Gene: ML Pipeline

Description

Machine learning and AI pipeline capabilities. Combines HuggingFace skills for datasets, model training, evaluation, and tool building.

Trigger Conditions

• User asks to work with machine learning models
• Dataset manipulation or analysis
• Model training or fine-tuning
• Evaluating ML models
• Building ML-powered tools

Capabilities

HuggingFace Datasets

• Load and manipulate datasets
• Data preprocessing and cleaning
• Train/test splits
• Dataset versioning

Model Trainer

• Fine-tune existing models
• Training loop setup
• Hyperparameter configuration
• Model evaluation

Evaluation

• Benchmark model performance
• Metrics calculation
• Comparison analysis
• Quality assessment

Tool Builder

• Build ML-powered tools
• Inference pipelines
• API integration with models

TrackIO

• Experiment tracking
• Metrics logging
• Run management

Execution Protocol

Step 1: Task Analysis

• Identify ML task type (classification, generation, etc.)
• Determine required models/datasets
• Plan pipeline steps

Step 2: Data Preparation

• Load datasets
• Preprocess data
• Handle missing values
• Create train/test splits

Step 3: Model Selection & Training

• Select appropriate model
• Configure training
• Fine-tune if needed
• Validate performance

Step 4: Evaluation & Deployment

• Run evaluation metrics
• Compare to baselines
• Prepare for deployment

Common Tools

• transformers library
• datasets library
• accelerate
• peft (parameter-efficient fine-tuning)
• wandb for tracking

Guardrails

• Don't make claims without evidence
• Report uncertainty in predictions
• Consider ethical implications
• Validate on appropriate test sets

Integration

• Works with: code-execution, research-analysis
• Essential for ML/AI workflows