Supported Formats

Entrenar supports multiple model serialization formats for different use cases.

Format Comparison

Format	Extension	Binary	HF Compatible	Use Case
SafeTensors	`.safetensors`	Yes	Yes	Production, sharing
JSON	`.json`	No	No	Debugging, inspection
YAML	`.yaml`	No	No	Configuration
GGUF	`.gguf`	Yes	No	Quantized models

Choosing a Format

SafeTensors (Recommended)

Use SafeTensors for:

Production deployments
Uploading to HuggingFace Hub
Large models (supports memory mapping)
Security-sensitive applications

#![allow(unused)]
fn main() {
let config = SaveConfig::new(ModelFormat::SafeTensors);
save_model(&model, "model.safetensors", &config)?;
}

JSON

Use JSON for:

Debugging and inspection
Small models
Human-readable output
Version control diffs

#![allow(unused)]
fn main() {
let config = SaveConfig::new(ModelFormat::Json).with_pretty(true);
save_model(&model, "model.json", &config)?;
}

YAML

Use YAML for:

Configuration files
Human-friendly syntax
Small models with metadata focus

#![allow(unused)]
fn main() {
let config = SaveConfig::new(ModelFormat::Yaml);
save_model(&model, "model.yaml", &config)?;
}

GGUF (Future)

GGUF format will be supported for:

Quantized model export
LLaMA.cpp compatibility
Integration with Realizar crate

Format Detection

Format is auto-detected from file extension:

#![allow(unused)]
fn main() {
// Auto-detect based on extension
let model = load_model("model.safetensors")?;  // SafeTensors
let model = load_model("model.json")?;         // JSON
let model = load_model("model.yaml")?;         // YAML
let model = load_model("config.yml")?;         // YAML (alternate extension)
}

Performance Characteristics

Format	100MB Save	100MB Load	Compression
SafeTensors	~100ms	~50ms	1x
JSON	~3s	~2.5s	~0.33x
YAML	~4s	~3.5s	~0.29x

Entrenar - Training & Optimization Library