extract-hyperparameters
Identify and document model hyperparameters from papers. Use when setting up training configurations.
When & Why to Use This Skill
The Extract Hyperparameters skill is a specialized tool designed to automate the identification and documentation of machine learning model configurations from research papers. By systematically parsing academic documents, it extracts critical values such as learning rates, batch sizes, and architectural details, significantly reducing the manual effort required to translate research findings into executable training configurations.
Use Cases
- Reproducing Research Results: Quickly gather all necessary numeric values and architectural parameters from a PDF to replicate a study's experimental setup with high fidelity.
- Automated Training Configuration: Generate implementation-ready configuration files in formats like YAML or JSON by extracting optimizer and regularization settings directly from source literature.
- Hyperparameter Benchmarking: Compare and contrast hyperparameter choices across multiple research papers to identify industry standards or optimal baseline configurations for new projects.
- Experiment Planning: Use extracted data to define informed search spaces and ranges for hyperparameter tuning and optimization workflows.
| name | extract-hyperparameters |
|---|---|
| description | "Identify and document model hyperparameters from papers. Use when setting up training configurations." |
| mcp_fallback | none |
| category | analysis |
| tier | 2 |
| user-invocable | false |
Extract Hyperparameters
Locate and document all hyperparameters mentioned in research papers including learning rates, batch sizes, and model configurations.
When to Use
- Reproducing paper results
- Setting up model training configurations
- Comparing hyperparameter choices across papers
- Planning hyperparameter tuning experiments
Quick Reference
# Extract numeric values and parameters from papers
pdftotext paper.pdf - | grep -i "learning rate\|batch\|epochs\|weight decay\|dropout" | head -20
# Common pattern search
grep -E "\\b(lr|batch_size|epochs|momentum|dropout|layers)\\s*[=:]" config.py
Workflow
- Find hyperparameter table: Look for "Table 1" or "Hyperparameters" section
- Document architecture parameters: Layer sizes, activation functions, normalization
- Extract training parameters: Learning rate, batch size, epochs, optimizers
- Note regularization: Dropout, weight decay, batch normalization
- Create configuration file: Translate to implementation format (YAML/JSON/Mojo)
Output Format
Hyperparameter documentation:
- Model architecture (layers, sizes, activations)
- Training parameters (LR, batch size, epochs)
- Optimizer configuration (type, momentum, decay)
- Regularization settings (dropout, L1/L2)
- Data preprocessing (normalization, augmentation)
- Hardware and precision (float32, float64)
References
- See
prepare-datasetskill for data configuration - See
train-modelskill for training implementation - See
/notes/review/mojo-ml-patterns.mdfor Mojo configuration patterns