{"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/do-not-let-low-probability-tokens-over","title":"Do Not Let Low-Probability Tokens Over-Dominate in RL for LLMs","arxiv_id":"2505.12929","date":"2025-05-19","proceeding":null,"authors":["Zhihe Yang","Xufang Luo","Zilong Wang","Dongqi Han","Zhiyuan He","Dongsheng Li","Yunjian Xu"],"abstract":"Reinforcement learning (RL) has become a cornerstone for enhancing the reasoning capabilities of large language models (LLMs), with recent innovations such as Group Relative Policy Optimization (GRPO) demonstrating exceptional effectiveness. In this study, we identify a critical yet underexplored issue in RL training: low-probability tokens disproportionately influence model updates due to their large gradient magnitudes. This dominance hinders the effective learning of high-probability tokens, whose gradients are essential for LLMs' performance but are substantially suppressed. To mitigate this interference, we propose two novel methods: Advantage Reweighting and Low-Probability Token Isolation (Lopti), both of which effectively attenuate gradients from low-probability tokens while emphasizing parameter updates driven by high-probability tokens. Our approaches promote balanced updates across tokens with varying probabilities, thereby enhancing the efficiency of RL training. Experimental results demonstrate that they substantially improve the performance of GRPO-trained LLMs, achieving up to a 46.2% improvement in K&K Logic Puzzle reasoning tasks. Our implementation is available at https://github.com/zhyang2226/AR-Lopti.","url_abs":"https://arxiv.org/abs/2505.12929v1","url_pdf":"https://arxiv.org/pdf/2505.12929v1.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":"do-not-let-low-probability-tokens-over","repo_url":"https://github.com/zhyang2226/ar-lopti","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"Apache-2.0"}}],"tasks":[{"task_slug":"reinforcement-learning-1","task_name":"Reinforcement Learning (RL)"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=2505.12929","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2505.12929"}},"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/zhyang2226/ar-lopti","reach":{"status":"ok","spdx":"Apache-2.0"}}],"summary":{"ran":2,"ran_honours":1,"ran_fixture":1,"ran_draft_wrong":1,"unverified":9},"by_repo_kind":{"official":{"samples":14,"ran":5,"repositories":1}},"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":0,"samples":[{"code_sha256_prefix":"360c671d6041ba3b","entry":"create_huggingface_actor","repo":"zhyang2226/ar-lopti","repo_kind":"official","path":"verl/utils/model.py","file_url":"https://github.com/zhyang2226/ar-lopti/blob/HEAD/verl/utils/model.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":"360c671d6041ba3b"}},{"code_sha256_prefix":"216e28c040173a61","entry":"extract_step","repo":"zhyang2226/ar-lopti","repo_kind":"official","path":"verl/trainer/fsdp_sft_trainer.py","file_url":"https://github.com/zhyang2226/ar-lopti/blob/HEAD/verl/trainer/fsdp_sft_trainer.py","link_basis":"harvester_set","language":"python","status":"ran_honours","verification_level":1,"contract_check":"HONOURS","metamorphic_tier":"well_formed","behaviour_fingerprint":true,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"216e28c040173a61"}},{"code_sha256_prefix":"42a7954a8cfa89e0","entry":"is_transformers_version_in_range","repo":"zhyang2226/ar-lopti","repo_kind":"official","path":"verl/models/transformers/monkey_patch.py","file_url":"https://github.com/zhyang2226/ar-lopti/blob/HEAD/verl/models/transformers/monkey_patch.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":"42a7954a8cfa89e0"}},{"code_sha256_prefix":"21331a58f93375e4","entry":"union_tensor_dict","repo":"zhyang2226/ar-lopti","repo_kind":"official","path":"verl/protocol.py","file_url":"https://github.com/zhyang2226/ar-lopti/blob/HEAD/verl/protocol.py","link_basis":"harvester_set","language":"python","status":"ran_fixture","verification_level":1,"contract_check":"DEP_MISSING","metamorphic_tier":"deterministic","behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"21331a58f93375e4"}},{"code_sha256_prefix":"25f0ea3f460f6ce1","entry":"unpad_dataproto","repo":"zhyang2226/ar-lopti","repo_kind":"official","path":"verl/protocol.py","file_url":"https://github.com/zhyang2226/ar-lopti/blob/HEAD/verl/protocol.py","link_basis":"harvester_set","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":"25f0ea3f460f6ce1"}},{"code_sha256_prefix":"1fb4adb621eefddc","entry":"apply_monkey_patch","repo":"zhyang2226/ar-lopti","repo_kind":"official","path":"verl/models/transformers/monkey_patch.py","file_url":"https://github.com/zhyang2226/ar-lopti/blob/HEAD/verl/models/transformers/monkey_patch.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":"1fb4adb621eefddc"}},{"code_sha256_prefix":"13606d906fd1dfa3","entry":"extract_answers","repo":"zhyang2226/ar-lopti","repo_kind":"official","path":"dataset_format.py","file_url":"https://github.com/zhyang2226/ar-lopti/blob/HEAD/dataset_format.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":"13606d906fd1dfa3"}},{"code_sha256_prefix":"33ed081f30218012","entry":"extract_prompts","repo":"zhyang2226/ar-lopti","repo_kind":"official","path":"dataset_format.py","file_url":"https://github.com/zhyang2226/ar-lopti/blob/HEAD/dataset_format.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":"33ed081f30218012"}},{"code_sha256_prefix":"82115dc84c0c9c2e","entry":"find_first_eos_position","repo":"zhyang2226/ar-lopti","repo_kind":"official","path":"verl/trainer/fig_plot.py","file_url":"https://github.com/zhyang2226/ar-lopti/blob/HEAD/verl/trainer/fig_plot.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":"82115dc84c0c9c2e"}},{"code_sha256_prefix":"88bf147f4ef431b1","entry":"find_max_length","repo":"zhyang2226/ar-lopti","repo_kind":"official","path":"verl/trainer/fig_plot.py","file_url":"https://github.com/zhyang2226/ar-lopti/blob/HEAD/verl/trainer/fig_plot.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":"88bf147f4ef431b1"}},{"code_sha256_prefix":"19472d4343d0a082","entry":"get_huggingface_actor_config","repo":"zhyang2226/ar-lopti","repo_kind":"official","path":"verl/utils/model.py","file_url":"https://github.com/zhyang2226/ar-lopti/blob/HEAD/verl/utils/model.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":"19472d4343d0a082"}},{"code_sha256_prefix":"919cf310ad4c189a","entry":"get_weight_loader","repo":"zhyang2226/ar-lopti","repo_kind":"official","path":"verl/models/weight_loader_registry.py","file_url":"https://github.com/zhyang2226/ar-lopti/blob/HEAD/verl/models/weight_loader_registry.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":"919cf310ad4c189a"}},{"code_sha256_prefix":"3b15e2ac7497c441","entry":"squeeze","repo":"zhyang2226/ar-lopti","repo_kind":"official","path":"verl/utils/model.py","file_url":"https://github.com/zhyang2226/ar-lopti/blob/HEAD/verl/utils/model.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":"3b15e2ac7497c441"}},{"code_sha256_prefix":"bbb3cd7b7c99574a","entry":"wrap_text","repo":"zhyang2226/ar-lopti","repo_kind":"official","path":"verl/trainer/fig_plot.py","file_url":"https://github.com/zhyang2226/ar-lopti/blob/HEAD/verl/trainer/fig_plot.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":"bbb3cd7b7c99574a"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}