Datasetpublic

DiogenesChen122/Dr.Sparse-GPT56Luna-eval-b200-rl562-spgemm

Dr.Sparse — GPT-5.6 Luna SpGEMM eval on the 562-matrix pool Snapshot state: complete (generated 2026-09-19 08:03 UTC) phase results explore (3 branches x 5 iterations) 1686 / 1686 exploit (top-2 branches x 10 iterations) 1124 / 1124 matrices covered 562 / 562 What this is Tree-search SpGEMM kernel optimization over the 562 matrices of KinGeorge/Dr.Sparse-RL-train-562, run on NVIDIA B200 (sm_100) and scored against cuSPARSE. Model:… See the full description on the dataset page: https://huggingface.co/datasets/DiogenesChen122/Dr.Sparse-GPT56Luna-eval-b200-rl562-spgemm.

sourceHugging Facecc-by-4.0updated 6d agoView on Hugging Face

0likes86downloads

Dataset Card

Dr.Sparse — GPT-5.6 Luna SpGEMM eval on the 562-matrix pool

Snapshot state: complete (generated 2026-09-19 08:03 UTC)

phase	results
explore (3 branches x 5 iterations)	1686 / 1686
exploit (top-2 branches x 10 iterations)	1124 / 1124
matrices covered	562 / 562

What this is

Tree-search SpGEMM kernel optimization over the 562 matrices of KinGeorge/Dr.Sparse-RL-train-562, run on NVIDIA B200 (sm_100) and scored against cuSPARSE.

—Model: openai/gpt-5.6-luna via OpenRouter, reasoning: {effort: high} on every call.
—Search: per matrix, 3 explore branches of 5 iterations, then the top 2 branches get 10 more iterations. 2,810 GPU elements total, 30 concurrent, one B200 per element so no benchmark shares a GPU.
—Scoring: compile -> compute-sanitizer memcheck -> benchmark vs cuSPARSE -> NCU on correct kernels.

Layout (inside the tarball)

rl562_b200_GPT56Luna_openrouter/spgemm/<level>/SpGEMM_<matrix>/
    tree_search_state.json          # the 3 planned strategies
    branch_<i>_<label>/             # per-branch workspace: kernel_iter_*.cu, chat_iter_*.json, NCU
    branch_<i>_result.json          # explore result (status, best_speedup, iterations)
    branch_<i>_exploit_result.json  # exploit result for the top-2 branches
branch_array/gpt56luna-rl562-spgemm-42497038/
    explore_manifest.csv exploit_manifest.csv config.yaml
    plan_logs/ explore_logs/ exploit_logs/ master-*.log

Unpack with tar -I zstd -xf rl562_b200_GPT56Luna_spgemm.tar.zst.