{"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/reinforcement-fine-tuning-naturally-mitigates","title":"Reinforcement Fine-Tuning Naturally Mitigates Forgetting in Continual Post-Training","arxiv_id":"2507.05386","date":"2025-07-07","proceeding":null,"authors":["Song Lai","Haohan Zhao","Rong Feng","Changyi Ma","Wenzhuo LIU","Hongbo Zhao","Xi Lin","Dong Yi","Min Xie","Qingfu Zhang","Hongbin Liu","Gaofeng Meng","Fei Zhu"],"abstract":"Continual post-training (CPT) is a popular and effective technique for adapting foundation models like multimodal large language models to specific and ever-evolving downstream tasks. While existing research has primarily concentrated on methods like data replay, model expansion, or parameter regularization, the fundamental role of the learning paradigm within CPT remains largely unexplored. This paper presents a comparative analysis of two core post-training paradigms: supervised fine-tuning (SFT) and reinforcement fine-tuning (RFT), investigating their respective impacts on knowledge retention during CPT. Our experiments are conducted on a benchmark comprising seven diverse multimodal tasks, utilizing Qwen2.5-VL-7B-Instruct as the base model for continual post-training. The investigation yields two significant findings: (1) When continuously learning on downstream tasks, SFT leads to catastrophic forgetting of previously learned tasks. In contrast, RFT inherently preserves prior knowledge and achieve performance comparable to multi-task training. (2) RFT successfully protects and even enhances the model's general knowledge on standard benchmarks (e.g., MMMU and MMLU-Pro). Conversely, SFT degrades general model capabilities severely. Further analysis shows that explicit mechanisms, such as KL penalty and chain-of-thought reasoning, are not the primary factors. Instead, we find that the implicit regularization inherent to RFT is a key factor in mitigating forgetting. Finally, we propose a rollout-based instance filtering algorithm to improve the stability and efficiency of RFT. Our comprehensive study demonstrates the superiority of RFT as a robust paradigm for continual post-training.","url_abs":"https://arxiv.org/abs/2507.05386v1","url_pdf":"https://arxiv.org/pdf/2507.05386v1.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":[],"tasks":[{"task_slug":"general-knowledge","task_name":"General Knowledge"},{"task_slug":"mmlu","task_name":"MMLU"}],"methods":[{"method_slug":"base","method_name":"BASE"},{"method_slug":"sft","method_name":"SFT"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2507.05386","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2507.05386"}},"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":"deterministic:regex_extraction","url":"https://github.com/zhhvvv/rft_vs_sft","reach":{"status":"ok"}}],"summary":{"ran":3,"unverified":7},"by_repo_kind":{"found_in_text":{"samples":10,"ran":3,"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":10,"samples":[{"code_sha256_prefix":"33a3cccef610855c","entry":"get_kl_controller","repo":"zhhvvv/rft_vs_sft","repo_kind":"found_in_text","path":"EasyR1/verl/trainer/core_algos.py","file_url":"https://github.com/zhhvvv/rft_vs_sft/blob/HEAD/EasyR1/verl/trainer/core_algos.py","link_basis":"plan_row","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":"33a3cccef610855c"}},{"code_sha256_prefix":"4049cf08f994464e","entry":"prepare_fa2_from_position_ids","repo":"zhhvvv/rft_vs_sft","repo_kind":"found_in_text","path":"EasyR1/verl/models/transformers/flash_attention_utils.py","file_url":"https://github.com/zhhvvv/rft_vs_sft/blob/HEAD/EasyR1/verl/models/transformers/flash_attention_utils.py","link_basis":"harvester_set","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":"4049cf08f994464e"}},{"code_sha256_prefix":"a91c18341292d50e","entry":"unpad_dataproto","repo":"zhhvvv/rft_vs_sft","repo_kind":"found_in_text","path":"EasyR1/verl/protocol.py","file_url":"https://github.com/zhhvvv/rft_vs_sft/blob/HEAD/EasyR1/verl/protocol.py","link_basis":"harvester_set","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":"a91c18341292d50e"}},{"code_sha256_prefix":"d3ee7b3d8f0cf2fa","entry":"compute_gae_advantage_return","repo":"zhhvvv/rft_vs_sft","repo_kind":"found_in_text","path":"EasyR1/verl/trainer/core_algos.py","file_url":"https://github.com/zhhvvv/rft_vs_sft/blob/HEAD/EasyR1/verl/trainer/core_algos.py","link_basis":"plan_row","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":"d3ee7b3d8f0cf2fa"}},{"code_sha256_prefix":"b7dacadfa17e35d2","entry":"compute_grpo_outcome_advantage","repo":"zhhvvv/rft_vs_sft","repo_kind":"found_in_text","path":"EasyR1/verl/trainer/core_algos.py","file_url":"https://github.com/zhhvvv/rft_vs_sft/blob/HEAD/EasyR1/verl/trainer/core_algos.py","link_basis":"plan_row","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":"b7dacadfa17e35d2"}},{"code_sha256_prefix":"2fba3c22367e0778","entry":"eval_logit_processor","repo":"zhhvvv/rft_vs_sft","repo_kind":"found_in_text","path":"LLaMA-Factory/src/llamafactory/train/sft/metric.py","file_url":"https://github.com/zhhvvv/rft_vs_sft/blob/HEAD/LLaMA-Factory/src/llamafactory/train/sft/metric.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":"2fba3c22367e0778"}},{"code_sha256_prefix":"d71e909038b965a1","entry":"flash_attention_forward","repo":"zhhvvv/rft_vs_sft","repo_kind":"found_in_text","path":"EasyR1/verl/models/transformers/flash_attention_utils.py","file_url":"https://github.com/zhhvvv/rft_vs_sft/blob/HEAD/EasyR1/verl/models/transformers/flash_attention_utils.py","link_basis":"plan_row","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":"d71e909038b965a1"}},{"code_sha256_prefix":"c5aeb7231bf64063","entry":"pad_dataproto_to_divisor","repo":"zhhvvv/rft_vs_sft","repo_kind":"found_in_text","path":"EasyR1/verl/protocol.py","file_url":"https://github.com/zhhvvv/rft_vs_sft/blob/HEAD/EasyR1/verl/protocol.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":"c5aeb7231bf64063"}},{"code_sha256_prefix":"b5b031edf82a3af8","entry":"process_for_parquet","repo":"zhhvvv/rft_vs_sft","repo_kind":"found_in_text","path":"rollout_filter.py","file_url":"https://github.com/zhhvvv/rft_vs_sft/blob/HEAD/rollout_filter.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":"b5b031edf82a3af8"}},{"code_sha256_prefix":"ad658a1d6fe829ba","entry":"union_tensor_dict","repo":"zhhvvv/rft_vs_sft","repo_kind":"found_in_text","path":"EasyR1/verl/protocol.py","file_url":"https://github.com/zhhvvv/rft_vs_sft/blob/HEAD/EasyR1/verl/protocol.py","link_basis":"harvester_set","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":"ad658a1d6fe829ba"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}