docs(algorithms): placement policy — core canon vs experimental variants (RFC) - #280
Closed
haonan3 wants to merge 1 commit into
Closed
docs(algorithms): placement policy — core canon vs experimental variants (RFC)#280haonan3 wants to merge 1 commit into
haonan3 wants to merge 1 commit into
Conversation
2 tasks
…nts (RFC) Writes down the placement rules for loss algorithms agreed during the experimental-tier design: core keeps the canonical family bases and the team-flagship published algorithms; model-specific one-offs and unpublished variants live in the experimental package that needs them (selected via _target_ — trainers are unaffected either way); movement in either direction is a deliberate PR with config dotpaths migrated in the same change. No algorithms move in this PR — it is the policy RFC.
haonan3
force-pushed
the
docs/algorithms-placement-policy
branch
from
July 31, 2026 07:28
b34db90 to
7cf4ae4
Compare
2 tasks
Collaborator
Author
|
Closing: rules 2–3 here restate the experimental-tier contract that #279 has since merged (with lint enforcement); the only net-new content is the rule-1 canon list, which is a rot-prone maintenance surface. If a real placement dispute arises, adjudicate it on that PR and fold the one-line outcome into the tier contract — case law over pre-legislation. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
RFC: writes the algorithm placement policy into
unirl/algorithms/README.md. Stacked on #210 (references the experimental tier); the diff collapses to the last commit once #210 merges.Three rules: (1) core keeps the canon — reference family bases (GRPO, FlowGRPO, SFT) and team-flagship published algorithms (CPPO, DRPO, FlowDPPO); (2) model-specific one-offs and unpublished variants start in the
experimental/package that needs them, selected via_target_exactly like core algorithms (trainers unaffected); (3) movement in either direction is a deliberate PR with config dotpaths migrated in the same change (check_recipe_targetsgates stale paths).No algorithms move in this PR — it establishes the policy so in-flight and future algorithm PRs choose their home at land time. Comment here if you want the lines drawn differently.
Test Plan
Docs-only.
pre-commit run --all-filesgreen.Compatibility / Risk
None (documentation). Adoption applies to future PRs; existing algorithms are untouched until a dedicated placement PR moves them.
AI-assisted; reviewed and directed by the maintainer.