Get up and running with Atrex-Bench in minutes using Docker and the built-in CLI tools.
The ROCm runtime image bundles PyTorch, AITER, FlyDSL, and evaluation tools.
docker run -it --rm \
--device=/dev/kfd --device=/dev/dri --group-add video \
-e ANTHROPIC_API_KEY -e OPENAI_API_KEY \
-v $PWD:/workspace \
<YOUR_REGISTRY>/atrex-bench:rocm7.2 Evaluate a candidate kernel against a reference operator.
python scripts/run_eval.py \
--input path/to/candidate.py \
--reference-dir data/fused_moe \
--output results/run Use an LLM CLI to generate a candidate kernel from a reference.
python scripts/run_generate.py \
--operator fused_moe \
--backend triton \
--cli codex