+import openai  # use the official client for correctness check
+import pytest
+# using Ray for overall ease of process management, parallel requests,
+# and debugging.
+import ray
+
+from ..utils import ServerRunner
+
+MODEL_NAME = "llava-hf/llava-1.5-7b-hf"
+CHAT_TEMPLATE = (Path(__file__).parent.parent.parent /
+                 "examples/template_llava.jinja")
+assert CHAT_TEMPLATE.exists()
+
+# Test different image extensions (JPG/PNG) and formats (gray/RGB/RGBA)
+TEST_IMAGE_URLS = [
+    "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg",
+    "https://upload.wikimedia.org/wikipedia/commons/f/fa/Grayscale_8bits_palette_sample_image.png",
+    "https://upload.wikimedia.org/wikipedia/commons/thumb/9/91/Venn_diagram_rgb.svg/1280px-Venn_diagram_rgb.svg.png",
+    "https://upload.wikimedia.org/wikipedia/commons/0/0b/RGBA_comp.png",
+]
+
+pytestmark = pytest.mark.openai
+
+
+@pytest.fixture(scope="module")
+def server():
+    ray.init()
+    server_runner = ServerRunner.remote([
+        "--model",
+        MODEL_NAME,
+        # use half precision for speed and memory savings in CI environment
+        "--dtype",
+        "bfloat16",
+        "--max-model-len",
+        "4096",
+        "--enforce-eager",
+        # vision language config below
+        "--image-input-type",
+        "pixel_values",
+        "--image-token-id",
+        "32000",
+        "--image-input-shape",
+        "1,3,336,336",
+        "--image-feature-size",
+        "576",
+        # chat template required for LLaVA
+        "--chat-template",
+        str(CHAT_TEMPLATE),
+    ])
+    ray.get(server_runner.ready.remote())
+    yield server_runner
+    ray.shutdown()
+
+
+@pytest.fixture(scope="session")
+def client():
+    client = openai.AsyncOpenAI(
+        base_url="http://localhost:8000/v1",
+        api_key="token-abc123",
+    )
+    yield client
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+@pytest.mark.parametrize("image_url", TEST_IMAGE_URLS)
+async def test_single_chat_session_image(server, client: openai.AsyncOpenAI,
+                                         model_name: str, image_url: str):
+    messages = [{
+        "role":
+        "user",
+        "content": [
+            {
+                "type": "image_url",
+                "image_url": {
+                    "url": image_url
+                }
+            },
+            {
+                "type": "text",
+                "text": "What's in this image?"
+            },
+        ],
+    }]
+
+    # test single completion
+    chat_completion = await client.chat.completions.create(model=model_name,
+                                                           messages=messages,
+                                                           max_tokens=10,
+                                                           logprobs=True,
+                                                           top_logprobs=5)
+    assert chat_completion.id is not None
+    assert len(chat_completion.choices) == 1
+
+    choice = chat_completion.choices[0]
+    assert choice.finish_reason == "length"
+    assert chat_completion.usage == openai.types.CompletionUsage(
+        completion_tokens=10, prompt_tokens=594, total_tokens=604)
+
+    message = choice.message
+    assert message.content is not None and len(message.content) >= 10
+    assert message.role == "assistant"
+    messages.append({"role": "assistant", "content": message.content})
+
+    # test multi-turn dialogue
+    messages.append({"role": "user", "content": "express your result in json"})
+    chat_completion = await client.chat.completions.create(
+        model=model_name,
+        messages=messages,
+        max_tokens=10,
+    )
+    message = chat_completion.choices[0].message
+    assert message.content is not None and len(message.content) >= 0
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("model_name", [MODEL_NAME])
+@pytest.mark.parametrize("image_url", TEST_IMAGE_URLS)
+async def test_chat_streaming_image(server, client: openai.AsyncOpenAI,
+                                    model_name: str, image_url: str):
+    messages = [{
+        "role":
+        "user",
+        "content": [
+            {
+                "type": "image_url",
+                "image_url": {
+                    "url": image_url
+                }
+            },
+            {
+                "type": "text",
+                "text": "What's in this image?"
+            },
+        ],
+    }]
+
+    # test single completion
+    chat_completion = await client.chat.completions.create(
+        model=model_name,
+        messages=messages,
+        max_tokens=10,
+        temperature=0.0,
+    )
+    output = chat_completion.choices[0].message.content
+    stop_reason = chat_completion.choices[0].finish_reason
+
+    # test streaming
+    stream = await client.chat.completions.create(
+        model=model_name,
+        messages=messages,
+        max_tokens=10,
+        temperature=0.0,
+        stream=True,
+    )
+    chunks = []
+    finish_reason_count = 0
+    async for chunk in stream:
+        delta = chunk.choices[0].delta
+        if delta.role:
+            assert delta.role == "assistant"
+        if delta.content:
+            chunks.append(delta.content)
+        if chunk.choices[0].finish_reason is not None:
+            finish_reason_count += 1
+    # finish reason should only return in last block
+    assert finish_reason_count == 1
+    assert chunk.choices[0].finish_reason == stop_reason
+    assert delta.content
+    assert "".join(chunks) == output
+
+
+if __name__ == "__main__":
+    pytest.main([__file__])