[Feature] Adds basic support for image content in OpenAI chat routes (#113)

python/sglang/test/test_conversation.py

@@ -0,0 +1,46 @@
+from sglang.srt.conversation import generate_chat_conv
+from sglang.srt.managers.openai_protocol import (
+    ChatCompletionMessageGenericParam,
+    ChatCompletionMessageContentImagePart,
+    ChatCompletionMessageContentImageURL,
+    ChatCompletionMessageContentTextPart,
+    ChatCompletionMessageUserParam,
+    ChatCompletionRequest,
+)
+
+
+def test_chat_completion_to_conv_image():
+    """Test that we can convert a chat image request to a convo"""
+    request = ChatCompletionRequest(
+        model="default",
+        messages=[
+            ChatCompletionMessageGenericParam(
+                role="system", content="You are a helpful AI assistant"
+            ),
+            ChatCompletionMessageUserParam(
+                role="user",
+                content=[
+                    ChatCompletionMessageContentTextPart(
+                        type="text", text="Describe this image"
+                    ),
+                    ChatCompletionMessageContentImagePart(
+                        type="image_url",
+                        image_url=ChatCompletionMessageContentImageURL(
+                            url="https://someurl.com"
+                        ),
+                    ),
+                ],
+            ),
+        ],
+    )
+    conv = generate_chat_conv(request, "vicuna_v1.1")
+    assert conv.messages == [
+        ["USER", "Describe this image<image>"],
+        ["ASSISTANT", None],
+    ]
+    assert conv.system_message == "You are a helpful AI assistant"
+    assert conv.image_data == ["https://someurl.com"]
+
+
+if __name__ == "__main__":
+    test_chat_completion_to_conv_image()

python/sglang/test/test_openai_protocol.py

@@ -0,0 +1,51 @@
+from sglang.srt.managers.openai_protocol import (
+    ChatCompletionMessageGenericParam,
+    ChatCompletionMessageContentImagePart,
+    ChatCompletionMessageContentImageURL,
+    ChatCompletionMessageContentTextPart,
+    ChatCompletionMessageUserParam,
+    ChatCompletionRequest,
+)
+
+
+def test_chat_completion_request_image():
+    """Test that Chat Completion Requests with images can be converted."""
+
+    image_request = {
+        "model": "default",
+        "messages": [
+            {"role": "system", "content": "You are a helpful AI assistant"},
+            {
+                "role": "user",
+                "content": [
+                    {"type": "text", "text": "Describe this image"},
+                    {"type": "image_url", "image_url": {"url": "https://someurl.com"}},
+                ],
+            },
+        ],
+        "temperature": 0,
+        "max_tokens": 64,
+    }
+    request = ChatCompletionRequest(**image_request)
+    assert len(request.messages) == 2
+    assert request.messages[0] == ChatCompletionMessageGenericParam(
+        role="system", content="You are a helpful AI assistant"
+    )
+    assert request.messages[1] == ChatCompletionMessageUserParam(
+        role="user",
+        content=[
+            ChatCompletionMessageContentTextPart(
+                type="text", text="Describe this image"
+            ),
+            ChatCompletionMessageContentImagePart(
+                type="image_url",
+                image_url=ChatCompletionMessageContentImageURL(
+                    url="https://someurl.com"
+                ),
+            ),
+        ],
+    )
+
+
+if __name__ == "__main__":
+    test_chat_completion_request_image()