benchmarks/gpu_mode/trimul/README.md

GPU Mode: Triangle Multiplicative Update (TriMul)

Evolve a Triton kernel for the TriMul operator using SkyDiscover.

Core operation for AlphaFold3, Chai, Protenix protein structure models.

Quick Start

From the repo root:

uv run skydiscover-run \
  benchmarks/gpu_mode/trimul/initial_program.py \
  benchmarks/gpu_mode/trimul/evaluator.py \
  -c benchmarks/gpu_mode/trimul/config.yaml \
  -s [your_algorithm] -i 50

Scoring

• Correctness: Must match reference output (rtol=0.02, atol=0.02 vs PyTorch reference)
• Score: SCORE_SCALE / geom_mean_us where SCORE_SCALE = 3000.0
• Higher is better (faster runtime = higher score)

Modal Cloud GPU Support

GPUMODE_USE_MODAL=true GPUMODE_MODAL_GPU=H100 \
  uv run skydiscover-run \
  benchmarks/gpu_mode/trimul/initial_program.py \
  benchmarks/gpu_mode/trimul/evaluator.py \
  -c benchmarks/gpu_mode/trimul/config.yaml \
  -s [your_algorithm] -i 50