Expand description
NAPI-RS bindings for training utilities
Provides Node.js bindings for:
- Loss functions (InfoNCE, LocalContrastive, SpectralRegularization)
- Optimizers (SGD, Adam, AdamW)
- Learning rate schedulers
- Curriculum learning
- Negative mining
Structs§
- Adam
Optimizer - Adam optimizer with bias correction
- AdamW
Optimizer - AdamW optimizer (Adam with decoupled weight decay)
- Curriculum
Scheduler - Curriculum scheduler for progressive training
- Curriculum
Stage Config - Curriculum stage configuration
- Hard
Negative Miner - Hard negative miner for selecting informative negatives
- InBatch
Miner - In-batch negative mining utility
- InfoNCE
Loss - InfoNCE contrastive loss for representation learning
- Learning
Rate Scheduler - Learning rate scheduler with warmup and cosine decay
- Local
Contrastive Loss - Local contrastive loss for neighborhood preservation
- Loss
With Gradients - Loss computation result with gradients
- SGDOptimizer
- SGD optimizer with optional momentum and weight decay
- Spectral
Regularization - Spectral regularization for smooth representations
- Temperature
Annealing - Temperature annealing scheduler
Enums§
- Decay
Type - Decay type for temperature annealing
- Mining
Strategy - Mining strategy for negative selection