Module prune 

Neural network pruning integration for Entrenar

This module provides training-time pruning capabilities that integrate with the Aprender pruning primitives. It implements:

• Pruning Schedules: OneShot, Gradual, and Cubic sparsity schedules
• Pruning Callback: Integration with the training loop
• Calibration Pipeline: Activation collection for Wanda/SparseGPT
• Prune-Finetune Pipeline: End-to-end pruning workflow

Toyota Way Principles

• Kaizen (Continuous Improvement): Gradual pruning schedules
• Jidoka (Quality at Source): Validates sparsity at each step
• Genchi Genbutsu (Go and See): Uses real activation data

Example

use entrenar::prune::{PruningSchedule, PruningCallback, PruningConfig};
use aprender::pruning::{MagnitudeImportance, WandaPruner};

let schedule = PruningSchedule::Gradual {
    start_step: 1000,
    end_step: 10000,
    initial_sparsity: 0.0,
    final_sparsity: 0.5,
    frequency: 100,
};

let config = PruningConfig::default()
    .with_schedule(schedule)
    .with_target_sparsity(0.5);

let callback = PruningCallback::new(config);
trainer.add_callback(callback);

References

• Han, S., et al. (2015). Learning both weights and connections. NeurIPS.
• Sun, M., et al. (2023). A simple and effective pruning approach. arXiv:2306.11695.
• Zhu, M., & Gupta, S. (2017). To prune, or not to prune. arXiv:1710.01878.

Structs

CalibrationCollector

Calibration data collector for pruning.

CalibrationConfig

Configuration for calibration data collection.

CalibrationDataConfig

Configuration for calibration data loading.

CalibrationDataLoader

Calibration data loader for pruning.

PruneFinetunePipeline

Prune-Finetune pipeline orchestrator.

PruneTrainer

Prune-finetune trainer that orchestrates the full pipeline.

PruneTrainerConfig

Configuration for the prune-finetune trainer.

PruningCallback

Callback for applying pruning during training.

PruningConfig

Configuration for pruning operations.

PruningMetrics

Metrics collected during pruning.

Enums

PruneMethod

Pruning method selection.

PruningSchedule

Pruning schedule defining when sparsity increases during training.

PruningStage

Current stage of the pruning pipeline.

SparsityPatternConfig

Sparsity pattern selection.