exp(glm45air): loss ablation configs (dppo / ipo / kimi-k1.5 scaffold)#2996
Draft
mikasenghaas wants to merge 1 commit into
Draft
exp(glm45air): loss ablation configs (dppo / ipo / kimi-k1.5 scaffold)#2996mikasenghaas wants to merge 1 commit into
mikasenghaas wants to merge 1 commit into
Conversation
…affold) Based on configs/glm45air/swe.toml, ablating [trainer.loss]: - swe_dppo.toml: DefaultLossConfig, symmetric dppo_mask_low=high=0.1 - swe_ipo.toml: IPOLossConfig, ipo_threshold=0.1 - swe_kimi.toml: CustomLossConfig scaffold — Kimi k1.5 loss is NOT implemented in prime-rl; needs a custom loss fn before it runs. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Loss-function ablations on
configs/glm45air/swe.toml(GLM-4.5-Air on scaleswe, eval on SWE-Bench-Verified). Each config only changes[trainer.loss](+ its own wandb name / job_name / sandbox label).configs/glm45air/swe_dppo.tomltype="default"), symmetricdppo_mask_low=high=0.1configs/glm45air/swe_ipo.tomltype="ipo"),ipo_threshold=0.1configs/glm45air/swe_kimi.tomltype="custom")import_pathpoints at akimi_k15_lossthat must be added firstDPPO and IPO are first-class loss types in
src/prime_rl/trainer/rl/loss.py(default_loss_fn/ipo_loss_fn) — pure config. Kimi k1.5 needs a newLossInputs→LossOutputsfn (config wired via thecustomloss type, left as a flagged TODO).