How to query vision models

Scaleway’s Generative APIs service allows users to interact with powerful vision models hosted on the platform.

There are several ways to interact with vision models:

• The Scaleway console provides a complete playground, aiming to test models, adapt parameters, and observe how these changes affect the output in real-time.
• Via the Chat API

Before you start

To complete the actions presented below, you must have:

• A Scaleway account logged into the console
• Owner status or IAM permissions allowing you to perform actions in the intended Organization
• A valid API key for API authentication
• Python 3.7+ installed on your system

Accessing the playground

Scaleway provides a web playground for vision models hosted on Generative APIs.

1. Navigate to Generative APIs under the AI section of the Scaleway console side menu. The list of models you can query displays.
2. Click the name of the vision model you want to try. Alternatively, click «See more Icon» next to the vision model, and click Try model in the menu.

The web playground displays.

Using the playground

1. Upload one or multiple images to the prompt area at the bottom of the page. Enter a prompt, for example, to describe the image(s) you attached.
2. Edit the hyperparameters listed on the right column, for example the default temperature for more or less randomness on the outputs.
3. Switch models at the top of the page, to observe the capabilities of chat and vision models offered via Generative APIs.
4. Click View code to get code snippets configured according to your settings in the playground.

Querying vision models via the API

The Chat API is an OpenAI-compatible REST API for generating and manipulating conversations.

You can query the vision models programmatically using your favorite tools or languages. Vision models take both text and images as inputs.

In the following example, we will use the OpenAI Python client.

Installing the OpenAI SDK

Install the OpenAI SDK using pip:

pip install openai

Initializing the client

Initialize the OpenAI client with your base URL and API key:

from openai import OpenAI
# Initialize the client with your base URL and API key
client = OpenAI(
    base_url="https://api.scaleway.ai/v1",  # Scaleway's Generative APIs service URL
    api_key="<SCW_SECRET_KEY>"  # Your unique API secret key from Scaleway
)

Generating a chat completion

You can now create a chat completion, for example with the pixtral-12b-2409 model:

# Create a chat completion using the 'pixtral-12b-2409' model
response = client.chat.completions.create(
    model="pixtral-12b-2409",
    messages=[
      {
        "role": "user", 
        "content": [
              {"type": "text", "text": "What is this image?"},
              {"type": "image_url", "image_url": {"url": "https://picsum.photos/id/32/512/512"}},
          ] #  Vision models will take a content array with text and image_url objects.
      }
    ],
    temperature=0.7,  # Adjusts creativity
    max_tokens=2048,   # Limits the length of the output
    top_p=0.9         # Controls diversity through nucleus sampling. You usually only need to use temperature.
)
# Print the generated response
print(response.choices[0].message.content)

This code sends messages, prompts and images, to the vision model and returns an answer based on your input. The temperature, max_tokens, and top_p parameters control the response’s creativity, length, and diversity, respectively.

A conversation style may include a default system prompt. You may set this prompt by setting the first message with the role system. For example:

[
  {
  	"role": "system",
  	"content": "You are Xavier Niel."
  }
]

Passing images to Pixtral

1. Image URLs: If the image is available online, you can just include the image URL in your request as demonstrated above. This approach is simple and does not require any encoding.
2. Base64 encoded: image Base64 encoding is a standard way to transform binary data, like images, into a text format, making it easier to transmit over the internet.

The following Python code sample shows you how to encode an image in base64 format and pass it to your request payload.

import base64
from io import BytesIO
from PIL import Image
def encode_image(img):
    buffered = BytesIO()
    img.save(buffered, format="JPEG")
    encoded_string = base64.b64encode(buffered.getvalue()).decode("utf-8")
    return encoded_string
img = Image.open("path_to_your_image.jpg")
base64_img = encode_image(img)
payload = {
    "messages": [
        {
            "role": "user",
            "content": [
                {
                    "type": "text", 
                    "text": "What is this image?"
                },
                {
                    "type": "image_url",
                    "image_url": {
                        "url": f"data:image/jpeg;base64,{base64_img}"
                    }
                }
            ]
        }
    ],
    ... # other parameters
}

Model parameters and their effects

The following parameters will influence the output of the model:

• messages: A list of message objects that represent the conversation history. Each message should have a role (e.g., “system”, “user”, “assistant”) and content. The content is an array that can contain text and/or image objects.
• temperature: Controls the output’s randomness. Lower values (e.g., 0.2) make the output more deterministic, while higher values (e.g., 0.8) make it more creative.
• max_tokens: The maximum number of tokens (words or parts of words) in the generated output.
• top_p: Recommended for advanced use cases only. You usually only need to use temperature. top_p controls the diversity of the output, using nucleus sampling, where the model considers the tokens with top probabilities until the cumulative probability reaches top_p.
• stop: A string or list of strings where the model will stop generating further tokens. This is useful for controlling the end of the output.

Streaming

By default, the outputs are returned to the client only after the generation process is complete. However, a common alternative is to stream the results back to the client as they are generated. This is particularly useful in chat applications, where it allows the client to view the results incrementally as each token is produced. The following example shows how to use the chat completion API:

from openai import OpenAI
client = OpenAI(
    base_url="https://api.scaleway.ai/v1",  # Scaleway's Generative APIs service URL
    api_key="<SCW_API_KEY>"  # Your unique API key from Scaleway
)
response = client.chat.completions.create(
  model="pixtral-12b-2409",
  messages=[{
      "role": "user", 
      "content": [
          {"type": "text", "text": "What is this image?"},
          {"type": "image_url", "image_url": {"url": "https://picsum.photos/id/32/512/512"}},
      ]
  }],
  stream=True,
)
for chunk in response:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")

Async

The service also supports asynchronous mode for any chat completion.

import asyncio
from openai import AsyncOpenAI
client = AsyncOpenAI(
    base_url="https://api.scaleway.ai/v1",  # Scaleway's Generative APIs service URL
    api_key="<SCW_API_KEY>"  # Your unique API key from Scaleway
)
async def main():
    stream = await client.chat.completions.create(
        model="pixtral-12b-2409",
        messages=[{
            "role": "user", 
            "content": [
                {"type": "text", "text": "What is this image?"},
                {"type": "image_url", "image_url": {"url": "https://picsum.photos/id/32/512/512"}},
            ]
        }],
        stream=True,
    )
    async for chunk in stream:
        print(chunk.choices[0].delta.content, end="")
asyncio.run(main())