Installation

Prerequisites

  • Python 3.11 or higher

Installation with pip

git clone https://github.com/imec-int/hespas.git
cd hespas
pip install .

Alternatively, you can use uv:

uv sync

Optional dependencies

Hespas supports several optional estimator backends. Install them as needed:

# IREE compiler and runtime (CPU/GPU execution)
pip install ".[iree]"

# ZigZag architectural simulator
pip install ".[zigzag_onnx]"

# External StableHLO to ONNX translation package
pip install ".[stablehlo_to_onnx]"

Running Tests

python -m pytest tests

For verbose output with debug logging:

python -m pytest -s --log-cli-level=DEBUG tests

Running a Simple Simulation

Generate a Chakra trace for a sample workload:

hespas_chakra_gen tests/fixtures/configs/config_roofline_a100.json

The output will be written to output/basic/roofline_a100/ and includes:

  • mini_module_*.mlir — individual MLIR compute modules

  • intermediate_results.json — execution times and compilation status

  • dev.*.et — Chakra trace files for use with ASTRA-sim