Fine-Tuning

Upload your training data to a designated S3 bucket. Data should include:

• Positive examples (good matches)
• Negative examples (poor matches)
• Relevance scores
• Content pairs with similarity annotations

Our team will provide detailed specifications for your use case.

Use the Mixpeek Dashboard to schedule a fine-tuning job:

{
  "base_model_id": "multimodal",
  "training_data": "s3://your-bucket/training-data",
  "specs": {
    "epochs": 10,
    "batch_size": 32,
    "learning_rate": 2e-5,
    "loss_function": "contrastive"
  }
}

This returns a new model_id (e.g., model_1askdh2390) for use in your API calls.

Track model versions in your metadata for reproducibility:

{
  "metadata": {
    "model_version": "model_1askdh2390",
    "base_model": "multimodal",
    "training_date": "2024-03-15",
    "description": "Optimized for technical documentation"
  }
}