Home

FlashOptim Documentation

Welcome to the FlashOptim documentation — a comprehensive model optimization toolkit for FlashVision models.

Overview

FlashOptim provides state-of-the-art model compression and optimization techniques:

• Quantization — Reduce model precision (INT8, FP16) for faster inference
• Pruning — Remove redundant weights and channels
• Distillation — Transfer knowledge from large to small models
• NAS — Automatically search for efficient architectures
• LoRA — Low-Rank Adaptation for efficient fine-tuning
• Export — Deploy to ONNX, TensorRT, OpenVINO

Navigation

Page	Description
Installation	Setup and installation guide
Quick Start	Get running in minutes
Quantization	PTQ and QAT guide
Pruning	Weight and channel pruning
Distillation	Knowledge distillation
NAS	Neural Architecture Search
FAQ	Frequently asked questions

Requirements

• Python >= 3.8
• PyTorch >= 2.0.0
• CUDA >= 11.8 (recommended for GPU acceleration)