label-free-rl-post-training-reasoning-models-deriving-each-model-s CLM-0582In label-free RL post-training of reasoning models, deriving each model's reward from a separately-trained peer model's majority vote — rather than from its own majority vote — avoids the reward collapse and length degeneration that self-rewarding runs show, and the benefit grows as the paired models differ in family or size because their errors overlap less.
Capability: Getting more from several models than from one
Observed on
1.7B-12B open LLMs and VLMs trained with GRPO on unlabeled math/multimodal-math prompts; pseudo-labels from majority vote over K rollouts..
Sources
- Validated by: Average gains of 3.0-8.6% over base across seven text benchmarks and 2.3-7.2% across four multimodal benchmarks, beating TTRL, RENT, Intuitor, Co-rewarding-II, MAPoRL and CoMAS (+4.0% over CoMAS with half the agents), with a compute-matched control where two TTRL-trained models are ensembled at inference.. Measured on 3B/7B/8B LLMs and 2B-12B VLMs with published baselines and a budget-matched ensemble control; [truncated]
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- For small instruction-tuned models (3-4B) answering closed-form true/false questions under a stated user opinion, scoring each sampled response by how much more common its answer is than the group itself predicted, and using that score as the GRPO reward, reduces answer flips under user pressure and raises accuracy without any labels, roughly matching label-supervised desycophancy fine-tuning at higher compute cost.Telling the user what they want to hear · reviewed by AI
- For small instruction-tuned models (3-4B) answering closed-form true/false questions under a stated user opinion, scoring each sampled response by how much more common its answer is than the group itself predicted, and using that score as the GRPO reward, reduces answer flips under user pressure and raises accuracy without any labels, roughly matching label-supervised desycophancy fine-tuning at higher compute cost.Telling the user what they want to hear · reviewed by AI
- Used to steer test-time compute, generative LLM judges (7B-70B) rerank candidate responses about as well as outcome reward models, are consistently worse than process reward models at step-level beam search, and their natural-language critiques, the one thing only a judge can produce, did not guide generators to better responses on math, code or instruction following.Judging another's work as well as a careful reviewer would · reviewed by AI
- For 8B-14B generative reward models trained as judges, making the model generate input-adaptive evaluation criteria inside its own reasoning chain and rewarding the quality of those criteria with a stronger teacher model raises agreement with preference labels, and the gain scales with the teacher's capability rather than with the presence of the rubric step alone.Judging another's work as well as a careful reviewer would · reviewed by AI
- For agentic retrieval-augmented multi-hop QA with 7B-class policy models, scoring each intermediate step for both logical validity and grounding in the retrieved passages — rather than scoring steps by the final answer — yields higher EM/F1 than outcome-reward and final-answer-based process-reward training, because it removes preference pairs where flawed retrieval happened to reach the right answer.Carrying a chain of inference through to its conclusion · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: medium. Falsifier as drafted: A run where peer-rewarded cohorts collapse at the same rate as self-rewarding, or where same-family pairs match cross-family pairs on pseudo-label accuracy and final score once compute is matched.