Fused Linear JSD Kernel Optimization

Source provenance: research/problems/fused_linear_jsd from Frontier-CS.

Optimize a fused dual-linear Jensen-Shannon divergence Triton kernel.

What To Submit

Submit solution.py whose Solution.solve(spec_path) returns code or a program path defining the required kernel function.

Scoring

Correctness is mandatory; source benchmark score is reported on a 0-100 scale.

Public And Official Data

Committed public data is a small smoke benchmark. Official evaluation uses private benchmark data from official-runs.zip.

Risks

Requires actual CUDA hardware for smoke; private metadata is visible to participant code but contains no secrets.