{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:EYIGROPEF4Z3F7JKKGSVKO7G2F","short_pith_number":"pith:EYIGROPE","schema_version":"1.0","canonical_sha256":"261068b9e42f33b2fd2a51a5553be6d178197fff1d3b780168ac799c749b7383","source":{"kind":"arxiv","id":"2412.09871","version":1},"attestation_state":"computed","paper":{"title":"Byte Latent Transformer: Patches Scale Better Than Tokens","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Ari Holtzman, Artidoro Pagnoni, Benjamin Muller, Chunting Zhou, Gargi Ghosh, Jason Weston, John Nguyen, Lili Yu, Luke Zettlemoyer, Margaret Li, Mike Lewis, Pedro Rodriguez, Ram Pasunuru, Srinivasan Iyer","submitted_at":"2024-12-13T05:33:32Z","abstract_excerpt":"We introduce the Byte Latent Transformer (BLT), a new byte-level LLM architecture that, for the first time, matches tokenization-based LLM performance at scale with significant improvements in inference efficiency and robustness. BLT encodes bytes into dynamically sized patches, which serve as the primary units of computation. Patches are segmented based on the entropy of the next byte, allocating more compute and model capacity where increased data complexity demands it. We present the first FLOP controlled scaling study of byte-level models up to 8B parameters and 4T training bytes. Our resu"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2412.09871","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-12-13T05:33:32Z","cross_cats_sorted":[],"title_canon_sha256":"39903fd6b9c5d47d68b43eb21f4be23eaac7779998532d401e8c3e5804adddf4","abstract_canon_sha256":"7ee84f07476b54206d5e9c93169bdf8d500d197ac8dd1d70ca33e2263b81d32d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:48:43.086831Z","signature_b64":"C899/LCsEwtYM71C1Q6q1yK8kyr2/44OiyPWYvZe5SDgytu1XIDgQrYNydHfG0KBqtE/Qx8pSLGQuh1Qq/S4Ag==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"261068b9e42f33b2fd2a51a5553be6d178197fff1d3b780168ac799c749b7383","last_reissued_at":"2026-07-05T09:48:43.086337Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:48:43.086337Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Byte Latent Transformer: Patches Scale Better Than Tokens","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Ari Holtzman, Artidoro Pagnoni, Benjamin Muller, Chunting Zhou, Gargi Ghosh, Jason Weston, John Nguyen, Lili Yu, Luke Zettlemoyer, Margaret Li, Mike Lewis, Pedro Rodriguez, Ram Pasunuru, Srinivasan Iyer","submitted_at":"2024-12-13T05:33:32Z","abstract_excerpt":"We introduce the Byte Latent Transformer (BLT), a new byte-level LLM architecture that, for the first time, matches tokenization-based LLM performance at scale with significant improvements in inference efficiency and robustness. BLT encodes bytes into dynamically sized patches, which serve as the primary units of computation. Patches are segmented based on the entropy of the next byte, allocating more compute and model capacity where increased data complexity demands it. We present the first FLOP controlled scaling study of byte-level models up to 8B parameters and 4T training bytes. Our resu"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.09871","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2412.09871/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2412.09871","created_at":"2026-07-05T09:48:43.086399+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.09871v1","created_at":"2026-07-05T09:48:43.086399+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.09871","created_at":"2026-07-05T09:48:43.086399+00:00"},{"alias_kind":"pith_short_12","alias_value":"EYIGROPEF4Z3","created_at":"2026-07-05T09:48:43.086399+00:00"},{"alias_kind":"pith_short_16","alias_value":"EYIGROPEF4Z3F7JK","created_at":"2026-07-05T09:48:43.086399+00:00"},{"alias_kind":"pith_short_8","alias_value":"EYIGROPE","created_at":"2026-07-05T09:48:43.086399+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":19,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.05691","citing_title":"Where to cut, how deep: BPE and Unigram-LM on chemistry SMILES","ref_index":34,"is_internal_anchor":true},{"citing_arxiv_id":"2606.20993","citing_title":"Phonemes to the Rescue: Multilingual Tokenization Based on International Phonetic Alphabet","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2606.19172","citing_title":"User as Engram: Internalizing Per-User Memory as Local Parametric Edits","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2607.02087","citing_title":"SUNTA: Hierarchical Video Prediction with Surprise-based Chunking","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04171","citing_title":"MimeLens: Position-Agnostic Content-Type Detection for Binary Fragments","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2606.02834","citing_title":"Large Byte Model: Teaching Language Models About Compiled Code","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2606.27550","citing_title":"EntMTP: Accelerating LLM Inference with Entropy Guided Multi Token Prediction","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2606.20587","citing_title":"Protocol-Aware Tokenization and Architecture Co-Design for Wireless Packet Foundation Models","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28929","citing_title":"Cybersecurity is the True Frontier for Generative AI Success or Failure","ref_index":67,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29459","citing_title":"Kronecker Embeddings: Byte-Level Structured Token Representations for Parameter-Efficient Language Models","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2606.14122","citing_title":"Beyond Perplexity: UTF-8 Validity in Byte-aware Language Models","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2605.21070","citing_title":"Towards Understanding Self-Pretraining for Sequence Classification","ref_index":181,"is_internal_anchor":false},{"citing_arxiv_id":"2506.14123","citing_title":"Sampling from Your Language Model One Byte at a Time","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2507.12720","citing_title":"FLEXITOKENS: Flexible Tokenization for Evolving Language Models","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2510.04800","citing_title":"Hybrid Architectures for Language Models: Systematic Analysis and Design Insights","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2603.06351","citing_title":"DC-DiT: Adaptive Compute and Elastic Inference for Visual Generation via Dynamic Chunking","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11829","citing_title":"Bin Latent Transformer (BiLT): A shift-invariant autoencoder for calibration-free spectral unmixing of turbid media","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09630","citing_title":"Scratchpad Patching: Decoupling Compute from Patch Size in Byte-Level Language Models","ref_index":72,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00253","citing_title":"Lost in State Space: Probing Frozen Mamba Representations","ref_index":9,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EYIGROPEF4Z3F7JKKGSVKO7G2F","json":"https://pith.science/pith/EYIGROPEF4Z3F7JKKGSVKO7G2F.json","graph_json":"https://pith.science/api/pith-number/EYIGROPEF4Z3F7JKKGSVKO7G2F/graph.json","events_json":"https://pith.science/api/pith-number/EYIGROPEF4Z3F7JKKGSVKO7G2F/events.json","paper":"https://pith.science/paper/EYIGROPE"},"agent_actions":{"view_html":"https://pith.science/pith/EYIGROPEF4Z3F7JKKGSVKO7G2F","download_json":"https://pith.science/pith/EYIGROPEF4Z3F7JKKGSVKO7G2F.json","view_paper":"https://pith.science/paper/EYIGROPE","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.09871&json=true","fetch_graph":"https://pith.science/api/pith-number/EYIGROPEF4Z3F7JKKGSVKO7G2F/graph.json","fetch_events":"https://pith.science/api/pith-number/EYIGROPEF4Z3F7JKKGSVKO7G2F/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EYIGROPEF4Z3F7JKKGSVKO7G2F/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EYIGROPEF4Z3F7JKKGSVKO7G2F/action/storage_attestation","attest_author":"https://pith.science/pith/EYIGROPEF4Z3F7JKKGSVKO7G2F/action/author_attestation","sign_citation":"https://pith.science/pith/EYIGROPEF4Z3F7JKKGSVKO7G2F/action/citation_signature","submit_replication":"https://pith.science/pith/EYIGROPEF4Z3F7JKKGSVKO7G2F/action/replication_record"}},"created_at":"2026-07-05T09:48:43.086399+00:00","updated_at":"2026-07-05T09:48:43.086399+00:00"}