File size: 10,001 Bytes
818282c | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 | {
"audited_at": "2026-07-26T05:38:59Z",
"post_run1_correction": {
"active_run1_process_or_shards_changed": false,
"behavior": "Known Hub schemas select their real path column and fail closed if it is missing or empty. A leading StarCoderData reponame metadata line is removed only for syntax parsing, while the original text remains the training payload.",
"corpus_script": {
"path": "scripts/corpus.py",
"sha256": "05afba561a6258812d2baa802456c29c00d30955bec7f5c46389b90b2dad9e93"
},
"effective_scope": "future_source_streaming_only",
"quality_gate_test": {
"path": "scripts/test_corpus_quality.py",
"sha256": "b90f72e30d0eb5040ded59aaf4ae5a7a4a90c8951bbc66fe4fceb0cdd2ba5eed"
}
},
"publication_documents": {
"data_document": {
"path": "DATA.md",
"sha256": "fac5dce1b46f29821e3ed509332922ba395d4f3a1660b2e95929e94393e25ec7"
},
"model_card_template": {
"path": "MODEL_CARD.md",
"sha256": "ecd9611c82d8a05da30f36ef41b5ec0642ec7a4e6137c20f5951d84a51d55d9f"
}
},
"registered_run1_config": {
"path": "config/run1.json",
"sha256": "f1683a7b00bf1b1e93654ebb366fa0072ba6092b0dbad10272e9353196c3532d"
},
"registered_run2_config": {
"path": "config/run2_no_fim.json",
"sha256": "5692f15944dcef9a8a64a96ea5b5888b39024de321c2bb9a302f1c522bc3ea12"
},
"required_publication_disclosures": [
"Do not claim that every Python or JSON training document passed an extension parser.",
"Do not claim that the run 1 shards were locally verified as permissive-only.",
"State that original StarCoderData repository terms and relevant attribution clauses still apply.",
"State that FineWeb-Edu is ODC-By 1.0 and remains subject to Common Crawl terms.",
"State that Apache 2.0 covers the Wisp artifact and does not override source-data or generated-code terms.",
"State that the tokenizer sampled source entries round-robin by document rather than using the configured training-token weights.",
"Do not claim that deterministic run 1 token normalization proves exact original examples or boundaries.",
"State that run 1 applied fim_rate 0.7 independently per tokenized chunk, not per source document or sampled training window."
],
"run1_fim_application": {
"build_log": {
"path": "evidence/run1_corpus_build.log",
"sha256": "ef5de5c46aac1ff601158b43a3cde481ae6090ea04eefc772c531ff2ac78295e"
},
"build_source": {
"git_blob_sha1": "18b7e158ecec3467be28e1b18a5bab72c0ee1c77",
"git_commit": "a534de4d542167bdcea8adfda8fbf25d6cd0db44",
"path": "scripts/prepare_data.py",
"sha256": "6ebbd49a92de87582c429e2c0a5e2fd22792b7db1cbf44e37651b4eceaef7ff6"
},
"configured_rate_is_per_source_document": false,
"configured_rate_is_per_training_window": false,
"configured_transform_probability_per_chunk": 0.7,
"maximum_chunk_tokens": 1024,
"selected_orderings": {
"psm_probability": 0.5,
"spm_probability": 0.5
},
"selection_unit": "tokenized_chunk",
"statement": "Run 1 split each tokenized source document into chunks of at most 1024 tokens and selected FIM independently for each chunk. The configured 0.7 is not a per-document or per-window rate.",
"training_window_tokens": 2051
},
"run1_final_build_evidence": {
"final_index": {
"path": "data/shards/index.json",
"sha256": "862b1a9b7cc6c3c0d31299e21b352e2b736de767a99bf7fa38213d6c60fc0db0"
},
"log": {
"path": "evidence/run1_corpus_build.log",
"sha256": "ef5de5c46aac1ff601158b43a3cde481ae6090ea04eefc772c531ff2ac78295e"
},
"realized_train_percent": {
"fineweb_edu": 7.993028,
"implementation_code": 84.997663,
"markdown": 7.009309,
"starcoderdata_including_markdown": 92.006972
},
"realized_train_tokens": {
"HuggingFaceFW/fineweb-edu:sample-10BT": 400290897,
"bigcode/starcoderdata:c": 250363883,
"bigcode/starcoderdata:go": 451081764,
"bigcode/starcoderdata:java": 400525337,
"bigcode/starcoderdata:javascript": 650722190,
"bigcode/starcoderdata:markdown": 351026251,
"bigcode/starcoderdata:python": 1200566506,
"bigcode/starcoderdata:rust": 451511705,
"bigcode/starcoderdata:shell": 150445965,
"bigcode/starcoderdata:sql": 100765073,
"bigcode/starcoderdata:typescript": 600701168
},
"scope": "Exact aggregate train-token totals only. Row identities, rejection counts, per-source validation overshoot, and row-level obligations remain unavailable.",
"total_train_tokens": 5008000739
},
"run1_provenance_limitations": {
"exact_original_unit_boundaries_proven": false,
"locally_verified_permissive_only": false,
"per_row_attribution_index_available": false,
"per_row_license_mapping_preserved": false,
"per_source_validation_counts_recorded": false,
"realized_per_source_train_tokens_recovered_from_final_log": true,
"source_revisions_captured_during_build": false,
"source_row_metadata_preserved_in_shards": false,
"statement": "Run 1 preserves configured source weights, exact aggregate train-token totals from the recovered final build log, post-build revision evidence, and a post-build hash of every current shard. It does not preserve ordered raw rows, repository paths, rejection counts, per-source validation overshoot, per-row licenses, attribution mapping, or exact original unit boundaries needed for a local permissive-only and example-exact audit."
},
"run1_quality_gate": {
"configured_path_field": null,
"corpus_script_at_build_sha256": "7180e0d69a543fa2ddcf76ef6fa035a14bab2f7e0dfc8a31413b416ad891886e",
"extension_parser_activated_for_hub_rows": false,
"path_field_used": "path",
"reason": "The run 1 iterator requested the absent path column, so Hub rows reached the structural filters without a file extension. The Python ast.parse and JSON json.loads branches therefore did not activate.",
"starcoderdata_path_field": "max_stars_repo_path",
"structural_filters_applied": true
},
"run1_shard_integrity": {
"attestation_kind": "post_build_current_bytes_and_visible_grammar",
"boundary_recovery": {
"detectable_reassembly_groups": 27,
"exact_original_units_proven": false,
"mode": "deterministic_visible_grammar_normalization",
"restored_internal_eos_tokens": 33
},
"normalized_splits": {
"train": {
"derived_tokens": 4992043184,
"fim_units": 5319185,
"source_tokens": 5008000739,
"units": 7629643
},
"val": {
"derived_tokens": 19949502,
"fim_units": 18870,
"source_tokens": 20006112,
"units": 27087
}
},
"receipt": {
"path": "config/run1_shard_integrity_receipt.json",
"sha256": "5831ecd4a471fbe07e19b212bc3de44bed0b0b6b456083e888f66802937bf471"
},
"scheduled_run2_training_positions": 4999872512,
"source_bytes": 10056013702,
"source_files": 52,
"source_index_sha256": "862b1a9b7cc6c3c0d31299e21b352e2b736de767a99bf7fa38213d6c60fc0db0"
},
"schema_version": 1,
"sources": {
"HuggingFaceFW/fineweb-edu": {
"configured_weight": 0.08,
"current_main_revision_at_audit": "87f09149ef4734204d70ed1d046ddc9ca3f2b8f9",
"dataset_card": {
"license_label": "odc-by",
"sha256": "a0cc8998a20499432b28b6575f3046b714938eb8e11b8d59a1d25ddf3716061e",
"terms": "The dataset is distributed under ODC-By 1.0 and remains subject to Common Crawl terms.",
"url": "https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu/blob/87f09149ef4734204d70ed1d046ddc9ca3f2b8f9/README.md"
},
"observed_stream_row_schema": {
"canonical_fields_sha256": "a7b0323d3e758514f936736e75a919bda456e98164299c1c1ce5970f65678f91",
"fields": [
"dump",
"file_path",
"id",
"int_score",
"language",
"language_score",
"score",
"text",
"token_count",
"url"
],
"subset": "sample-10BT"
},
"post_build_cache_ref_revision": "87f09149ef4734204d70ed1d046ddc9ca3f2b8f9"
},
"bigcode/starcoderdata": {
"configured_weight": 0.92,
"current_main_revision_at_audit": "9fc30b578cedaec69e47302df72cf00feed7c8c4",
"dataset_card": {
"license_label": "other",
"sha256": "7a3e42cc82fb48b6b81f2ef06eab94af33e605eff743c6a4b8a3b1852ced7c0a",
"terms": "Original repository licenses apply, including attribution clauses when relevant. Users must follow the source dataset update and removal terms.",
"url": "https://huggingface.co/datasets/bigcode/starcoderdata/blob/9fc30b578cedaec69e47302df72cf00feed7c8c4/README.md"
},
"observed_stream_row_schema": {
"canonical_fields_sha256": "ddfa03121c2f5e5766eada883df62dcaef2a04540a49d849692831ae81fbddd4",
"fields": [
"content",
"id",
"max_stars_count",
"max_stars_repo_name",
"max_stars_repo_path"
],
"subset": "python"
},
"post_build_cache_ref_revision": "9fc30b578cedaec69e47302df72cf00feed7c8c4"
}
},
"status": "limitations_registered",
"tokenizer_sampling": {
"build_log": {
"path": "evidence/tokenizer_build.log",
"sha256": "7c28f91dc527e0cc37d23c520ba183aef845f948b64ea822f3b3ca17de264467"
},
"configured_token_weights_applied": false,
"documents_requested": 400000,
"exact_row_manifest_preserved": false,
"source_entries": 11,
"statement": "The tokenizer sample included FineWeb-Edu and sampled source entries evenly by document, not according to the later training-token weights.",
"strategy": "round_robin_by_configured_source_entry",
"tokenizer": {
"path": "tokenizer/code32k.json",
"sha256": "401a28c1f079050c48f6438830ca772d161d897e3cf2f30588d9ddc587dc6081"
}
}
}
|