rafmacalaba commited on
Commit
3797360
·
verified ·
1 Parent(s): 3497e8a

datause-ner: single-type DATA_MENTION NER views (ner + ner_w_neg) at probe operating point

Browse files
.gitattributes CHANGED
@@ -58,3 +58,9 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
61
+ gliner2_holdout.jsonl filter=lfs diff=lfs merge=lfs -text
62
+ gliner2_train.jsonl filter=lfs diff=lfs merge=lfs -text
63
+ gliner2_val.jsonl filter=lfs diff=lfs merge=lfs -text
64
+ gliner_holdout.jsonl filter=lfs diff=lfs merge=lfs -text
65
+ gliner_train.jsonl filter=lfs diff=lfs merge=lfs -text
66
+ gliner_val.jsonl filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ task_categories:
3
+ - token-classification
4
+ tags:
5
+ - ner
6
+ - span-extraction
7
+ - data-mentions
8
+ license: apache-2.0
9
+ configs:
10
+ - config_name: gliner
11
+ data_files:
12
+ - split: train
13
+ path: gliner_train.jsonl
14
+ - split: val
15
+ path: gliner_val.jsonl
16
+ - split: holdout
17
+ path: gliner_holdout.jsonl
18
+ - config_name: gliner2
19
+ data_files:
20
+ - split: train
21
+ path: gliner2_train.jsonl
22
+ - split: val
23
+ path: gliner2_val.jsonl
24
+ - split: holdout
25
+ path: gliner2_holdout.jsonl
26
+ ---
27
+
28
+ # Datause NER (single-type probe operating point)
29
+
30
+ NER views of the `rafmacalaba/gliner-datause-v3-probe` operating point over `rafmacalaba/datause-probe-v3` passages (29,346 spans grouped into passage examples). Single entity type **DATA_MENTION**: a span is DATA_MENTION iff its probe head score is at or above the per-origin optimal threshold, else NON_MENTION. No NAMED/DESCRIPTIVE/VAGUE subtypes.
31
+
32
+ ## Per-origin thresholds (head best-F1, published holdout sweep)
33
+
34
+ | origin | threshold |
35
+ |---|---|
36
+ | `fcv_pads_east_africa` | 0.5 |
37
+ | `general_prwp` | 0.3 |
38
+ | `jad_paddy_docs` | 0.3 |
39
+ | `jdc_operational` | 0.5 |
40
+ | `refugee_pads` | 0.6 |
41
+ | `reliefweb` | 0.3 |
42
+
43
+ ## Columns
44
+
45
+ - `gliner`: `tokenized_text` + `ner` (DATA_MENTION only; NON_MENTION spans left untagged as hard negatives) + `ner_w_neg` (NON_MENTION spans tagged too).
46
+ - `gliner2`: `input` + `output` (data_mention only) + `output_w_neg` (data_mention + non_mention).
47
+ - Traceability on every row: `corpus_id`, `page`, `chunk`, `split`, `spans` (per-span text, pred, luna_label, head_score, threshold, char offsets, key, source). Luna keep/drop verdicts are preserved in `spans[].luna_label` for audit, but tagging follows the head-threshold operating point.
48
+
49
+ ## Splits
50
+
51
+ Doc-disjoint `train` / `val` / `holdout`, inherited from `rafmacalaba/datause-probe-v3` (`outputs/probe_v3_data`), so NER splits match probe splits. See `split_stats.json`.
52
+
53
+ ## Usage
54
+
55
+ ```python
56
+ from datasets import load_dataset
57
+ gliner = load_dataset("rafmacalaba/datause-ner", "gliner")
58
+ gliner2 = load_dataset("rafmacalaba/datause-ner", "gliner2")
59
+ # DATA_MENTION-only supervision:
60
+ # gliner['train']['ner'] / gliner2['train']['output']
61
+ # with NON_MENTION tagged:
62
+ # gliner['train']['ner_w_neg'] / gliner2['train']['output_w_neg']
63
+ ```
64
+
65
+ ## Files
66
+
67
+ - `gliner_{train,val,holdout}.jsonl`, `gliner2_{train,val,holdout}.jsonl`
68
+ - `labels.json`, `thresholds.json`, `split_stats.json`
gliner2_holdout.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fb1716b59ea33f3379d5d2e158a85e9e70c6f06b09daac06dc42fcec353237a6
3
+ size 11566860
gliner2_train.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ed87e203d8b2c89ea95291739c47d9a90b0ccea57a6854f466cfc34fe3dbe28d
3
+ size 51507467
gliner2_val.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6e264d7ce86c94614e1c2467b2c77b8d0b5be90b64ce046f43ecacfb4ab60e90
3
+ size 11234586
gliner_holdout.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3d9552e57e16e0665cfb06678b618a2948504ec046c8d74f6acb42d3e25849a7
3
+ size 12359715
gliner_train.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e62a3b99faf280b5323ef71b1c78055191e340910eaa7d76c321cccd28bd5f7b
3
+ size 54910631
gliner_val.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:69fd25382396c8cbfb26f6ab64e18686676368479f3d7114d6d9b0cf6a5ec0c8
3
+ size 12028251
labels.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "labels": [
3
+ "DATA_MENTION",
4
+ "NON_MENTION"
5
+ ]
6
+ }
split_stats.json ADDED
@@ -0,0 +1,58 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "train": {
3
+ "examples": 18751,
4
+ "positive_DATA_MENTION": 11910,
5
+ "negative_only": 6841,
6
+ "DATA_MENTION_spans": 12770,
7
+ "NON_MENTION_spans": 7512,
8
+ "origin_counts": {
9
+ "fcv_pads_east_africa": 5301,
10
+ "jad_paddy_docs": 79,
11
+ "jdc_operational": 647,
12
+ "general_prwp": 7960,
13
+ "refugee_pads": 1606,
14
+ "reliefweb": 3158
15
+ },
16
+ "corpus_counts": {
17
+ "fcv": 10791,
18
+ "prwp": 7960
19
+ }
20
+ },
21
+ "val": {
22
+ "examples": 4107,
23
+ "positive_DATA_MENTION": 2596,
24
+ "negative_only": 1511,
25
+ "DATA_MENTION_spans": 2754,
26
+ "NON_MENTION_spans": 1653,
27
+ "origin_counts": {
28
+ "fcv_pads_east_africa": 1134,
29
+ "jdc_operational": 89,
30
+ "general_prwp": 1756,
31
+ "refugee_pads": 404,
32
+ "reliefweb": 724
33
+ },
34
+ "corpus_counts": {
35
+ "fcv": 2351,
36
+ "prwp": 1756
37
+ }
38
+ },
39
+ "holdout": {
40
+ "examples": 4203,
41
+ "positive_DATA_MENTION": 2679,
42
+ "negative_only": 1524,
43
+ "DATA_MENTION_spans": 2963,
44
+ "NON_MENTION_spans": 1694,
45
+ "origin_counts": {
46
+ "fcv_pads_east_africa": 1075,
47
+ "jad_paddy_docs": 77,
48
+ "jdc_operational": 198,
49
+ "general_prwp": 1784,
50
+ "refugee_pads": 375,
51
+ "reliefweb": 694
52
+ },
53
+ "corpus_counts": {
54
+ "fcv": 2419,
55
+ "prwp": 1784
56
+ }
57
+ }
58
+ }
thresholds.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "fcv_pads_east_africa": 0.5,
3
+ "general_prwp": 0.3,
4
+ "jad_paddy_docs": 0.3,
5
+ "jdc_operational": 0.5,
6
+ "refugee_pads": 0.6,
7
+ "reliefweb": 0.3
8
+ }