entrenar 0.7.13

Training & Optimization library with autograd, LoRA, quantization, and model merging
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43

//! Quantization: QAT and PTQ
//!
//! Provides quantization for QLoRA and Quantization-Aware Training:
//! - 4-bit block-wise quantization for QLoRA
//! - Fake quantization with STE for QAT
//! - PTQ calibration (min-max, percentile, moving average)
//! - GGUF-compatible Q4_0/Q8_0 formats
//! - Per-channel vs per-tensor quantization granularity
//! - Quantization error analysis and metrics
//! - Accuracy degradation benchmarks

mod benchmarks;
mod calibration;
mod double_quant;
mod error_analysis;
mod fake_quantize;
mod gguf_quant;
mod granularity;
mod quant4bit;

pub use benchmarks::{
    accuracy_retention, compare_bit_width_degradation, generate_gaussian_weights,
    generate_multi_channel_weights, generate_uniform_weights, generate_weights_with_outliers,
    run_benchmark, run_full_benchmark_suite, BenchmarkSuite, QuantBenchmarkResult,
};
pub use calibration::{
    calibrate_min_max, calibrate_percentile, CalibrationMethod, CalibrationResult, Calibrator,
};
pub use double_quant::{
    dequantize_4bit_double, quantize_4bit_double, DoubleQuantized4Bit, DOUBLE_QUANT_BLOCK_SIZE,
};
pub use error_analysis::{
    analyze_error, analyze_outlier_impact, compare_bit_widths, error_within_bounds,
    scale_sensitivity, theoretical_max_error, theoretical_sqnr, QuantErrorStats,
};
pub use fake_quantize::{fake_quantize, ste_backward, FakeQuantConfig, FakeQuantize};
pub use gguf_quant::{GGUFQuantType, GGUF_BLOCK_SIZE, Q4_0, Q8_0};
pub use granularity::{
    calibrate_per_channel, calibrate_per_group, calibrate_per_tensor, compare_granularities,
    dequantize_tensor, dequantize_with_params, quantization_mse, quantize_tensor,
    quantize_with_params, QuantGranularity, QuantMode, QuantParams, QuantizedTensor,
};
pub use quant4bit::{dequantize_4bit, quantize_4bit, Quantized4Bit, BLOCK_SIZE};