Skip to content

docs: add training recommendations for epochs, shots, and noise upscaling#72

Merged
ivanbasov merged 1 commit intomainfrom
worktree-add-docs
Apr 22, 2026
Merged

docs: add training recommendations for epochs, shots, and noise upscaling#72
ivanbasov merged 1 commit intomainfrom
worktree-add-docs

Conversation

@ivanbasov
Copy link
Copy Markdown
Collaborator

Summary

  • Models 1, 4, 5 (uncorrelated matching): document the requirement to train for at least 100 epochs; added to both the README.md model-selection section and the TRAINING.md env-var table (PREDECODER_TRAIN_EPOCHS).
  • Shots per epoch: recommend 67 million shots/epoch with 8 GPUs (PREDECODER_TRAIN_SAMPLES=67108864); fewer shots produce worse results. Added in the same two locations.
  • Noise upscaling algorithm: added a concise "Algorithm in brief" paragraph to the existing noise upscaling section in README.md, explicitly stating the p_max → rescale-to-0.6% recipe so the mechanism is clear at a glance.

Test plan

  • Documentation-only change — no code modified, no tests required.
  • CI spdx-header-check unaffected (no source files changed).
  • Verify rendered Markdown renders correctly on GitHub.

🤖 Generated with Claude Code

…ling

- Advise ≥100 epochs for models 1, 4, 5 (uncorrelated matching) in both
  README and TRAINING env-var table
- Document 67 M shots/epoch with 8 GPUs as the recommended sample count
- Add concise algorithm summary to the noise upscaling section (p_max →
  rescale to 0.6 %)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
@ivanbasov ivanbasov requested review from bmhowe23 and tlubowe April 21, 2026 18:33
@ivanbasov ivanbasov merged commit 9d41843 into main Apr 22, 2026
17 checks passed
@ivanbasov ivanbasov deleted the worktree-add-docs branch April 22, 2026 15:15
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.

2 participants