Skip to content

Import KernelBook external corpora as verifiable task seeds - #35

Merged
ai-hpc merged 2 commits into
mainfrom
feat/external-task-seeds
Jul 26, 2026
Merged

Import KernelBook external corpora as verifiable task seeds#35
ai-hpc merged 2 commits into
mainfrom
feat/external-task-seeds

Conversation

@ai-hpc

@ai-hpc ai-hpc commented Jul 22, 2026

Copy link
Copy Markdown
Member

Summary

Implements seed → re-prove for external Triton corpora:

  • In: KernelBook, opus multi-turn, gpt-oss traces
  • Out: prompts.jsonl with origin kernelbook_seed (PyTorch tasks only)
  • Blocked: KernelBench as training (fingerprints + source filter); non-permissive licenses; external CoT/messages/Inductor kernels
  • Next step: sparkproof-triton-generate (multi-turn episodes) on CC → release-gate → registry

Also strengthens AST decontam (class names stripped) so KernelBench structural matches work.

Test plan

  • uv run pytest tests/ -q (313 passed, 9 skipped)
  • Optional: scripts/import_external_tasks.sh --limit 5 with HF network
  • Optional: CC smoke generate on seed prompts

ai-hpc added a commit that referenced this pull request Jul 22, 2026
Add docs/EXTERNAL_SEEDS.md (policy, licenses, decontam, commands, FAQ),
miner-guide section, and README/CHANGELOG links for PR #35.
ai-hpc added 2 commits July 26, 2026 09:18
Add sparkproof-import-external-tasks: extract PyTorch problems from
KernelBook and related traces, filter licenses, drop KernelBench, and
decontaminate — never copy external CoT. Re-prove with multi-turn
generate on a CC VM for SN74-verifiable rows.
Add docs/EXTERNAL_SEEDS.md (policy, licenses, decontam, commands, FAQ),
miner-guide section, and README/CHANGELOG links for PR #35.
@ai-hpc
ai-hpc force-pushed the feat/external-task-seeds branch from c6ddd90 to 28d0e40 Compare July 26, 2026 09:19
@ai-hpc
ai-hpc merged commit ff22d87 into main Jul 26, 2026
1 check passed
@ai-hpc
ai-hpc deleted the feat/external-task-seeds branch July 26, 2026 09:20
@ai-hpc ai-hpc mentioned this pull request Jul 26, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant