{ "domain": "computer-science", "image": "python:3.12-slim", "data": "The twelve problems and their input shapes are in problems/; nothing else is needed.", "benchmark": { "task": "GPU kernel optimization", "metric": { "name": "speed-up", "better": "higher" }, "subtasks": "kernels", "guardrails": [ "Outputs match the PyTorch reference within atol 1e-2 and rtol 1e-2 on five random inputs", "Timed on the pinned H100 runner, the median of 100 runs after warm-up", "No library calls the reference does not make" ], "aggregate": "geomean", "baseline": { "name": "PyTorch eager", "score": 1, "subtasks": { "matmul": 1, "conv2d": 1, "softmax": 1, "layernorm": 1, "gelu": 1, "attention": 1, "groupnorm": 1, "maxpool": 1, "cumsum": 1, "argmax": 1, "mse-loss": 1, "lstm-cell": 1 } } } }