{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2020:HEST645VO3T7B32DIDGQMEM3QL","short_pith_number":"pith:HEST645V","canonical_record":{"source":{"id":"2004.03720","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2020-04-07T21:21:06Z","cross_cats_sorted":[],"title_canon_sha256":"ad254bfe2d5e8f58e712922c7f0411d020905e951a0eb1a7f0b6548d714ca9d7","abstract_canon_sha256":"44be804fc1905e2fc90800b42ba36f3537d86af3b704e0afa215ae7b1eafe2b5"},"schema_version":"1.0"},"canonical_sha256":"39253f73b576e7f0ef4340cd06119b82f6980e4527073e667894add414d233d9","source":{"kind":"arxiv","id":"2004.03720","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2004.03720","created_at":"2026-07-05T01:40:17Z"},{"alias_kind":"arxiv_version","alias_value":"2004.03720v2","created_at":"2026-07-05T01:40:17Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2004.03720","created_at":"2026-07-05T01:40:17Z"},{"alias_kind":"pith_short_12","alias_value":"HEST645VO3T7","created_at":"2026-07-05T01:40:17Z"},{"alias_kind":"pith_short_16","alias_value":"HEST645VO3T7B32D","created_at":"2026-07-05T01:40:17Z"},{"alias_kind":"pith_short_8","alias_value":"HEST645V","created_at":"2026-07-05T01:40:17Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2020:HEST645VO3T7B32DIDGQMEM3QL","target":"record","payload":{"canonical_record":{"source":{"id":"2004.03720","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2020-04-07T21:21:06Z","cross_cats_sorted":[],"title_canon_sha256":"ad254bfe2d5e8f58e712922c7f0411d020905e951a0eb1a7f0b6548d714ca9d7","abstract_canon_sha256":"44be804fc1905e2fc90800b42ba36f3537d86af3b704e0afa215ae7b1eafe2b5"},"schema_version":"1.0"},"canonical_sha256":"39253f73b576e7f0ef4340cd06119b82f6980e4527073e667894add414d233d9","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T01:40:17.266133Z","signature_b64":"/USku8LlA43PKOSoiZOfECOK4ZMQb5WHhEJC/eogfzdAe/WdUqt5etqCpoIC2geqbU18h4AdmjjQmn5jWz96AQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"39253f73b576e7f0ef4340cd06119b82f6980e4527073e667894add414d233d9","last_reissued_at":"2026-07-05T01:40:17.265654Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T01:40:17.265654Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2004.03720","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T01:40:17Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"QEYyifgqxGSXzN5WoY4gHycThj5HNOegaEvWv7Ee04p6GEZ1N0Kf2cnpy6vy7wzyj9eBNircf/jNNHAkKNffBQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-07T22:28:50.191178Z"},"content_sha256":"afe6ca2109efb929b551a4335040f46f45d44f57e1fda3d4b35e93c585725567","schema_version":"1.0","event_id":"sha256:afe6ca2109efb929b551a4335040f46f45d44f57e1fda3d4b35e93c585725567"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2020:HEST645VO3T7B32DIDGQMEM3QL","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Byte Pair Encoding is Suboptimal for Language Model Pretraining","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Greg Durrett, Kaj Bostrom","submitted_at":"2020-04-07T21:21:06Z","abstract_excerpt":"The success of pretrained transformer language models (LMs) in natural language processing has led to a wide range of pretraining setups. In particular, these models employ a variety of subword tokenization methods, most notably byte-pair encoding (BPE) (Sennrich et al., 2016; Gage, 1994), the WordPiece method (Schuster and Nakajima, 2012), and unigram language modeling (Kudo, 2018), to segment text. However, to the best of our knowledge, the literature does not contain a direct evaluation of the impact of tokenization on language model pretraining. We analyze differences between BPE and unigr"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2004.03720","kind":"arxiv","version":2},"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/2004.03720/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T01:40:17Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"pgjnVMtn0plywSZ42SASGroD3n0IRHv0DlHCYSLF6jj0YNZhMhsQTW5xju8OemRNlKoEP9pX+wqoA8lCFqO5CA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-07T22:28:50.191746Z"},"content_sha256":"74a442fcc7f9621cc0c7eaac4691c7218e9cef0974f0cb94887d5c125c608772","schema_version":"1.0","event_id":"sha256:74a442fcc7f9621cc0c7eaac4691c7218e9cef0974f0cb94887d5c125c608772"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/HEST645VO3T7B32DIDGQMEM3QL/bundle.json","state_url":"https://pith.science/pith/HEST645VO3T7B32DIDGQMEM3QL/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/HEST645VO3T7B32DIDGQMEM3QL/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-08-07T22:28:50Z","links":{"resolver":"https://pith.science/pith/HEST645VO3T7B32DIDGQMEM3QL","bundle":"https://pith.science/pith/HEST645VO3T7B32DIDGQMEM3QL/bundle.json","state":"https://pith.science/pith/HEST645VO3T7B32DIDGQMEM3QL/state.json","well_known_bundle":"https://pith.science/.well-known/pith/HEST645VO3T7B32DIDGQMEM3QL/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2020:HEST645VO3T7B32DIDGQMEM3QL","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"44be804fc1905e2fc90800b42ba36f3537d86af3b704e0afa215ae7b1eafe2b5","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2020-04-07T21:21:06Z","title_canon_sha256":"ad254bfe2d5e8f58e712922c7f0411d020905e951a0eb1a7f0b6548d714ca9d7"},"schema_version":"1.0","source":{"id":"2004.03720","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2004.03720","created_at":"2026-07-05T01:40:17Z"},{"alias_kind":"arxiv_version","alias_value":"2004.03720v2","created_at":"2026-07-05T01:40:17Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2004.03720","created_at":"2026-07-05T01:40:17Z"},{"alias_kind":"pith_short_12","alias_value":"HEST645VO3T7","created_at":"2026-07-05T01:40:17Z"},{"alias_kind":"pith_short_16","alias_value":"HEST645VO3T7B32D","created_at":"2026-07-05T01:40:17Z"},{"alias_kind":"pith_short_8","alias_value":"HEST645V","created_at":"2026-07-05T01:40:17Z"}],"graph_snapshots":[{"event_id":"sha256:74a442fcc7f9621cc0c7eaac4691c7218e9cef0974f0cb94887d5c125c608772","target":"graph","created_at":"2026-07-05T01:40:17Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2004.03720/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"The success of pretrained transformer language models (LMs) in natural language processing has led to a wide range of pretraining setups. In particular, these models employ a variety of subword tokenization methods, most notably byte-pair encoding (BPE) (Sennrich et al., 2016; Gage, 1994), the WordPiece method (Schuster and Nakajima, 2012), and unigram language modeling (Kudo, 2018), to segment text. However, to the best of our knowledge, the literature does not contain a direct evaluation of the impact of tokenization on language model pretraining. We analyze differences between BPE and unigr","authors_text":"Greg Durrett, Kaj Bostrom","cross_cats":[],"headline":"","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2020-04-07T21:21:06Z","title":"Byte Pair Encoding is Suboptimal for Language Model Pretraining"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2004.03720","kind":"arxiv","version":2},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:afe6ca2109efb929b551a4335040f46f45d44f57e1fda3d4b35e93c585725567","target":"record","created_at":"2026-07-05T01:40:17Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"44be804fc1905e2fc90800b42ba36f3537d86af3b704e0afa215ae7b1eafe2b5","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2020-04-07T21:21:06Z","title_canon_sha256":"ad254bfe2d5e8f58e712922c7f0411d020905e951a0eb1a7f0b6548d714ca9d7"},"schema_version":"1.0","source":{"id":"2004.03720","kind":"arxiv","version":2}},"canonical_sha256":"39253f73b576e7f0ef4340cd06119b82f6980e4527073e667894add414d233d9","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"39253f73b576e7f0ef4340cd06119b82f6980e4527073e667894add414d233d9","first_computed_at":"2026-07-05T01:40:17.265654Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T01:40:17.265654Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"/USku8LlA43PKOSoiZOfECOK4ZMQb5WHhEJC/eogfzdAe/WdUqt5etqCpoIC2geqbU18h4AdmjjQmn5jWz96AQ==","signature_status":"signed_v1","signed_at":"2026-07-05T01:40:17.266133Z","signed_message":"canonical_sha256_bytes"},"source_id":"2004.03720","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:afe6ca2109efb929b551a4335040f46f45d44f57e1fda3d4b35e93c585725567","sha256:74a442fcc7f9621cc0c7eaac4691c7218e9cef0974f0cb94887d5c125c608772"],"state_sha256":"e7017e372fc0f36bb47c902a8c02334c341c98d7a0a6c391f85764a69583e688"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"u/oN2Ba9VgTsJxrBMvtqQw0PVxTVfxOBzvHDs/KWOnKxqwvYn9N39bsPi5FkGWi8fVcNc0ka1UONnR1f7CSeCw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-07T22:28:50.197851Z","bundle_sha256":"6a54fe4f86c11ca65634d91b327ed65de7ee8481074951fd449a03bebdb63d0e"}}