{"id":"https://openalex.org/W4415315847","doi":"https://doi.org/10.48550/arxiv.2509.12464","title":"Reasoning Models Can be Accurately Pruned Via Chain-of-Thought Reconstruction","display_name":"Reasoning Models Can be Accurately Pruned Via Chain-of-Thought Reconstruction","publication_year":2025,"publication_date":"2025-09-15","ids":{"openalex":"https://openalex.org/W4415315847","doi":"https://doi.org/10.48550/arxiv.2509.12464"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2509.12464","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2509.12464","pdf_url":"https://arxiv.org/pdf/2509.12464","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2509.12464","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5100535184","display_name":"Ryan Lucas","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I63966007","display_name":"Massachusetts Institute of Technology","ror":"https://ror.org/042nb2s44","country_code":"US","type":"education","lineage":["https://openalex.org/I63966007"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Lucas, Ryan","raw_affiliation_strings":["Massachusetts Institute of Technology , Cambridge , MA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Massachusetts Institute of Technology , Cambridge , MA","institution_ids":["https://openalex.org/I63966007"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5075423316","display_name":"Kayhan Behdin","orcid":"https://orcid.org/0000-0003-3482-0421","observed_orcids":["https://orcid.org/0000-0003-3482-0421"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Behdin, Kayhan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100424178","display_name":"Zhipeng Wang","orcid":"https://orcid.org/0000-0003-4632-9170","observed_orcids":["https://orcid.org/0000-0003-4632-9170"]},"institutions":[{"id":"https://openalex.org/I63966007","display_name":"Massachusetts Institute of Technology","ror":"https://ror.org/042nb2s44","country_code":"US","type":"education","lineage":["https://openalex.org/I63966007"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Wang, Zhipeng","raw_affiliation_strings":["Massachusetts Institute of Technology , Cambridge , MA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Massachusetts Institute of Technology , Cambridge , MA","institution_ids":["https://openalex.org/I63966007"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5003977430","display_name":"Qingquan Song","orcid":"https://orcid.org/0000-0002-4307-0725","observed_orcids":["https://orcid.org/0000-0002-4307-0725"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Song, Qingquan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5056385133","display_name":"Shao Tang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tang, Shao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5045271820","display_name":"Rahul Mazumder","orcid":"https://orcid.org/0000-0003-1384-9743","observed_orcids":["https://orcid.org/0000-0003-1384-9743","https://orcid.org/0000-0003-4285-7400"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Mazumder, Rahul","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.9545999765396118,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.9545999765396118,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.02370000071823597,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.00279999990016222,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/chain-of-thought-reasoning","display_name":"chain-of-thought reasoning","score":0.7979999780654907},{"id":"https://openalex.org/keywords/neural-network-pruning","display_name":"neural network pruning","score":0.9190000295639038},{"id":"https://openalex.org/keywords/deepseek-r1","display_name":"DeepSeek-R1","score":0.6690000295639038},{"id":"https://openalex.org/keywords/inference-cost-reduction","display_name":"inference cost reduction","score":0.6940000057220459},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.4869999885559082}],"concepts":[{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.8287000060081482},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8015000224113464},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.6557000279426575},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.5999000072479248},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5914000272750854},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5533000230789185},{"id":"https://openalex.org/C177212765","wikidata":"https://www.wikidata.org/wiki/Q627335","display_name":"Workflow","level":2,"score":0.5085999965667725},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.4934000074863434},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4812000095844269},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.4666000008583069},{"id":"https://openalex.org/C2983448237","wikidata":"https://www.wikidata.org/wiki/Q1078276","display_name":"Language understanding","level":2,"score":0.33239999413490295},{"id":"https://openalex.org/C180016635","wikidata":"https://www.wikidata.org/wiki/Q2712821","display_name":"Compression (physics)","level":2,"score":0.3321000039577484},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.29679998755455017},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.2937999963760376},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2915000021457672},{"id":"https://openalex.org/C43126263","wikidata":"https://www.wikidata.org/wiki/Q128751","display_name":"Source code","level":2,"score":0.2669999897480011},{"id":"https://openalex.org/C2984842247","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep neural networks","level":3,"score":0.2556000053882599},{"id":"https://openalex.org/C2780586882","wikidata":"https://www.wikidata.org/wiki/Q7520643","display_name":"Simple (philosophy)","level":2,"score":0.2531000077724457}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2509.12464","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2509.12464","pdf_url":"https://arxiv.org/pdf/2509.12464","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2509.12464","doi":"https://doi.org/10.48550/arxiv.2509.12464","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2509.12464","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2509.12464","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2509.12464","pdf_url":"https://arxiv.org/pdf/2509.12464","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4415315847.pdf","grobid_xml":"https://content.openalex.org/works/W4415315847.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7168326986","https://openalex.org/W7221849804","https://openalex.org/W4417002136","https://openalex.org/W4415192111","https://openalex.org/W4414812677","https://openalex.org/W4415024500","https://openalex.org/W7125833972","https://openalex.org/W4416949876","https://openalex.org/W7126081047","https://openalex.org/W7166860003"],"abstract_inverted_index":{"Reasoning":[0],"language":[1,39],"models":[2],"such":[3,27,122],"as":[4,28,123],"DeepSeek-R1":[5],"produce":[6,57],"long":[7],"chain-of-thought":[8,110],"traces":[9],"during":[10,97],"inference":[11],"time":[12],"which":[13],"make":[14,47],"them":[15],"costly":[16],"to":[17,56,72],"deploy":[18],"at":[19],"scale.":[20],"We":[21,65,91],"show":[22,66],"that":[23,67,75],"using":[24],"compression":[25],"techniques":[26],"neural":[29],"network":[30],"pruning":[31,78,98,120],"produces":[32],"greater":[33],"performance":[34,128],"loss":[35],"than":[36],"in":[37,43,134],"typical":[38],"modeling":[40],"tasks,":[41],"and":[42,106,125],"some":[44],"cases":[45],"can":[46,137],"the":[48,54,73,104,107,132,135],"model":[49,55],"slower":[50],"since":[51],"they":[52],"cause":[53],"more":[58],"thinking":[59],"tokens":[60],"but":[61],"with":[62],"worse":[63],"performance.":[64],"this":[68],"is":[69,87],"partly":[70],"due":[71],"fact":[74],"standard":[76],"LLM":[77],"methods":[79],"often":[80],"focus":[81],"on":[82],"input":[83,105],"reconstruction,":[84],"whereas":[85],"reasoning":[86],"a":[88,93],"decode-dominated":[89],"task.":[90],"introduce":[92],"simple,":[94],"drop-in":[95],"fix:":[96],"we":[99],"jointly":[100],"reconstruct":[101],"activations":[102],"from":[103],"model's":[108],"on-policy":[109],"traces.":[111],"This":[112],"\"Reasoning-Aware":[113],"Compression\"":[114],"(RAC)":[115],"integrates":[116],"seamlessly":[117],"into":[118],"existing":[119],"workflows":[121],"SparseGPT,":[124],"boosts":[126],"their":[127],"significantly.":[129],"Code":[130],"reproducing":[131],"results":[133],"paper":[136],"be":[138],"found":[139],"at:":[140],"https://github.com/RyanLucas3/RAC":[141]},"counts_by_year":[],"updated_date":"2026-10-16T09:59:22.235173","created_date":"2025-10-18T00:00:00"}