pub mod activation_capture;
pub mod benchmark;
pub mod calibration;
pub mod domain_mapper;
pub mod export;
pub mod import;
pub mod kv_calculator;
pub mod profiler;
pub mod pruner;
pub mod quant_allocator;
pub mod sensitivity;
pub mod streaming;
pub mod tptf_format;
pub use activation_capture::{
load_domain_maps_from_dir, ActivationCallback, ActivationCapture, ActivationMap,
LayerActivations,
};
pub use benchmark::QualityBenchmark;
pub use calibration::CalibrationGenerator;
pub use domain_mapper::DomainMapper;
pub use export::detect::ModelFormat;
pub use import::gguf::{GgufImporter, GgufModel};
pub use kv_calculator::KvCacheCalculator;
pub use profiler::HardwareProfiler;
pub use pruner::SurgicalPruner;
pub use quant_allocator::MixedPrecisionAllocator;
pub use sensitivity::{LayerSensitivityMap, SensitivityConfig};
pub use streaming::StreamingLoader;
pub use tptf_format::{TensorBlock, TptfHeader, TptfWriter};
use std::path::PathBuf;
#[derive(Debug, Clone)]
pub struct OptimizerConfig {
pub model_path: PathBuf,
pub output_path: PathBuf,
pub max_loss_fraction: f32,
pub domains_to_drop: Vec<String>,
pub force_streaming: bool,
}
impl OptimizerConfig {
pub fn new(model_path: impl Into<PathBuf>, output_path: impl Into<PathBuf>) -> Self {
OptimizerConfig {
model_path: model_path.into(),
output_path: output_path.into(),
max_loss_fraction: 0.05,
domains_to_drop: Vec::new(),
force_streaming: false,
}
}
}
#[derive(Debug, Clone)]
pub struct OptimizationResult {
pub output_path: PathBuf,
pub per_layer_bits: Vec<u8>,
pub pruned_domains: Vec<String>,
pub size_before_gb: f64,
pub size_after_gb: f64,
pub compression_ratio: f32,
pub ppl_delta_pct: f32,
}