Solayman-Sameh commited on
Commit
b2fd5df
·
verified ·
1 Parent(s): 6f44509

Upload export.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. export.json +82 -0
export.json ADDED
@@ -0,0 +1,82 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "distilbert",
3
+ "labels": [
4
+ "O",
5
+ "B-PER",
6
+ "I-PER",
7
+ "B-ORG",
8
+ "I-ORG",
9
+ "B-LOC",
10
+ "I-LOC",
11
+ "B-MISC",
12
+ "I-MISC"
13
+ ],
14
+ "length_policy": {
15
+ "max_subwords": 256,
16
+ "word_policy": "No recurrent word truncation; packed dynamic batches",
17
+ "char_max": 61,
18
+ "char_policy": "Train max; held-out/app longer chars clipped, count reported",
19
+ "transformer_policy": "Greedy whole-word chunks; no overlap; label-independent; reconstruct before scoring",
20
+ "app_recurrent_chunk_words": 256,
21
+ "app_max_characters": 20000,
22
+ "app_sentence_policy": "raw-v2: acronyms/titles and hyphenated words retained; format/control characters omitted from inference with original offsets preserved; standalone .!? and newlines split sentences"
23
+ },
24
+ "selection": {
25
+ "winner": "distilbert",
26
+ "best_validation_f1": 0.9391641388608022,
27
+ "eligible_within_0.001": [
28
+ "distilbert"
29
+ ],
30
+ "rule": "Strict validation micro F1; within absolute 0.001 prefer fewer trainable parameters, then measured validation seconds, then stable model name",
31
+ "experiment_id": "8a1b26004cfaa53318dee7fbb6289ba971e3152c3aaaf7171b08086838f4aac2",
32
+ "mode": "full",
33
+ "checkpoint_manifests": {
34
+ "distilbert": "b09581ff4d98bc34ec88817dc6357df0e38f57e33262f84ed7fca251888eca30",
35
+ "lstm": "55fd4158f08af157b3466c47fc5c2ed50f295a29f7e6e390107d699146733e5a",
36
+ "bilstm": "ed628fb397d73a3d130e5c91460d0b340402dcbea621cc3a7898e9d62606d04a",
37
+ "bilstm_crf": "81979fc4eff8545034b555f1c4906f0e785bcdc70a257df0668a3a2cb85ebb5a"
38
+ }
39
+ },
40
+ "experiment_id": "8a1b26004cfaa53318dee7fbb6289ba971e3152c3aaaf7171b08086838f4aac2",
41
+ "preprocessing": "Preserve casing/punctuation/word order; train-only vocab; lowercase only external GloVe lookup. No POS/chunk features.",
42
+ "span_policy": "Strict IOB2; invalid I predictions ignored rather than repaired. Half-open original character offsets.",
43
+ "app_tokenizer": "raw-v2: acronyms/titles and hyphenated words retained; format/control characters omitted from inference with original offsets preserved; standalone .!? and newlines split sentences",
44
+ "benchmark_tokenization_note": "Raw-text regex is deterministic but differs from human CoNLL tokenization; exported token API preserves benchmark tokens exactly.",
45
+ "embedding_provenance": {
46
+ "num_records": 400000,
47
+ "file_size": 69182535,
48
+ "base_dataset": "Wikipedia 2014 + Gigaword 5 (6B tokens, uncased)",
49
+ "reader_code": "https://github.com/RaRe-Technologies/gensim-data/releases/download/glove-wiki-gigaword-50/__init__.py",
50
+ "license": "http://opendatacommons.org/licenses/pddl/",
51
+ "parameters": {
52
+ "dimension": 50
53
+ },
54
+ "description": "Pre-trained vectors based on Wikipedia 2014 + Gigaword, 5.6B tokens, 400K vocab, uncased (https://nlp.stanford.edu/projects/glove/).",
55
+ "preprocessing": "Converted to w2v format with `python -m gensim.scripts.glove2word2vec -i <fname> -o glove-wiki-gigaword-50.txt`.",
56
+ "read_more": [
57
+ "https://nlp.stanford.edu/projects/glove/",
58
+ "https://nlp.stanford.edu/pubs/glove.pdf"
59
+ ],
60
+ "checksum": "c289bc5d7f2f02c6dc9f2f9b67641813",
61
+ "file_name": "glove-wiki-gigaword-50.gz",
62
+ "parts": 1
63
+ },
64
+ "training_mode": "full",
65
+ "full_training_results": true,
66
+ "inference_update": {
67
+ "revision": "raw-v2",
68
+ "date": "2026-11-10",
69
+ "training_experiment_unchanged": true,
70
+ "previous_app_archive": {
71
+ "bytes": 242761064,
72
+ "sha256": "079a09e2e8239c28f8818b414ba7810e8e51433016f7098d00f0dcae12c986b1"
73
+ },
74
+ "source_sha256": {
75
+ "common.py": "33b5d2a9dbb97e86ad46cf79c41433f8525ac76494ff9966f908f456d408e3a3",
76
+ "inference.py": "0286bfa88af230330333402052bc6ddc58581f212fc039e78cc32889ba1e418f",
77
+ "app.py": "09e7845cdfd24e85eca8cb34a5f043e8fee7a047866fe892352944fabe4acc30",
78
+ "models.py": "05519b51d548b7cdca9df3248d51d44e0916937f0711fb4ce3913bc0a6a25cb0"
79
+ },
80
+ "verification": "All 6 raw-text and offline CPU checkpoint regression tests passed, including former crashes, full acronym/hyphen boundaries, offsets and long input. Updated Kaggle training has not been rerun."
81
+ }
82
+ }