{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/rebel-reinforcement-learning-via-regressing","title":"REBEL: Reinforcement Learning via Regressing Relative Rewards","arxiv_id":"2404.16767","date":"2024-04-25","proceeding":null,"authors":["Zhaolin Gao","Jonathan D. Chang","Wenhao Zhan","Owen Oertell","Gokul Swamy","Kianté Brantley","Thorsten Joachims","J. Andrew Bagnell","Jason D. Lee","Wen Sun"],"abstract":"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.","url_abs":"https://arxiv.org/abs/2404.16767v4","url_pdf":"https://arxiv.org/pdf/2404.16767v4.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"rebel-reinforcement-learning-via-regressing","repo_url":"https://github.com/Owen-Oertell/rlcm","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"rebel-reinforcement-learning-via-regressing","repo_url":"https://github.com/zhaolingao/rebel","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"Apache-2.0"}},{"paper_slug":"rebel-reinforcement-learning-via-regressing","repo_url":"https://github.com/ziyizhang27/sdpo","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"continuous-control","task_name":"Continuous Control"},{"task_slug":"image-generation","task_name":"Image Generation"},{"task_slug":"language-modeling","task_name":"Language Modeling"},{"task_slug":"language-modelling","task_name":"Language Modelling"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"},{"task_slug":"reinforcement-learning-1","task_name":"Reinforcement Learning (RL)"},{"task_slug":"continuous-control","task_name":"continuous-control"},{"task_slug":"reinforcement-learning-2","task_name":"reinforcement-learning"}],"methods":[{"method_slug":"dpo","method_name":"DPO"},{"method_slug":"entropy-regularization","method_name":"Entropy Regularization"},{"method_slug":"ppo","method_name":"PPO"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2404.16767","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2404.16767"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/ziyizhang27/sdpo","reach":null},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/Owen-Oertell/rlcm","reach":{"status":"ok"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/zhaolingao/rebel","reach":{"status":"ok","spdx":"Apache-2.0"}}],"summary":{"ran_honours":1,"ran":11,"ran_draft_wrong":3,"ran_fixture":1,"unverified":4},"by_repo_kind":{"official":{"samples":20,"ran":16,"repositories":2}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"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"},"n_pointer_only_for_licence":6,"samples":[{"code_sha256_prefix":"f1e3f7493aad4745","entry":"betas_for_alpha_bar","repo":"Owen-Oertell/rlcm","repo_kind":"official","path":"lcm_rl_pytorch/diffusers_patch/lcm_scheduler.py","file_url":"https://github.com/Owen-Oertell/rlcm/blob/HEAD/lcm_rl_pytorch/diffusers_patch/lcm_scheduler.py","link_basis":"harvester_set","language":"python","status":"ran_honours","verification_level":1,"contract_check":"HONOURS","metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"f1e3f7493aad4745"}},{"code_sha256_prefix":"1cf3d9a4597a653f","entry":"compressibility","repo":"Owen-Oertell/rlcm","repo_kind":"official","path":"lcm_rl_pytorch/rewards/rewards.py","file_url":"https://github.com/Owen-Oertell/rlcm/blob/HEAD/lcm_rl_pytorch/rewards/rewards.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"1cf3d9a4597a653f"}},{"code_sha256_prefix":"b44d4582c73da402","entry":"filter_same_responses","repo":"zhaolingao/rebel","repo_kind":"official","path":"src/ultrafeedback_largebatch/filter_tokenize.py","file_url":"https://github.com/zhaolingao/rebel/blob/HEAD/src/ultrafeedback_largebatch/filter_tokenize.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"b44d4582c73da402"}},{"code_sha256_prefix":"579f0202bb30e003","entry":"first_true_indices","repo":"zhaolingao/rebel","repo_kind":"official","path":"src/nectar/rebel.py","file_url":"https://github.com/zhaolingao/rebel/blob/HEAD/src/nectar/rebel.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"579f0202bb30e003"}},{"code_sha256_prefix":"51062aca9db3f84c","entry":"generate","repo":"zhaolingao/rebel","repo_kind":"official","path":"src/tldr/sft.py","file_url":"https://github.com/zhaolingao/rebel/blob/HEAD/src/tldr/sft.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"51062aca9db3f84c"}},{"code_sha256_prefix":"70c923c8290a4fcc","entry":"get_cosine_schedule_with_warmup","repo":"zhaolingao/rebel","repo_kind":"official","path":"src/ultrafeedback_largebatch/rebel.py","file_url":"https://github.com/zhaolingao/rebel/blob/HEAD/src/ultrafeedback_largebatch/rebel.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"70c923c8290a4fcc"}},{"code_sha256_prefix":"dc68fb166b4b7b89","entry":"get_message","repo":"zhaolingao/rebel","repo_kind":"official","path":"src/ultrafeedback_largebatch/filter_tokenize.py","file_url":"https://github.com/zhaolingao/rebel/blob/HEAD/src/ultrafeedback_largebatch/filter_tokenize.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"dc68fb166b4b7b89"}},{"code_sha256_prefix":"93805966dfd5681a","entry":"get_message","repo":"zhaolingao/rebel","repo_kind":"official","path":"src/ultrafeedback_largebatch/generate.py","file_url":"https://github.com/zhaolingao/rebel/blob/HEAD/src/ultrafeedback_largebatch/generate.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"93805966dfd5681a"}},{"code_sha256_prefix":"689d710104523822","entry":"get_message","repo":"zhaolingao/rebel","repo_kind":"official","path":"src/ultrafeedback_largebatch/rank.py","file_url":"https://github.com/zhaolingao/rebel/blob/HEAD/src/ultrafeedback_largebatch/rank.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"689d710104523822"}},{"code_sha256_prefix":"fc7498a95a58a987","entry":"get_model_prediction","repo":"Owen-Oertell/rlcm","repo_kind":"official","path":"lcm_rl_pytorch/util/model.py","file_url":"https://github.com/Owen-Oertell/rlcm/blob/HEAD/lcm_rl_pytorch/util/model.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"fc7498a95a58a987"}},{"code_sha256_prefix":"724222affceacea6","entry":"get_reward","repo":"zhaolingao/rebel","repo_kind":"official","path":"src/tldr/rebel.py","file_url":"https://github.com/zhaolingao/rebel/blob/HEAD/src/tldr/rebel.py","link_basis":"plan_row","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"MISDECLARED","metamorphic_tier":"deterministic","behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"724222affceacea6"}},{"code_sha256_prefix":"c02add7716c931ba","entry":"get_reward","repo":"zhaolingao/rebel","repo_kind":"official","path":"src/tldr/rm.py","file_url":"https://github.com/zhaolingao/rebel/blob/HEAD/src/tldr/rm.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"c02add7716c931ba"}},{"code_sha256_prefix":"11c329fa52d72b13","entry":"incompressibility","repo":"Owen-Oertell/rlcm","repo_kind":"official","path":"lcm_rl_pytorch/rewards/rewards.py","file_url":"https://github.com/Owen-Oertell/rlcm/blob/HEAD/lcm_rl_pytorch/rewards/rewards.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"11c329fa52d72b13"}},{"code_sha256_prefix":"9ad5922df265477f","entry":"layer_init","repo":"zhaolingao/rebel","repo_kind":"official","path":"src/tldr/rebel.py","file_url":"https://github.com/zhaolingao/rebel/blob/HEAD/src/tldr/rebel.py","link_basis":"plan_row","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":"deterministic","behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"9ad5922df265477f"}},{"code_sha256_prefix":"64b2196de5dcf1e5","entry":"rescale_zero_terminal_snr","repo":"Owen-Oertell/rlcm","repo_kind":"official","path":"lcm_rl_pytorch/diffusers_patch/lcm_scheduler.py","file_url":"https://github.com/Owen-Oertell/rlcm/blob/HEAD/lcm_rl_pytorch/diffusers_patch/lcm_scheduler.py","link_basis":"plan_row","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"MISDECLARED","metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"64b2196de5dcf1e5"}},{"code_sha256_prefix":"e78dac11b89dad63","entry":"whiten","repo":"zhaolingao/rebel","repo_kind":"official","path":"src/nectar/rebel.py","file_url":"https://github.com/zhaolingao/rebel/blob/HEAD/src/nectar/rebel.py","link_basis":"harvester_set","language":"python","status":"ran_fixture","verification_level":1,"contract_check":"RAISES","metamorphic_tier":"invariant","behaviour_fingerprint":true,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"e78dac11b89dad63"}},{"code_sha256_prefix":"44735222f08d757a","entry":"aesthetic","repo":"Owen-Oertell/rlcm","repo_kind":"official","path":"lcm_rl_pytorch/rewards/rewards.py","file_url":"https://github.com/Owen-Oertell/rlcm/blob/HEAD/lcm_rl_pytorch/rewards/rewards.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"44735222f08d757a"}},{"code_sha256_prefix":"d58b842cdc7fda2b","entry":"evaluate","repo":"zhaolingao/rebel","repo_kind":"official","path":"src/tldr/rm.py","file_url":"https://github.com/zhaolingao/rebel/blob/HEAD/src/tldr/rm.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"d58b842cdc7fda2b"}},{"code_sha256_prefix":"b45408e42f101791","entry":"gather_logprob","repo":"zhaolingao/rebel","repo_kind":"official","path":"src/ultrafeedback_largebatch/rebel.py","file_url":"https://github.com/zhaolingao/rebel/blob/HEAD/src/ultrafeedback_largebatch/rebel.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"b45408e42f101791"}},{"code_sha256_prefix":"62ece7427179dea2","entry":"truncate_response","repo":"zhaolingao/rebel","repo_kind":"official","path":"src/nectar/rebel.py","file_url":"https://github.com/zhaolingao/rebel/blob/HEAD/src/nectar/rebel.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"62ece7427179dea2"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}