Skip to main content

User guide

This guide walks you through the process of fine-tuning large language models (LLMs) on the platform using a no-code interface, enabling custom model development without writing any code.

Pre-requisites​

  1. Sufficient Account Credits

    • Ensure your account has enough credits for training and deployment. Add credits if necessary to avoid interruptions during resource-intensive operations.

  2. Clean, Preprocessed CSV File

    • Prepare your dataset in a CSV format.
    • The data should be cleaned and preprocessed (no missing values, consistent formatting, etc.).
    • Each row should represent a unique training example, and clearly labeled columns for the model’s input and output.

  3. Available Instance for Deployment

    • You’ll need a cloud GPU instance to deploy your tuned model. For guidance, refer to the instance launching steps outlined in the platform documentation.

Steps to Use No Code Tuning​

1. Navigate to Model Tuning Page​

  • From the platform’s Home page, select LLM Tools.
  • Then click Fine Tuning to reach the no-code model tuning dashboard.

2. Select Model Category and Model​

MODEL_CATEGORY_SELECTION

  • Choose the model category that matches your use case (e.g., Text Generation, Code Generation).
  • Pick a model suited for your task, such as Falcon-11B, by clicking on its name or selection button.

3. Upload Your CSV File​

FILE_UPLOAD

  • Use the upload interface to add your prepared CSV file.
  • Ensure your data meets the platform’s requirements; refer to documentation for supported formats and size limits.
  • The system may provide a preview or validation before proceeding.

4. Set Training Parameters​

SET_PARAMS

  • Specify which columns in your CSV will be used for model input and desired output.
  • Configure essential training options:
    • Batch Size: Controls how many samples are processed before updating the model.
    • Learning Rate: Determines the adjustment step size during model optimization.
    • Epochs: How many times the training data is passed through the model.
  • Additional advanced settings (if available) can further tailor the tuning process to your needs.

5. Initiate Tuning​

  • Review your configuration summary for correctness.
  • Click Start Tuning to begin.
  • The system automatically creates an isolated container to run your fine-tuning job, keeping it resource-contained and safe.

6. Check Tuning Status​

  • Monitor the progress of your job on the dashboard.
  • Wait until the model’s status changes from “In Progress” or “Tuning” to Tuning Completed.
  • Notifications or logs may be available for detailed tracking.

Tuning_Status

7. Deploy the Model​

  • When fine-tuning finishes, deploy the model:
    • Go to Actions → Deploy.
    • Select one of your active GPU instances. Note: Instances incur hourly billing once running.
    • Deployment spins up the necessary environment to serve your model for inference.

8. Verify Deployment​

  • Confirm deployment was successful by checking that the model status is Deployment Ready on your dashboard or model listing.
  • The platform may provide logs or a test endpoint for quick checks.

9. Use the Model​

  • Integrate and test the deployed model:
    • GetCode: Instantly generate code snippets for terminal, Python scripts, or React frontend integration.
    • Interactive UI: Use the built-in browser interface for rapid testing, parameter tuning, and live demonstration.

10. Stop or Delete the Model (Optional)​

  • Stop the Model: Pause resource usage and billing by stopping deployment when the model isn’t in use.
  • Delete the Model: When finalized, delete it from the dashboard. Ensure status shows Deployment Stopped before deleting to safely free resources.

Additional Tips​

  • Data Backup: Always keep local copies of important datasets and trained models.
  • Resource Monitoring: Track usage and budget from the dashboard.
  • Parameter Optimization: Experiment with different training settings for optimal performance.
  • Support: Reach out to helpdesk if you encounter issues or need advanced customization.
    • Join our Discord Group to connect with the community and share your feedback directly or email us at digital@qubrid.com.

By following these steps, you can efficiently train, deploy, and utilize tailored LLMs for your use cases—entirely through a no-code interface, saving time and effort while ensuring robust results.