{"id":"https://openalex.org/W4417439573","doi":"https://doi.org/10.48550/arxiv.2507.16323","title":"SpeLLM: Character-Level Multi-Head Decoding","display_name":"SpeLLM: Character-Level Multi-Head Decoding","publication_year":2025,"publication_date":"2025-07-22","ids":{"openalex":"https://openalex.org/W4417439573","doi":"https://doi.org/10.48550/arxiv.2507.16323"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2507.16323","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2507.16323","pdf_url":"https://arxiv.org/pdf/2507.16323","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2507.16323","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5114329082","display_name":"Amit Ben Artzy","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I197251160","display_name":"Hebrew University of Jerusalem","ror":"https://ror.org/03qxff017","country_code":"IL","type":"education","lineage":["https://openalex.org/I197251160"]}],"countries":["IL"],"is_corresponding":false,"raw_author_name":"Ben-Artzy, Amit","raw_affiliation_strings":["The Hebrew University of Jerusalem"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The Hebrew University of Jerusalem","institution_ids":["https://openalex.org/I197251160"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5007903277","display_name":"Roy Schwartz","orcid":"https://orcid.org/0000-0003-3487-5713","observed_orcids":["https://orcid.org/0000-0003-3487-5713"]},"institutions":[{"id":"https://openalex.org/I197251160","display_name":"Hebrew University of Jerusalem","ror":"https://ror.org/03qxff017","country_code":"IL","type":"education","lineage":["https://openalex.org/I197251160"]}],"countries":["IL"],"is_corresponding":false,"raw_author_name":"Schwartz, Roy","raw_affiliation_strings":["The Hebrew University of Jerusalem"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The Hebrew University of Jerusalem","institution_ids":["https://openalex.org/I197251160"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I197251160"],"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/T10028","display_name":"Topic Modeling","score":0.8490999937057495,"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"}},"topics":[{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.8490999937057495,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.13809999823570251,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.0031999999191612005,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/vocabulary-expansion","display_name":"vocabulary enrichment","score":0.5260000228881836},{"id":"https://openalex.org/keywords/self-distillation","display_name":"self-distillation","score":0.9480000138282776},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.675000011920929},{"id":"https://openalex.org/keywords/low-resource-languages","display_name":"low-resource languages","score":0.7950000166893005}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6456999778747559},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.6032000184059143},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.5723000168800354},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.5299000144004822},{"id":"https://openalex.org/C2777601683","wikidata":"https://www.wikidata.org/wiki/Q6499736","display_name":"Vocabulary","level":2,"score":0.5070000290870667},{"id":"https://openalex.org/C129844170","wikidata":"https://www.wikidata.org/wiki/Q41299","display_name":"Quadratic equation","level":2,"score":0.49309998750686646},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.3806999921798706},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.37940001487731934},{"id":"https://openalex.org/C152124472","wikidata":"https://www.wikidata.org/wiki/Q1204361","display_name":"Redundancy (engineering)","level":2,"score":0.3772999942302704},{"id":"https://openalex.org/C205711294","wikidata":"https://www.wikidata.org/wiki/Q176953","display_name":"Rendering (computer graphics)","level":2,"score":0.3702000081539154},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.31929999589920044},{"id":"https://openalex.org/C18648836","wikidata":"https://www.wikidata.org/wiki/Q381892","display_name":"Compact space","level":2,"score":0.31470000743865967},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.29440000653266907},{"id":"https://openalex.org/C2780861071","wikidata":"https://www.wikidata.org/wiki/Q1062934","display_name":"Character (mathematics)","level":2,"score":0.29280000925064087},{"id":"https://openalex.org/C2779227376","wikidata":"https://www.wikidata.org/wiki/Q6505497","display_name":"Layer (electronics)","level":2,"score":0.2849000096321106},{"id":"https://openalex.org/C207467116","wikidata":"https://www.wikidata.org/wiki/Q4385666","display_name":"Inverse","level":2,"score":0.28459998965263367},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.2720000147819519},{"id":"https://openalex.org/C2778572836","wikidata":"https://www.wikidata.org/wiki/Q380933","display_name":"Space (punctuation)","level":2,"score":0.27059999108314514},{"id":"https://openalex.org/C2776207758","wikidata":"https://www.wikidata.org/wiki/Q5303302","display_name":"Downstream (manufacturing)","level":2,"score":0.2632000148296356},{"id":"https://openalex.org/C127162648","wikidata":"https://www.wikidata.org/wiki/Q16858953","display_name":"Channel (broadcasting)","level":2,"score":0.2621999979019165}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2507.16323","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2507.16323","pdf_url":"https://arxiv.org/pdf/2507.16323","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2507.16323","doi":"https://doi.org/10.48550/arxiv.2507.16323","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2507.16323","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:2507.16323","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2507.16323","pdf_url":"https://arxiv.org/pdf/2507.16323","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4417439573.pdf","grobid_xml":"https://content.openalex.org/works/W4417439573.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7167971427","https://openalex.org/W7154347950","https://openalex.org/W4407371101","https://openalex.org/W4406059547","https://openalex.org/W4416034672","https://openalex.org/W7092286881","https://openalex.org/W4405561577","https://openalex.org/W4402672007","https://openalex.org/W7204772780","https://openalex.org/W4408354809"],"abstract_inverted_index":{"Scaling":[0],"LLM":[1,18,96,133],"vocabulary":[2,34],"is":[3],"often":[4],"used":[5],"to":[6,24,75,97],"reduce":[7],"input":[8,47],"sequence":[9],"length":[10],"and":[11,48,141],"alleviate":[12],"attention's":[13],"quadratic":[14],"cost.":[15],"Yet,":[16],"current":[17],"architectures":[19],"impose":[20],"a":[21,43,68,77,89,94,98,128],"critical":[22],"bottleneck":[23],"this":[25],"procedure:":[26],"the":[27,63,73],"output":[28,49,57,80],"projection":[29],"layer":[30],"scales":[31],"linearly":[32],"with":[33,102],"size,":[35],"rendering":[36],"substantial":[37],"expansion":[38],"impractical.":[39],"We":[40,87],"propose":[41],"SpeLLM,":[42,60],"method":[44],"that":[45],"decouples":[46],"vocabularies":[50],"by":[51,119],"predicting":[52],"character-level":[53],"strings":[54],"through":[55],"multiple":[56],"heads.":[58,86],"In":[59],"each":[61],"of":[62],"$k$":[64],"linear":[65,85],"heads":[66],"predicts":[67],"single":[69],"character":[70],"simultaneously,":[71],"enabling":[72],"model":[74],"represent":[76],"much":[78],"larger":[79],"space":[81],"using":[82],"smaller,":[83],"independent":[84],"present":[88],"self-distillation":[90],"approach":[91,126],"for":[92,131,138],"converting":[93],"standard":[95],"SpeLLM.":[99],"Our":[100,125],"experiments":[101],"four":[103],"pre-trained":[104],"LLMs":[105],"show":[106],"their":[107],"SpeLLM":[108],"variants":[109],"achieve":[110],"competitive":[111],"performance":[112],"on":[113,121],"downstream":[114],"tasks":[115],"while":[116,135],"reducing":[117,132],"runtime":[118],"5.1%":[120],"average":[122],"across":[123],"models.":[124],"provides":[127],"potential":[129],"avenue":[130],"costs,":[134],"increasing":[136],"support":[137],"underrepresented":[139],"languages":[140],"domains.":[142]},"counts_by_year":[],"updated_date":"2026-11-04T10:54:05.112534","created_date":"2025-10-10T00:00:00"}