{"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/self-training-elicits-concise-reasoning-in","title":"Self-Training Elicits Concise Reasoning in Large Language Models","arxiv_id":"2502.20122","date":"2025-02-27","proceeding":null,"authors":["Tergel Munkhbat","Namgyu Ho","Seohyun Kim","Yongjin Yang","Yujin Kim","Se-Young Yun"],"abstract":"Chain-of-thought (CoT) reasoning has enabled large language models (LLMs) to utilize additional computation through intermediate tokens to solve complex tasks. However, we posit that typical reasoning traces contain many redundant tokens, incurring extraneous inference costs. Upon examination of the output distribution of current LLMs, we find evidence on their latent ability to reason more concisely, relative to their default behavior. To elicit this capability, we propose simple fine-tuning methods which leverage self-generated concise reasoning paths obtained by best-of-N sampling and few-shot conditioning, in task-specific settings. Our combined method achieves a 30% reduction in output tokens on average, across five model families on GSM8K and MATH, while maintaining average accuracy. By exploiting the fundamental stochasticity and in-context learning capabilities of LLMs, our self-training approach robustly elicits concise reasoning on a wide range of models, including those with extensive post-training. Code is available at https://github.com/TergelMunkhbat/concise-reasoning","url_abs":"https://arxiv.org/abs/2502.20122v1","url_pdf":"https://arxiv.org/pdf/2502.20122v1.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":"self-training-elicits-concise-reasoning-in","repo_url":"https://github.com/tergelmunkhbat/concise-reasoning","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"gsm8k","task_name":"GSM8K"},{"task_slug":"in-context-learning","task_name":"In-Context Learning"},{"task_slug":"math","task_name":"Math"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=2502.20122","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2502.20122"}},"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/tergelmunkhbat/concise-reasoning","reach":{"status":"ok","spdx":"MIT"}}],"summary":{"ran":2,"unverified":14},"by_repo_kind":{"official":{"samples":16,"ran":2,"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":"6390dea763ce122e","entry":"extract_boxed_answers","repo":"tergelmunkhbat/concise-reasoning","repo_kind":"official","path":"src/math_parser.py","file_url":"https://github.com/tergelmunkhbat/concise-reasoning/blob/HEAD/src/math_parser.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"6390dea763ce122e"}},{"code_sha256_prefix":"5abed389274c4c03","entry":"strip_string","repo":"tergelmunkhbat/concise-reasoning","repo_kind":"official","path":"src/math_parser.py","file_url":"https://github.com/tergelmunkhbat/concise-reasoning/blob/HEAD/src/math_parser.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"5abed389274c4c03"}},{"code_sha256_prefix":"07c47b62bc5a9ca7","entry":"add_parsed_budget","repo":"tergelmunkhbat/concise-reasoning","repo_kind":"official","path":"src/estimated_budget_process.py","file_url":"https://github.com/tergelmunkhbat/concise-reasoning/blob/HEAD/src/estimated_budget_process.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"07c47b62bc5a9ca7"}},{"code_sha256_prefix":"8b057c6909544771","entry":"check_patterns","repo":"tergelmunkhbat/concise-reasoning","repo_kind":"official","path":"src/estimated_budget_process.py","file_url":"https://github.com/tergelmunkhbat/concise-reasoning/blob/HEAD/src/estimated_budget_process.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"8b057c6909544771"}},{"code_sha256_prefix":"46e84dfabd0695d0","entry":"compare_answers","repo":"tergelmunkhbat/concise-reasoning","repo_kind":"official","path":"src/preprocess.py","file_url":"https://github.com/tergelmunkhbat/concise-reasoning/blob/HEAD/src/preprocess.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"46e84dfabd0695d0"}},{"code_sha256_prefix":"5ee5c0c3dc69f851","entry":"filter_rationales","repo":"tergelmunkhbat/concise-reasoning","repo_kind":"official","path":"src/preprocess.py","file_url":"https://github.com/tergelmunkhbat/concise-reasoning/blob/HEAD/src/preprocess.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"5ee5c0c3dc69f851"}},{"code_sha256_prefix":"84abbf776f80eaf8","entry":"generate_text","repo":"tergelmunkhbat/concise-reasoning","repo_kind":"official","path":"src/training_utils.py","file_url":"https://github.com/tergelmunkhbat/concise-reasoning/blob/HEAD/src/training_utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"84abbf776f80eaf8"}},{"code_sha256_prefix":"98694e37b00066e2","entry":"get_config_dir","repo":"tergelmunkhbat/concise-reasoning","repo_kind":"official","path":"src/utils.py","file_url":"https://github.com/tergelmunkhbat/concise-reasoning/blob/HEAD/src/utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"98694e37b00066e2"}},{"code_sha256_prefix":"ed87ec15885fe0aa","entry":"get_latest_checkpoint","repo":"tergelmunkhbat/concise-reasoning","repo_kind":"official","path":"src/model.py","file_url":"https://github.com/tergelmunkhbat/concise-reasoning/blob/HEAD/src/model.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"ed87ec15885fe0aa"}},{"code_sha256_prefix":"8a0e38e1ecf7a971","entry":"get_train_dir","repo":"tergelmunkhbat/concise-reasoning","repo_kind":"official","path":"src/utils.py","file_url":"https://github.com/tergelmunkhbat/concise-reasoning/blob/HEAD/src/utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"8a0e38e1ecf7a971"}},{"code_sha256_prefix":"ab895046a21a97a7","entry":"load_model","repo":"tergelmunkhbat/concise-reasoning","repo_kind":"official","path":"src/model.py","file_url":"https://github.com/tergelmunkhbat/concise-reasoning/blob/HEAD/src/model.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"ab895046a21a97a7"}},{"code_sha256_prefix":"f60f5bc6df5e6841","entry":"load_model_with_flash_attention","repo":"tergelmunkhbat/concise-reasoning","repo_kind":"official","path":"src/model.py","file_url":"https://github.com/tergelmunkhbat/concise-reasoning/blob/HEAD/src/model.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"f60f5bc6df5e6841"}},{"code_sha256_prefix":"d00d21477f6cf430","entry":"parse_fraction","repo":"tergelmunkhbat/concise-reasoning","repo_kind":"official","path":"src/math_parser.py","file_url":"https://github.com/tergelmunkhbat/concise-reasoning/blob/HEAD/src/math_parser.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"d00d21477f6cf430"}},{"code_sha256_prefix":"3edbfe5817ad1b08","entry":"process_file","repo":"tergelmunkhbat/concise-reasoning","repo_kind":"official","path":"src/estimated_budget_process.py","file_url":"https://github.com/tergelmunkhbat/concise-reasoning/blob/HEAD/src/estimated_budget_process.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"3edbfe5817ad1b08"}},{"code_sha256_prefix":"c687c5d4e753d025","entry":"process_file","repo":"tergelmunkhbat/concise-reasoning","repo_kind":"official","path":"src/preprocess.py","file_url":"https://github.com/tergelmunkhbat/concise-reasoning/blob/HEAD/src/preprocess.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"c687c5d4e753d025"}},{"code_sha256_prefix":"0b6eb4d9aa2a6729","entry":"tokenize_function","repo":"tergelmunkhbat/concise-reasoning","repo_kind":"official","path":"src/training_utils.py","file_url":"https://github.com/tergelmunkhbat/concise-reasoning/blob/HEAD/src/training_utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"0b6eb4d9aa2a6729"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}