Quickstart
Five minutes from zero to your first capacity plan.
1. Install
bash
cargo install --path crates/hwledger-cli --root /tmp/hwl-install
export PATH="/tmp/hwl-install/bin:$PATH"
hwledger --version
Keyframes (0 — VLM-friendly)
Loading manifest…
2. Plan your first model
Pass any HF-style config.json:
bash
hwledger plan tests/golden/deepseek-v3.json --seq 32768 --users 2
Keyframes (0 — VLM-friendly)
Loading manifest…
3. Probe local GPUs
bash
hwledger probe list
Keyframes (0 — VLM-friendly)
Loading manifest…
Need deeper telemetry? Use hwledger probe watch --interval 1s to stream samples.
4. Export a plan
Convert a plan into flags for your inference engine:
bash
hwledger plan tests/golden/deepseek-v3.json --seq 32768 --export vllm
hwledger plan tests/golden/deepseek-v3.json --seq 32768 --export llama-cpp
hwledger plan tests/golden/deepseek-v3.json --seq 32768 --export mlxNext steps
- Installation options for Homebrew, cargo, and release DMG paths
- Fleet overview to wire up multiple hosts
- Math core for the formula tree behind the plan output
- Clients for SwiftUI, Streamlit, and the Rust CLI