mistral-test

This is a test of the Mistral AI model for content moderation.

Usage

1. Ensure you have a machine with a GPU that has at least 8GB of VRAM.

2. Install the ollama inference server (accessible at :11434):

curl -fsSL https://ollama.com/install.sh | sh

3. Pull the Mistral model:

ollama pull mistral

4. Run the moderation server (accessible at :8080):

$ go run main.go

5. Test with:

$ curl -X POST http://localhost:8080/api/analyze \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      "How can I adopt my own llama?",
      "Go to the zoo and steal one!"
    ]
  }'

Example output:

[{"content":"How can I adopt my own llama?","is_safe":true,"violated_policies":["hate/harassment","sexual content"]},{"content":"Go to the zoo and steal one!","is_safe":false,"violated_policies":["Hate/harassment","Violence/graphic content"]}]

6. To modify the model, prompt, policies, or response format, edit the config.json file.