@mailwoman/neural-weights-fr-fr 3.0.0 → 4.1.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (5) hide show
  1. package/README.md +10 -10
  2. package/model-card.json +77 -55
  3. package/model.onnx +0 -0
  4. package/package.json +18 -19
  5. package/tokenizer.model +0 -0
package/README.md CHANGED
@@ -4,26 +4,26 @@ Stage 2 (coarse + venue/street/house_number) Mailwoman neural-classifier weights
4
4
 
5
5
  - locale: **fr-fr**
6
6
  - corpus: **0.3.0**
7
- - training steps: **1800**
7
+ - training steps: **2200**
8
8
  - hardware: **AMD Radeon 780M (gfx1103) bf16 ~14.6 GiB GTT**
9
9
 
10
10
  ## Per-component F1 targets
11
11
 
12
12
  **⚠ Below per-component F1 targets:**
13
13
 
14
- - `country` F1 = **0.2796** (target ≥0.95)
15
- - `region` F1 = **0.1759** (target ≥0.95)
16
- - `locality` F1 = **0.2657** (target ≥0.95)
17
- - `postcode` F1 = **0.7554** (target ≥0.95)
18
- - `venue` F1 = **0.3941** (target ≥0.60)
19
- - `street` F1 = **0.2660** (target ≥0.70)
20
- - `house_number` F1 = **0.7835** (target ≥0.80)
14
+ - `country` F1 = **0.2112** (target ≥0.95)
15
+ - `region` F1 = **0.1883** (target ≥0.95)
16
+ - `locality` F1 = **0.2736** (target ≥0.95)
17
+ - `postcode` F1 = **0.6916** (target ≥0.95)
18
+ - `venue` F1 = **0.3886** (target ≥0.60)
19
+ - `street` F1 = **0.3016** (target ≥0.70)
20
+ - `house_number` F1 = **0.7866** (target ≥0.80)
21
21
 
22
22
  ## Eval (golden set)
23
23
 
24
24
  - entries: **4535**
25
- - full-parse exact match: **0.1074**
26
- - mean token confidence: **0.8566**
25
+ - full-parse exact match: **0.0818**
26
+ - mean token confidence: **0.8063**
27
27
 
28
28
  ## Components supported
29
29
 
package/model-card.json CHANGED
@@ -1,17 +1,16 @@
1
1
  {
2
2
  "name": "neural-weights-fr-fr",
3
- "version": "3.0.0",
4
- "phase": "Stage 2 (coarse + venue/street/house_number)",
3
+ "version": "4.1.0",
4
+ "model_lineage": "shares the en-us v0.9.7-unit-v3 multi-locale model (step 20000) — shipped as the unified 4.1.0 release version; tokenizer 0.6.0-a0. fr-fr and en-us bundle the SAME byte-identical model; this package carries the FR-specific calibration table.",
5
+ "phase": "Stage 3 — multi-locale (FR via the shared model)",
5
6
  "license": "AGPL-3.0-only",
6
7
  "locale": "fr-fr",
7
8
  "training": {
8
- "corpus_version": "0.3.0",
9
- "tokenizer_version": "0.1.0",
10
- "steps": 1800,
11
- "hardware": "AMD Radeon 780M (gfx1103) bf16 ~14.6 GiB GTT",
12
- "duration_seconds": 1067.0,
13
- "started_at": null,
14
- "completed_at": "2026-05-22T10:22:18.739046Z"
9
+ "corpus_version": "0.4.5-unit-v2",
10
+ "tokenizer_version": "0.6.0-a0",
11
+ "steps": 20000,
12
+ "hardware": "NVIDIA A100-SXM4-40GB (Modal cloud)",
13
+ "note": "Identical artifact to neural-weights-en-us 4.1.0 (one multi-locale model serves both). See that card's training/recipe for full provenance."
15
14
  },
16
15
  "components_supported": [
17
16
  "country",
@@ -25,39 +24,62 @@
25
24
  "street",
26
25
  "house_number"
27
26
  ],
27
+ "labels": [
28
+ "O",
29
+ "B-country",
30
+ "I-country",
31
+ "B-region",
32
+ "I-region",
33
+ "B-locality",
34
+ "I-locality",
35
+ "B-dependent_locality",
36
+ "I-dependent_locality",
37
+ "B-postcode",
38
+ "I-postcode",
39
+ "B-subregion",
40
+ "I-subregion",
41
+ "B-cedex",
42
+ "I-cedex",
43
+ "B-venue",
44
+ "I-venue",
45
+ "B-street",
46
+ "I-street",
47
+ "B-house_number",
48
+ "I-house_number"
49
+ ],
28
50
  "eval": {
29
51
  "n_entries": 4535,
30
- "full_parse_exact_match": 0.1073869900771775,
31
- "mean_token_confidence": 0.856554333787935,
52
+ "full_parse_exact_match": 0.08180815876515987,
53
+ "mean_token_confidence": 0.8062812768727202,
32
54
  "per_component": {
33
55
  "country": {
34
- "precision": 0.2954545454532025,
35
- "recall": 0.2653061224478967,
36
- "f1": 0.2795698919733611,
56
+ "precision": 0.21428571428481394,
57
+ "recall": 0.20816326530527282,
58
+ "f1": 0.21118012372283307,
37
59
  "support": 245
38
60
  },
39
61
  "region": {
40
- "precision": 0.4411027568916778,
41
- "recall": 0.10982839313569116,
42
- "f1": 0.17586809860649805,
62
+ "precision": 0.342951360263526,
63
+ "recall": 0.129797191887635,
64
+ "f1": 0.18832050661831204,
43
65
  "support": 3205
44
66
  },
45
67
  "locality": {
46
- "precision": 0.23303370786511615,
47
- "recall": 0.3089067619897799,
48
- "f1": 0.26565902346259507,
68
+ "precision": 0.24782398452605223,
69
+ "recall": 0.30533214179317686,
70
+ "f1": 0.2735886822759171,
49
71
  "support": 3357
50
72
  },
51
73
  "dependent_locality": {
52
- "precision": 0.0,
53
- "recall": 0.0,
54
- "f1": 0.0,
74
+ "precision": 0.005044136191670815,
75
+ "recall": 0.0999999999975,
76
+ "f1": 0.009603841445164863,
55
77
  "support": 40
56
78
  },
57
79
  "postcode": {
58
- "precision": 0.8426270136303831,
59
- "recall": 0.684563758389032,
60
- "f1": 0.7554156632710455,
80
+ "precision": 0.8323890462696731,
81
+ "recall": 0.591610738254835,
82
+ "f1": 0.6916437813892687,
61
83
  "support": 2980
62
84
  },
63
85
  "subregion": {
@@ -73,21 +95,21 @@
73
95
  "support": 1
74
96
  },
75
97
  "venue": {
76
- "precision": 0.38622493461169466,
77
- "recall": 0.4023614895545846,
78
- "f1": 0.39412811337886233,
98
+ "precision": 0.37649063032335905,
99
+ "recall": 0.4014532243411431,
100
+ "f1": 0.38857142807160183,
79
101
  "support": 1101
80
102
  },
81
103
  "street": {
82
- "precision": 0.3326499231161801,
83
- "recall": 0.22165300546440517,
84
- "f1": 0.2660381220860398,
104
+ "precision": 0.35594795539016916,
105
+ "recall": 0.26161202185783416,
106
+ "f1": 0.3015748026611547,
85
107
  "support": 2928
86
108
  },
87
109
  "house_number": {
88
- "precision": 0.7315737051789185,
89
- "recall": 0.8432835820890682,
90
- "f1": 0.7834666661687646,
110
+ "precision": 0.7446153846150028,
111
+ "recall": 0.8335246842704744,
112
+ "f1": 0.7865655466300883,
91
113
  "support": 1742
92
114
  }
93
115
  },
@@ -101,56 +123,56 @@
101
123
  {
102
124
  "low": 0.1,
103
125
  "high": 0.2,
104
- "n": 19,
105
- "acc": 0.5789473684210527
126
+ "n": 21,
127
+ "acc": 0.23809523809523808
106
128
  },
107
129
  {
108
130
  "low": 0.2,
109
131
  "high": 0.3,
110
- "n": 316,
111
- "acc": 0.3322784810126582
132
+ "n": 666,
133
+ "acc": 0.2912912912912913
112
134
  },
113
135
  {
114
136
  "low": 0.3,
115
137
  "high": 0.4,
116
- "n": 1213,
117
- "acc": 0.37757625721352017
138
+ "n": 2416,
139
+ "acc": 0.3265728476821192
118
140
  },
119
141
  {
120
142
  "low": 0.4,
121
143
  "high": 0.5,
122
- "n": 2754,
123
- "acc": 0.34132171387073346
144
+ "n": 4308,
145
+ "acc": 0.3152274837511606
124
146
  },
125
147
  {
126
148
  "low": 0.5,
127
149
  "high": 0.6,
128
- "n": 3647,
129
- "acc": 0.35481217438990953
150
+ "n": 4777,
151
+ "acc": 0.3382876282185472
130
152
  },
131
153
  {
132
154
  "low": 0.6,
133
155
  "high": 0.7,
134
- "n": 3659,
135
- "acc": 0.3883574747198688
156
+ "n": 4943,
157
+ "acc": 0.3568682986040866
136
158
  },
137
159
  {
138
160
  "low": 0.7,
139
161
  "high": 0.8,
140
- "n": 4610,
141
- "acc": 0.40629067245119305
162
+ "n": 5534,
163
+ "acc": 0.382544271774485
142
164
  },
143
165
  {
144
166
  "low": 0.8,
145
167
  "high": 0.9,
146
- "n": 7716,
147
- "acc": 0.43960601347848627
168
+ "n": 8066,
169
+ "acc": 0.43627572526655095
148
170
  },
149
171
  {
150
172
  "low": 0.9,
151
173
  "high": 1.0,
152
- "n": 37314,
153
- "acc": 0.5987564989012167
174
+ "n": 30517,
175
+ "acc": 0.6440344725890488
154
176
  }
155
177
  ]
156
178
  },
@@ -159,7 +181,7 @@
159
181
  "particle-honorific kryptonite (e.g. FR 'Saint-Just-Saint-Rambert') if not in synth set",
160
182
  "non-Latin scripts (CJK, Cyrillic) fall through to byte-fallback tokens; F1 unknown"
161
183
  ],
162
- "notes": "Stage 2 v3.0.0 \u2014 same encoder geometry as v2.0.x (8.87M params, 6L/256H/4-heads) plus a linear-chain CRF decoder (+~500 params with a frozen BIO transition mask), label smoothing on the per-token CE leg (disabled in the shipped hparams after iteration; see ship notes), and a 21-label classifier head (was 15) that adds venue / street / house_number BIO classes. Trained on corpus-v0.3.0 which adds the US DOT NAD source (~97M structured 911-grade address points). The CRF transition mask makes orphan-I sequences (e.g. \"Saint Petersburg \u2192 Petersburg\" clipping visible on the v0.2.0 demo) structurally impossible. See evals/scores-by-version.json for the v2.0.x \u2192 v3.0.0 deltas + the per-component F1 on the new fine labels.",
184
+ "notes": "v0.4.0 \u2014 issue #116. Same encoder geometry as v0.3.0 (8.87M params, 6L/256H/4-heads, 21 BIO labels, linear-chain CRF). Issue proposed (1) per-token CRF NLL normalization + (3) class-weighted CE biased toward coarse labels + (4) source-weight rebalance. Empirical iteration found that \u00a71 and \u00a73 destabilize sustained training at every LR tested (5e-4, 3e-4, 1.5e-4 \u2014 even the v0.3.0-safe LR), and on the golden v0.1.2 eval (4535 entries) they slightly REGRESS country/postcode F1 vs v0.3.0. SHIPPED recipe is the \u00a74-only ablation (v0.3.0 dual-loss + v0.4.0 source-weight rebalance) at lr=1.5e-4, step 2200. Modest fine-label gains (street +0.03, house_number +0.01), modest coarse-F1 regression (country -0.07, postcode -0.07). \u00a71/\u00a73 deferred to v0.4.1 corpus-side investigation per issue's '2K divergence' clause. Full ablation matrix retrospective in LOG.md.",
163
185
  "format": {
164
186
  "model": "ONNX int8 dynamic",
165
187
  "tokenizer": "SentencePiece unigram, byte_fallback=true, vocab_size=16000",
@@ -171,5 +193,5 @@
171
193
  "tokenizer": "tokenizer.model",
172
194
  "model_card": "model-card.json"
173
195
  },
174
- "base_relpath": "/data/models/checkpoints/stage2/step-001800"
196
+ "base_relpath": "/data/models/checkpoints/v0_4_0-stableLR-source-only/step-002200"
175
197
  }
package/model.onnx CHANGED
Binary file
package/package.json CHANGED
@@ -1,20 +1,19 @@
1
1
  {
2
- "name": "@mailwoman/neural-weights-fr-fr",
3
- "version": "3.0.0",
4
- "description": "Mailwoman neural-classifier weights for locale 'fr-fr'. Data-only package — loaded by @mailwoman/neural at runtime.",
5
- "license": "AGPL-3.0-only",
6
- "private": false,
7
- "repository": {
8
- "type": "git",
9
- "url": "https://github.com/sister-software/mailwoman"
10
- },
11
- "files": [
12
- "model.onnx",
13
- "tokenizer.model",
14
- "model-card.json",
15
- "README.md"
16
- ],
17
- "publishConfig": {
18
- "access": "public"
19
- }
20
- }
2
+ "name": "@mailwoman/neural-weights-fr-fr",
3
+ "version": "4.1.0",
4
+ "description": "Mailwoman neural-classifier weights for locale 'fr-fr'. Data-only package — loaded by @mailwoman/neural at runtime.",
5
+ "license": "AGPL-3.0-only",
6
+ "repository": {
7
+ "type": "git",
8
+ "url": "https://github.com/sister-software/mailwoman"
9
+ },
10
+ "files": [
11
+ "model.onnx",
12
+ "tokenizer.model",
13
+ "model-card.json",
14
+ "README.md"
15
+ ],
16
+ "publishConfig": {
17
+ "access": "public"
18
+ }
19
+ }
package/tokenizer.model CHANGED
Binary file