Merlin
MLIR/IREE-based compiler stack for CPU and custom RISC-V targets.
Merlin compiles ML models down to host CPU, the SpacemiT X60 board, and the Saturn OPU / FireSim flow, with first-class plugins for new dialects, HAL drivers, and target generators.
Pick your path
Use Merlin
Compile or run models on a supported target. Start here:
Bring up a target
Wire a new accelerator into the codegen + runtime pipeline:
Develop Merlin
Change compiler passes, HAL drivers, or plugins:
Build these docs locally
conda activate merlin-dev
MLIR_TBLGEN=build/host-vanilla-release/llvm-project/bin/mlir-tblgen \
uv run --with-requirements docs/requirements.txt python docs/hooks.py
uv run --with-requirements docs/requirements.txt zensical serve
Then open http://127.0.0.1:8000.