Files
zeekay b7ca3f46db Run cargo fmt for consistent formatting
Apply rustfmt across the entire codebase for CI compliance.
2026-01-17 17:27:31 -08:00

76 lines
2.3 KiB
Rust

//! Example: Train Llama 3.1 8B on zen-agentic-dataset
use anyhow::Result;
use hanzo_training::{init_logging, utils, Trainer, TrainingConfig};
use tracing::info;
#[tokio::main]
async fn main() -> Result<()> {
// Initialize logging
init_logging()?;
info!("🚀 Starting Llama 3.1 8B training on zen-agentic-dataset");
// Load configuration
let config = TrainingConfig::from_file("configs/llama3-8b.yaml")?;
// Validate configuration
config.validate()?;
info!("✅ Configuration validated");
// Print memory and time estimates
let estimated_memory = utils::estimate_memory_requirements(
config.model.hidden_size.unwrap_or(4096) * config.model.num_layers.unwrap_or(32),
config.training.batch_size,
config.model.max_seq_length,
2, // bf16
);
let estimated_time = utils::estimate_training_time(
config.model.hidden_size.unwrap_or(4096) * config.model.num_layers.unwrap_or(32),
10000, // estimated dataset size
config.effective_batch_size(),
config.training.epochs.unwrap_or(1),
1000, // tokens per second estimate
);
info!(
"📊 Estimated memory usage: {}",
utils::format_bytes(estimated_memory)
);
info!(
"⏱️ Estimated training time: {}",
utils::format_duration(estimated_time)
);
// Create output directory structure
utils::create_output_dirs("./output/llama3-8b-zen-agentic")?;
// Create trainer
let mut trainer = Trainer::new(config)?;
info!("✅ Trainer created");
// Start training with progress tracking
info!("🎯 Starting training...");
let progress = utils::ProgressBar::new(10000); // Estimated steps
progress.set_message("Training model...");
let result = trainer.train()?;
progress.finish_with_message("Training completed!");
// Print results
info!("🎉 Training completed successfully!");
info!("Final loss: {:.6}", result.final_loss);
info!("Total steps: {}", result.total_steps);
info!(
"Training time: {}",
utils::format_duration(result.training_time)
);
// Save final model
trainer.save_checkpoint(std::path::Path::new("./output/llama3-8b-zen-agentic"))?;
info!("💾 Model saved to ./output/llama3-8b-zen-agentic");
Ok(())
}