Skip to content

Comet Model Registry Integration

Overview

IdeaWeaver integrates with Comet for experiment tracking and model management. This guide shows you how to use Comet with IdeaWeaver for model versioning and deployment.

Setup

Train and Register a Model

source ideaweaver-env/bin/activate
ideaweaver train \
  --model sshleifer/tiny-distilbert-base-cased \
  --dataset ./datasets/training_data.csv \
  --comet-api-key <your-comet-api-key> \
  --comet-project-name <your-project> \
  --comet-workspace <your-workspace>

Example Output

🤗 Using model: sshleifer/tiny-distilbert-base-cased
🚀 Starting model training...

============================================================
🎉 TRAINING SUMMARY
============================================================
📂 Model Path:           ./my-model
🤖 Base Model:           sshleifer/tiny-distilbert-base-cased
📊 Dataset:              ./autotrain_projects/my-model

📊 KEY PERFORMANCE METRICS
----------------------------------------
📉 Final Train Loss:     1.0986
🎯 Overall Accuracy:     20.0%

============================================================
✨ Training completed successfully! Model is ready for use.
============================================================

✅ Training completed successfully!
📁 Model saved to: ./my-model

Comet UI Example

Once your model is registered, you can view it in the Comet UI:

Comet UI Screenshot

Features

  • Experiment tracking
  • Model versioning
  • Artifact storage
  • Performance metrics
  • Model deployment
  • Collaboration tools

Configuration

Required Parameters

  • --comet-api-key: Your Comet API key
  • --comet-project-name: Project name
  • --comet-workspace: Workspace name

Getting Your Comet API Key

  1. Go to Comet Account Settings
  2. Create a new API key
  3. Copy the key value
  4. Use it in your training command

Best Practices

  1. Project Organization
  2. Use meaningful experiment names
  3. Tag experiments appropriately

  4. Document model versions

  5. Model Management

  6. Version your models
  7. Add model descriptions

  8. Track model lineage

  9. Collaboration

  10. Share experiments with team
  11. Document findings
  12. Track changes

Troubleshooting

Common Issues

  1. Authentication Errors
  2. Verify API key is correct
  3. Check key permissions

  4. Ensure proper access

  5. Connection Issues

  6. Check internet connection
  7. Verify project exists
  8. Validate workspace name

Debug Mode

Enable verbose output for debugging:

ideaweaver train \
  --model sshleifer/tiny-distilbert-base-cased \
  --dataset ./datasets/training_data.csv \
  --comet-api-key <your-comet-api-key> \
  --verbose

Resources