give bytes eyes

vision.py

# /// script
# requires-python = ">=3.12"
# dependencies = [
#     "httpx",
#     "rich",
# ]
# ///

# script require llama.cpp: > llama-server -hf ggml-org/SmolVLM-500M-Instruct-GGUF --n-predict -1

import base64
from pathlib import Path

from httpx import Client


def image_to_base64(path: Path) -> str:
    data = base64.b64encode(path.read_bytes()).decode("utf-8")
    return f"data:image/jpeg;base64,{data}"


def describe_image(image: Path) -> dict:
    image_b64 = image_to_base64(image)

    base_url = "http://localhost:8080"
    payload = {
        "model": "SmolVLM",  # not used
        "messages": [
            {"role": "system", "content": "Describe what you see in 5 sentences"},
            {
                "role": "user",
                "content": [{"type": "image_url", "image_url": {"url": image_b64}}],
            },
        ],
        "temperature": 0.0,
    }

    with Client(
        base_url=base_url,
        headers={"Content-Type": "application/json"},
        timeout=None,
    ) as requests:
        response = requests.post("/v1/chat/completions", json=payload)

    return response.json()


if __name__ == "__main__":
    from rich.console import Console
    from rich.markdown import Markdown

    beer = Path("beer.png")
    results = describe_image(beer)

    text = results["choices"][0]["message"]["content"]

    console = Console()
    console.print(Markdown(text))