OpenAI Vision API Client#

Source vllm-project/vllm.

 1"""An example showing how to use vLLM to serve VLMs.
 2
 3Launch the vLLM server with the following command:
 4vllm serve llava-hf/llava-1.5-7b-hf --chat-template template_llava.jinja
 5"""
 6import base64
 7
 8import requests
 9from openai import OpenAI
10
11# Modify OpenAI's API key and API base to use vLLM's API server.
12openai_api_key = "EMPTY"
13openai_api_base = "http://localhost:8000/v1"
14
15client = OpenAI(
16    # defaults to os.environ.get("OPENAI_API_KEY")
17    api_key=openai_api_key,
18    base_url=openai_api_base,
19)
20
21models = client.models.list()
22model = models.data[0].id
23
24image_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"
25
26# Use image url in the payload
27chat_completion_from_url = client.chat.completions.create(
28    messages=[{
29        "role":
30        "user",
31        "content": [
32            {
33                "type": "text",
34                "text": "What’s in this image?"
35            },
36            {
37                "type": "image_url",
38                "image_url": {
39                    "url": image_url
40                },
41            },
42        ],
43    }],
44    model=model,
45    max_tokens=64,
46)
47
48result = chat_completion_from_url.choices[0].message.content
49print(f"Chat completion output:{result}")
50
51
52# Use base64 encoded image in the payload
53def encode_image_base64_from_url(image_url: str) -> str:
54    """Encode an image retrieved from a remote url to base64 format."""
55
56    with requests.get(image_url) as response:
57        response.raise_for_status()
58        result = base64.b64encode(response.content).decode('utf-8')
59
60    return result
61
62
63image_base64 = encode_image_base64_from_url(image_url=image_url)
64chat_completion_from_base64 = client.chat.completions.create(
65    messages=[{
66        "role":
67        "user",
68        "content": [
69            {
70                "type": "text",
71                "text": "What’s in this image?"
72            },
73            {
74                "type": "image_url",
75                "image_url": {
76                    "url": f"data:image/jpeg;base64,{image_base64}"
77                },
78            },
79        ],
80    }],
81    model=model,
82    max_tokens=64,
83)
84
85result = chat_completion_from_base64.choices[0].message.content
86print(f"Chat completion output:{result}")