lethalantidote commited on
Commit
8bdb0cf
·
verified ·
1 Parent(s): 4114d30

Publish quality-gated PHI token classifier

Browse files
Files changed (6) hide show
  1. README.md +33 -0
  2. config.json +46 -0
  3. model.safetensors +3 -0
  4. run_manifest.json +51 -0
  5. tokenizer.json +0 -0
  6. tokenizer_config.json +15 -0
README.md ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language: en
3
+ library_name: transformers
4
+ pipeline_tag: token-classification
5
+ base_model: emilyalsentzer/Bio_ClinicalBERT
6
+ tags:
7
+ - clinical
8
+ - phi
9
+ - ner
10
+ ---
11
+
12
+ # Synthetic PHI Detector
13
+
14
+ This token-classification model was fine-tuned from `emilyalsentzer/Bio_ClinicalBERT` using
15
+ Synthetic Synthea-derived clinical templates. It recognizes three entity types
16
+ with the BIO labels stored in `config.json`.
17
+
18
+ Training examples: 5997
19
+
20
+ | Entity | Strict seqeval F1 |
21
+ |---|---:|
22
+ | MEDICAL_RECORD | 1.0000 |
23
+ | DIAGNOSIS | 1.0000 |
24
+ | MEDICATION | 1.0000 |
25
+
26
+ Overall strict seqeval F1: 1.0000
27
+
28
+ ## Limitations
29
+
30
+ Synthetic templates do not represent every institution, document style, or
31
+ identifier format. False negatives and false positives are expected. This
32
+ model assists PHI detection and is not a compliance guarantee or substitute
33
+ for human review and layered detection controls.
config.json ADDED
@@ -0,0 +1,46 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_cross_attention": false,
3
+ "architectures": [
4
+ "BertForTokenClassification"
5
+ ],
6
+ "attention_probs_dropout_prob": 0.1,
7
+ "bos_token_id": null,
8
+ "classifier_dropout": null,
9
+ "dtype": "float32",
10
+ "eos_token_id": null,
11
+ "hidden_act": "gelu",
12
+ "hidden_dropout_prob": 0.1,
13
+ "hidden_size": 768,
14
+ "id2label": {
15
+ "0": "O",
16
+ "1": "B-MEDICAL_RECORD",
17
+ "2": "I-MEDICAL_RECORD",
18
+ "3": "B-DIAGNOSIS",
19
+ "4": "I-DIAGNOSIS",
20
+ "5": "B-MEDICATION",
21
+ "6": "I-MEDICATION"
22
+ },
23
+ "initializer_range": 0.02,
24
+ "intermediate_size": 3072,
25
+ "is_decoder": false,
26
+ "label2id": {
27
+ "B-DIAGNOSIS": 3,
28
+ "B-MEDICAL_RECORD": 1,
29
+ "B-MEDICATION": 5,
30
+ "I-DIAGNOSIS": 4,
31
+ "I-MEDICAL_RECORD": 2,
32
+ "I-MEDICATION": 6,
33
+ "O": 0
34
+ },
35
+ "layer_norm_eps": 1e-12,
36
+ "max_position_embeddings": 512,
37
+ "model_type": "bert",
38
+ "num_attention_heads": 12,
39
+ "num_hidden_layers": 12,
40
+ "pad_token_id": 0,
41
+ "tie_word_embeddings": true,
42
+ "transformers_version": "5.14.1",
43
+ "type_vocab_size": 2,
44
+ "use_cache": true,
45
+ "vocab_size": 28996
46
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7fc3730f96aca8d89a8483487be43546bad6b5f9f51364186d8de85c32fd4b74
3
+ size 430923588
run_manifest.json ADDED
@@ -0,0 +1,51 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "base_model": "emilyalsentzer/Bio_ClinicalBERT",
3
+ "base_model_revision": "d5892b39a4adaed74b92212a44081509db72f87b",
4
+ "bio_labels": [
5
+ "O",
6
+ "B-MEDICAL_RECORD",
7
+ "I-MEDICAL_RECORD",
8
+ "B-DIAGNOSIS",
9
+ "I-DIAGNOSIS",
10
+ "B-MEDICATION",
11
+ "I-MEDICATION"
12
+ ],
13
+ "code_fingerprint": "6ca8a25e9d2aa93d901e2f15d984e1d8b497b390b48195775a5f6d85fcc3e54c",
14
+ "dataset_manifest_sha256": "77443e9234a02f1b3b9eff317662f3a93c96bdff306e86eb1d5ecbdd508bd57d",
15
+ "dependency_versions": {
16
+ "python": "3.12.13",
17
+ "torch": "2.11.0+cu128",
18
+ "transformers": "5.14.1"
19
+ },
20
+ "metrics": {
21
+ "DIAGNOSIS_f1": 1.0,
22
+ "DIAGNOSIS_precision": 1.0,
23
+ "DIAGNOSIS_recall": 1.0,
24
+ "DIAGNOSIS_support": 940.0,
25
+ "MEDICAL_RECORD_f1": 1.0,
26
+ "MEDICAL_RECORD_precision": 1.0,
27
+ "MEDICAL_RECORD_recall": 1.0,
28
+ "MEDICAL_RECORD_support": 300.0,
29
+ "MEDICATION_f1": 1.0,
30
+ "MEDICATION_precision": 1.0,
31
+ "MEDICATION_recall": 1.0,
32
+ "MEDICATION_support": 702.0,
33
+ "loss": 0.00027296075131744146,
34
+ "overall_f1": 1.0,
35
+ "overall_precision": 1.0,
36
+ "overall_recall": 1.0,
37
+ "overall_support": 1942.0,
38
+ "runtime": 1.704,
39
+ "samples_per_second": 1139.665,
40
+ "steps_per_second": 35.798
41
+ },
42
+ "seed": 42,
43
+ "training_parameters": {
44
+ "epochs": 4.0,
45
+ "eval_batch_size": 32,
46
+ "learning_rate": 2e-05,
47
+ "max_length": 256,
48
+ "train_batch_size": 16,
49
+ "weight_decay": 0.01
50
+ }
51
+ }
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,15 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "backend": "tokenizers",
3
+ "cls_token": "[CLS]",
4
+ "do_lower_case": true,
5
+ "is_local": false,
6
+ "local_files_only": false,
7
+ "mask_token": "[MASK]",
8
+ "model_max_length": 1000000000000000019884624838656,
9
+ "pad_token": "[PAD]",
10
+ "sep_token": "[SEP]",
11
+ "strip_accents": null,
12
+ "tokenize_chinese_chars": true,
13
+ "tokenizer_class": "BertTokenizer",
14
+ "unk_token": "[UNK]"
15
+ }