CLI_QUICKSTART
Compress locally.
Deploy natively.
Execute LeanRoute natively via the interactive CLI. The automated engine handles downloading, scanning, routing, and precision allocation automatically.
leanroute-node-01 ~ HAQ Terminal
INSTALLATION_MATRIX
Universal Binary Distribution
LeanRoute CLI is compiled as a standalone binary with optimized bindings. Install via your preferred package manager to run locally.
go install github.com/leanroute-ai/leanroute/cmd/leanroute@latest
npm install -g leanroute-cli
pip install leanroute-haq
brew install leanroute/tap/leanroute
HAQ_PIPELINE
Adaptive Precision Mapping Architecture
Unlike standard 4-bit quantization (GPTQ/AWQ) that blindly compresses matrices uniformly leading to accuracy degradation, LeanRoute's HAQ engine runs a layer-wise sensitivity analysis before compression. Information-heavy layers remain in 8-bit, standard weights are pushed to 4-bit, and redundant blocks are pruned entirely.
Phase 1 Input
16-bit Original Model
Phase 2 Runtime Mapping
Hessian Trace Formulation
Pruned
Redundant Layers
3/4-bit
Standard Vectors
8-bit
Intelligence Core