Papers › REBEL: Reinforcement Learning via Regressing Relative Rewards

REBEL: Reinforcement Learning via Regressing Relative Rewards

25 Apr 2024arXiv:2404.16767archive 2025-07-28

Zhaolin Gao, Jonathan D. Chang, Wenhao Zhan, Owen Oertell, Gokul Swamy, Kianté Brantley, Thorsten Joachims, J. Andrew Bagnell, Jason D. Lee, Wen Sun

While originally developed for continuous control problems, Proximal Policy Optimization (PPO) has emerged as the work-horse of a variety of reinforcement learning (RL) applications, including the fine-tuning of generative models. Unfortunately, PPO requires multiple heuristics to enable stable convergence (e.g. value networks, clipping), and is notorious for its sensitivity to the precise implementation of these components. In response, we take a step back and ask what a minimalist RL algorithm for the era of generative models would look like. We propose REBEL, an algorithm that cleanly reduces the problem of policy optimization to regressing the relative reward between two completions to a prompt in terms of the policy, enabling strikingly lightweight implementation. In theory, we prove that fundamental RL algorithms like Natural Policy Gradient can be seen as variants of REBEL, which allows us to match the strongest known theoretical guarantees in terms of convergence and sample complexity in the RL literature. REBEL can also cleanly incorporate offline data and be extended to handle the intransitive preferences we frequently see in practice. Empirically, we find that REBEL provides a unified approach to language modeling and image generation with stronger or similar performance as PPO and DPO, all while being simpler to implement and more computationally efficient than PPO. When fine-tuning Llama-3-8B-Instruct, REBEL achieves strong performance in AlpacaEval 2.0, MT-Bench, and Open LLM Leaderboard.

PaperPDFCodeCode Syntology ran

In Syntology View this paper on Syntology, its page in Syntology's graph. That page adds which of its functions are also present in other papers' code Syntology holds, with dates, and its dated run record as JSON and a README badge. It also lists the repositories linked to the paper, the abstract and the calls for agents.

Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

For agents, on Syntology's MCP service (how to connect):

Code

Syntology 20 code samples harvested from 2 repositories linked to this paper: 16 ran (of which 0 constructed an object rather than computing a result; 8 with no instrument failure: 2 honoured, 1 violated, 5 with no contract checked; 8 where Syntology's instrument failed) · 4 unverified. Where Syntology's instrument failed: 7 where our contract draft was wrong; 1 where the contract check could not run; failures of Syntology's instrument, not of the code.

By repository: official repository: 20 samples from 2 repositories, 16 ran (8 with no instrument failure; 8 where Syntology's instrument failed). The run record, sample by sample. “Ran” means executed on a synthesized input, not that the code is correct or reproduces the paper.

No sample is linked to this paper from 1 of its repositories that Syntology harvested for other papers:

Levels written between 2026-09-22 and 2026-09-24 (UTC): the dates Syntology's graph wrote each sample's verification level, not when the code executed.

Owen-Oertell/rlcm officialmentioned in papermentioned on GitHubpytorchSyntology: 6 samples, 5 ran (3 instrument-failed) · 1 unverified report
zhaolingao/rebel officialmentioned in papermentioned on GitHubpytorchApache-2.0Syntology: 14 samples, 11 ran (5 instrument-failed) · 3 unverified report
ziyizhang27/sdpo mentioned on GitHubpytorchSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to Aligning Few-Step Diffusion Models with Dense Reward Difference Learning (arXiv:2411.11727; that paper's own run record: 0 ran · 4 unverified); nothing checks that they implement this paper's method. report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

A paper named beside a repository with no sample linked to this paper is shown with that paper's own run record, not this paper's: “ran” means executed on a synthesized input, not that the code is correct, and “where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code. Papers are listed in arXiv-id order, at most three per repository.

Code Syntology ran Syntology

20 samples harvested: 16 ran (of which 0 constructed an object rather than computing a result; 8 with no instrument failure: 2 honoured, 1 violated, 5 with no contract checked; 8 where Syntology's instrument failed) · 4 unverified. “Where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code. Syntology's record for this page last changed ; when this build read Syntology's graph is in the build record. That date is not when the samples ran.

2ran · honoured contract
1ran · violated contract
7ran · our draft was wrong
1ran; the contract check could not run
5ran
4unverified

Licence: 6 of the 20 samples are pointer only, meaning Syntology does not serve that copy's text. This page shows no code text for any sample; each one links to its file in the repository.

Harvested from 2 repositories linked to this paper, official or community; the list below is grouped by repository, each group headed by its name and what it is to the paper. “Ran” means the sample executed on a synthesized input. It does not mean the output is correct, and nothing here reproduces the paper's results. A line beginning “fixture:” under a sample that ran gives the arguments Syntology's fixture called the sample with, each as name:type/rank/element kind (? where none was recorded), and the type of what the fixture's run returned, exactly as the harness recorded them; the arguments and the returned type were recorded by two separate runs of the same fixture. No output value is shown, and a returned type says nothing about whether the output was right. “Level recorded” gives the UTC date when Syntology's graph wrote the sample's verification level, not when the code executed; where the level was written before Syntology's graph kept that date, the row says “level recorded, date not recorded”. “Honoured” and “violated” refer to a contract Syntology drafted from the code itself. “Our draft was wrong” and “the contract check could not run” are failures of Syntology's instrument, not of the code: the sample ran on Syntology's synthesized input, and then the drafted contract did not describe it, or the check itself raised an error (RAISES), lacked a dependency (DEP_MISSING) or ran out of time (TIMEOUT); hover the label for which. A metamorphic tier, where shown, is evidence of the output's shape, not that the sample implements the method.

Each sample ends with its code_sha256. A sample's code_sha256 is Syntology's identity for that exact code. It is printed in full (16 hexadecimal characters for every sample in this build), and get_code(code_sha256="…") takes it exactly as printed. An agent calls it on Syntology's MCP service (how to connect); click an identity to copy that call.

Repository labels, per sample. official repository: The archive marks this repository official for the paper. named in the paper: The archive records that the paper mentions this repository; it is not marked official. community (archive-listed): In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper. found in paper text by Syntology: Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted. community: Not in the archive's code links for this paper; a community repository Syntology harvested. Samples are grouped by repository: a repository marked official first, then by how directly the repository is tied to the paper, then in the archive's order; never by how many ran. Licence labels name the repository's licence as recorded at harvest. “Pointer only” means Syntology does not serve that copy's text, for one of six reasons: no licence file was found; the licence was not identified; the licence is recorded as permissive but that copy's record is not marked cleared; the licence is outside the permissive list Syntology serves text under (MIT, Apache-2.0, BSD and similar); this paper's copy was not located, so its licence is not recorded; or this paper's copy is cleared, but the stored copy of the same code that Syntology serves is not. Some licences outside that list permit redistribution, such as WTFPL, and GPL-3.0 under its conditions; they are simply not on the list. Hover a licence label for the reason. File links open the file on GitHub at the default branch, which may have changed since the harvest.

A function or class name links to its page when the name occurs in the harvested code of at least 5 papers; the name is a string, not an identity. A census tag appears when Syntology's census ran implementations with that name on one shared input per bucket: it gives how many were compared, from how many papers, and how many distinct outputs they produced, and names this sample's bucket when this sample was one of them.

Owen-Oertell/rlcm official repository 6 samples, 5 ran (3 instrument-failed) · 1 unverified
betas_for_alpha_bar in 123 papers lcm_rl_pytorch/diffusers_patch/lcm_scheduler.py ran · honoured contract no licence file found · pointer only · f1e3f7493aad4745 · report fixture: signature num_diffusion_timesteps:int/0/int, max_beta:float/0/float, alpha_transform_type:str/0/str · returned Tensor · level recorded 2026-09-22
compressibility lcm_rl_pytorch/rewards/rewards.py ran · our draft was wrong no licence file found · pointer only · 1cf3d9a4597a653f · report fixture: signature cfg:dict/0/? · returned function · level recorded 2026-09-24
get_model_prediction lcm_rl_pytorch/util/model.py ran no licence file found · pointer only · fc7498a95a58a987 · report fixture: signature pipeline:callable/0/?, latents:Tensor/2/float, timesteps:Tensor/1/int, w_embedding:Tensor/2/float, prompt_embeds:Tensor/3/float · returned Tensor · level recorded 2026-09-24
incompressibility lcm_rl_pytorch/rewards/rewards.py ran · our draft was wrong no licence file found · pointer only · 11c329fa52d72b13 · report fixture: signature cfg:dict/0/? · returned function · level recorded 2026-09-24
rescale_zero_terminal_snr in 14 papers lcm_rl_pytorch/diffusers_patch/lcm_scheduler.py ran · our draft was wrong fingerprinted no licence file found · pointer only · 64b2196de5dcf1e5 · report fixture: signature betas:Tensor/1/float · returned Tensor · level recorded 2026-09-22
aesthetic lcm_rl_pytorch/rewards/rewards.py unverified no licence file found · pointer only · 44735222f08d757a · report
zhaolingao/rebel official repository 14 samples, 11 ran (5 instrument-failed) · 3 unverified
filter_same_responses src/ultrafeedback_largebatch/filter_tokenize.py ran · violated contract Apache-2.0 (permissive) · b44d4582c73da402 · report fixture: signature row:dict/0/? · returned bool · level recorded 2026-09-24
first_true_indices in 6 papers src/nectar/rebel.py ran · honoured contract Apache-2.0 (permissive) · 579f0202bb30e003 · report fixture: signature bools:Tensor/2/bool, dtype:dtype/0/? · returned Tensor · level recorded 2026-09-24
generate in 143 papers
census: 5 implementations from 6 papers, 5 distinct outputs across 4 buckets, one shared input per bucket
src/tldr/sft.py ran Apache-2.0 (permissive) · 51062aca9db3f84c · report fixture: signature lm_backbone:callable/0/?, queries:Tensor/2/int, tokenizer:SimpleTokenizer/0/?, generation_config:SimpleGenerationConfig/0/? · returned Tensor · level recorded 2026-09-24
get_cosine_schedule_with_warmup in 42 papers src/ultrafeedback_largebatch/rebel.py ran Apache-2.0 (permissive) · 70c923c8290a4fcc · report fixture: signature optimizer:SGD/0/?, num_warmup_steps:int/0/int, num_training_steps:int/0/int, num_cycles:float/0/float, last_epoch:int/0/int · returned LambdaLR · level recorded 2026-09-24
get_message in 8 papers src/ultrafeedback_largebatch/filter_tokenize.py ran · our draft was wrong fingerprinted Apache-2.0 (permissive) · dc68fb166b4b7b89 · report fixture: signature instruction:str/0/str, response:str/0/str · returned list · level recorded 2026-09-24
get_message in 8 papers src/ultrafeedback_largebatch/generate.py ran · our draft was wrong fingerprinted Apache-2.0 (permissive) · 93805966dfd5681a · report fixture: signature instruction:str/0/str · returned list · level recorded 2026-09-24
get_message in 8 papers src/ultrafeedback_largebatch/rank.py ran fingerprinted Apache-2.0 (permissive) · 689d710104523822 · report fixture: signature instruction:str/0/str, response:str/0/str · returned list · level recorded 2026-09-24
get_reward in 20 papers src/tldr/rebel.py ran · our draft was wrong Apache-2.0 (permissive) · 724222affceacea6 · report fixture: signature model:callable/0/?, query_responses:Tensor/2/int, tokenizer:SimpleTokenizer/0/?, context_length:int/0/int · returned tuple · level recorded 2026-09-22
get_reward in 20 papers src/tldr/rm.py ran Apache-2.0 (permissive) · c02add7716c931ba · report fixture: signature model:callable/0/?, query_responses:Tensor/2/int, tokenizer:SimpleTokenizer/0/? · returned Tensor · level recorded 2026-09-24
layer_init in 46 papers src/tldr/rebel.py ran · our draft was wrong Apache-2.0 (permissive) · 9ad5922df265477f · report fixture: signature layer:callable/0/?, std:float/0/float, bias_const:float/0/float · returned Linear · level recorded 2026-09-22
whiten in 15 papers
census: 5 implementations from 6 papers, 3 distinct outputs across 3 buckets, one shared input per bucket this sample: bucket 2 of 3, arg 1: rank 2, kind float, dtype float32
src/nectar/rebel.py ran; the contract check could not run (RAISES) · metamorphic tier: invariant fingerprinted Apache-2.0 (permissive) · e78dac11b89dad63 · report fixture: signature values:Tensor/2/float, shift_mean:bool/0/bool · returned Tensor · level recorded 2026-09-22
evaluate in 514 papers src/tldr/rm.py unverified Apache-2.0 (permissive) · d58b842cdc7fda2b · report
gather_logprob src/ultrafeedback_largebatch/rebel.py unverified Apache-2.0 (permissive) · b45408e42f101791 · report
truncate_response src/nectar/rebel.py unverified Apache-2.0 (permissive) · 62ece7427179dea2 · report

Tasks

Continuous ControlImage GenerationLanguage ModelingLanguage ModellingReinforcement LearningReinforcement Learning (RL)continuous-controlreinforcement-learning

Results from the paper archive 2025-07-28

No leaderboard rows for this paper in the archive.

Methods

DPOEntropy RegularizationPPO

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections