Automated nightly build from main.

Commit: 5d021a9

This is a pre-release build and may be unstable.

What's New

SafeTensors Support

• Full save/load support with metadata preservation
• HuggingFace Hub compatibility - Direct upload/download
• Auto-detection - Format automatically detected from file extension
• Metadata serialization - Model name, architecture, version in headers

HuggingFace Pipeline

• HfModelFetcher - Download models from HuggingFace Hub
• SafeTensors parsing - Load models in SafeTensors format
• Projection layers - Handle dimension-mismatched distillation

New CLI Tools (Sub-Crates)

• entrenar-common - CLI utilities, progress bars, structured output
• entrenar-lora - LoRA optimization and memory planning
• entrenar-inspect - Model inspection and validation
• entrenar-shell - Interactive REPL shell
• entrenar-bench - Distillation benchmarking
• entrenar-distill - Distillation pipeline demo

Quality

• All clippy warnings resolved
• 1,900+ tests passing
• Book documentation updated

Crates Published

• entrenar v0.2.2
• entrenar-common v0.1.0
• entrenar-lora v0.1.0
• entrenar-inspect v0.1.0
• entrenar-shell v0.1.0
• entrenar-bench v0.1.0
• entrenar-distill v0.1.0

Installation

```bash
cargo install entrenar
cargo install entrenar-lora
cargo install entrenar-inspect
cargo install entrenar-shell
cargo install entrenar-bench
cargo install entrenar-distill
```

Full Changelog: v0.2.1...v0.2.2

Entrenar v0.1.0 - Production Release

Entrenar (Spanish: "to train") is a high-performance Rust library for training and optimizing neural networks with automatic differentiation, state-of-the-art optimizers, and memory-efficient LoRA/QLoRA fine-tuning.

🎉 First Production Release

This is the initial production-ready release of Entrenar, featuring a complete training infrastructure built with EXTREME TDD methodology.

✨ Features

Core Framework

• Autograd Engine - Tape-based automatic differentiation with backward propagation
  • Tensor abstraction with gradient tracking
  • BackwardOp trait for custom operations
  • Attention, matmul, softmax, layer norm operations
  • Property-based gradient checking (200K+ iterations)

Optimizers

• SGD with momentum support
• Adam optimizer with bias correction
• AdamW with decoupled weight decay
• Gradient clipping via L2 norm
• Learning rate scheduling (Cosine, Linear)
• SIMD acceleration for parameter updates via Trueno
• Convergence property tests for all optimizers

LoRA & QLoRA

• LoRA layers with configurable rank and alpha
• QLoRA with 4-bit quantized base weights
• Adapter management (save/load separately from base model)
• Memory benchmarks showing 4× reduction with QLoRA
• Gradient flow tests ensuring proper backpropagation

Model Merging (Arcee Methods)

• TIES (Task Inference via Elimination and Sign voting)
• DARE (Drop And REscale with Bernoulli masking)
• SLERP (Spherical Linear intERPolation)
• Property tests for permutation invariance
• Multi-model ensemble support

Knowledge Distillation

• Temperature-scaled KL divergence loss
• Multi-teacher ensemble distillation
• Progressive layer-wise distillation
• 44 distillation tests including 13 property tests
• Temperature smoothing validation

Training Loop

• High-level Trainer abstraction
• Batch processing with configurable batch size
• Metrics tracking (loss history, learning rates, steps)
• Gradient clipping integration
• Learning rate scheduling during training

Model I/O

• Save/load models with multiple formats (JSON, YAML)
• ModelMetadata with custom fields
• Round-trip integrity validation
• Automatic format detection from file extension

Declarative Configuration

• YAML-based training configuration (Ludwig-style)
• Schema validation with comprehensive error messages
• Auto-inference of feature types from data
• Single-command training via train_from_yaml()
• Builder pattern for optimizers and models from config

📊 Quality Metrics

• 258 tests passing (100% success rate)
  • 130 core library tests
  • 18 gradient checking tests
  • 35 architecture tests
  • 16 I/O and configuration tests
  • 13 property-based tests (13,000+ iterations)
  • 15 chaos engineering tests
  • 11 memory benchmark tests
  • 10+ additional integration tests
• 0 clippy warnings (strict mode)
• 0 TODOs remaining in codebase
• 55 Rust source files with full documentation
• 146 chapter mdBook documentation

🚀 Getting Started

Add to your Cargo.toml:

```toml
[dependencies]
entrenar = "0.1.0"
```

Quick example:

```rust
use entrenar::{Tensor, train::{Trainer, TrainConfig, MSELoss}, optim::Adam};

// Create trainer
let config = TrainConfig::new().with_log_interval(100);
let optimizer = Adam::default_params(0.001);
let mut trainer = Trainer::new(parameters, optimizer, config);
trainer.set_loss(Box::new(MSELoss));

// Train
let avg_loss = trainer.train_epoch(batches, |x| model.forward(x));
```

📚 Documentation

• Crates.io: https://crates.io/crates/entrenar
• API Docs: https://docs.rs/entrenar
• Book: https://github.com/paiml/entrenar/tree/main/book
• Examples: https://github.com/paiml/entrenar/tree/main/examples

🔗 Links

• Repository: https://github.com/paiml/entrenar
• CHANGELOG: https://github.com/paiml/entrenar/blob/main/CHANGELOG.md
• License: MIT

🙏 Acknowledgments

Built with:

• trueno 0.4.1 - SIMD-accelerated compute engine
• ndarray 0.16 - N-dimensional arrays
• serde 1.0 - Serialization framework

Part of the PAIML (Pragmatic AI Labs ML) ecosystem.

Quality Grade: A+ (Zero defects, 258 passing tests, EXTREME TDD methodology)