HoprLab is a CLI and research toolkit for simulating AI training math before spending time and money on model training.
The working MVP lives in hoprlab/. It estimates model size, activation memory, optimizer memory, approximate VRAM usage, training time, token budget, config risks, benchmark speed, and reliability.
cd hoprlab
python3 -m venv .venv
.venv/bin/pip install -e ".[dev]"
.venv/bin/hoprlab estimate configs/example_transformer.yaml
.venv/bin/hoprlab simulate configs/example_transformer.yaml
.venv/bin/hoprlab benchmark
.venv/bin/hoprlab test-reliability configs/example_transformer.yaml
Or use the root Makefile:
make setup
make test
make build
make reliability
HoprLab includes optional native benchmark modules:
-
Rust:
hoprlab/native/rust -
C:
hoprlab/native/c
The Python CLI falls back gracefully when native backends are unavailable.
MIT
source & further reading
github.com — original article