README.md

---
title: Smol VLM 256m Instruct Docker
emoji: 📚
colorFrom: purple
colorTo: yellow
sdk: docker
pinned: false
short_description: Api endpoint for SMOL VLM 256M
---
# 🧠 SmolVLM-256M: Vision + Language Inference API

This Space demonstrates how to deploy and serve the **SmolVLM-256M-Instruct** multimodal language model using a Docker-based backend. The API provides OpenAI-style `chat/completions` endpoints for image + text understanding — similar to how ChatGPT Vision works.
Example frontend app could be found here: https://text-rec-api.glitch.me/

## 🚀 Docker Setup

This Space uses a custom Dockerfile that downloads and launches the SmolVLM model with vision support using [llama.cpp](https://github.com/ggerganov/llama.cpp).

### Dockerfile

```Dockerfile
FROM ghcr.io/ggml-org/llama.cpp:full

# Install wget
RUN apt update && apt install wget -y

# Download the GGUF model file
RUN wget "https://huggingface.co/ggml-org/SmolVLM-256M-Instruct-GGUF/resolve/main/SmolVLM-256M-Instruct-Q8_0.gguf" -O /smoll.gguf

# Download the mmproj (multimodal projection) file
RUN wget "https://huggingface.co/ggml-org/SmolVLM-256M-Instruct-GGUF/resolve/main/mmproj-SmolVLM-256M-Instruct-Q8_0.gguf" -O /mmproj.gguf

# Run the server on port 7860 with moderate generation settings
CMD [ "--server", "-m", "/smoll.gguf", "--mmproj", "/mmproj.gguf", "--port", "7860", "--host", "0.0.0.0", "-n", "512", "-t", "2" ]
```
## 🧠 API Usage

The server exposes a `POST /v1/chat/completions` endpoint compatible with the OpenAI API format.

### 🔁 Request Format

Send a JSON payload structured like this:

```json
{
  "model": "SmolVLM-256M-Instruct",
  "messages": [
    {
      "role": "user",
      "content": [
        { "type": "text", "text": "What is in this image?" },
        {
          "type": "image_url",
          "image_url": {
            "url": "data:image/jpeg;base64,/9j/4AAQSkZJRgABAQABAAD..." 
          }
        }
      ]
    }
  ]
}
```