Hugging Face's logo

BiomedicalEvidencePlatform
/
pico-extractor

Token Classification
Transformers
Safetensors
MLX
English
bert
fp16
biomedical
pico
ner
Model card Files
xet
 Community
kubeheoriginal commited on
Commit
a5a5a1c
·
verified ·
1 Parent(s): 5776900

v1.0.0: convert zuzab/pico-extractor@ed4df86

Browse files
Files changed (5) hide show
  1. README.md +73 -0
  2. config.json +53 -0
  3. model.safetensors +3 -0
  4. tokenizer.json +0 -0
  5. tokenizer_config.json +21 -0
README.md ADDED
@@ -0,0 +1,73 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ base_model: zuzab/pico-extractor
4
+ library_name: transformers
5
+ pipeline_tag: token-classification
6
+ language:
7
+ - en
8
+ tags:
9
+ - mlx
10
+ - safetensors
11
+ - fp16
12
+ - biomedical
13
+ - pico
14
+ - ner
15
+ - bert
16
+ ---
17
+
18
+ # pico-extractor
19
+
20
+ BIO token classifier tagging PICO elements (Population, Intervention, Comparator, Outcome) in clinical-trial abstracts; used by Mineris for PICO span extraction.
21
+
22
+ This is a **format conversion** of [`zuzab/pico-extractor`](https://huggingface.co/zuzab/pico-extractor)
23
+ for Mineris' MLX inference stack (`mineris_inference`) — not a fine-tune. Weights are
24
+ the full upstream `BertForTokenClassification` (encoder + classification head) cast to fp16 and stored as `model.safetensors`; a
25
+ `tokenizer.json` is included so the model loads without `transformers` or `torch`.
26
+
27
+ ## Labels
28
+
29
+ | id | label |
30
+ |---|---|
31
+ | 0 | `O` |
32
+ | 1 | `B-P` |
33
+ | 2 | `B-I` |
34
+ | 3 | `B-C` |
35
+ | 4 | `B-O` |
36
+ | 5 | `I-P` |
37
+ | 6 | `I-I` |
38
+ | 7 | `I-C` |
39
+ | 8 | `I-O` |
40
+
41
+ ## Provenance
42
+
43
+ | | |
44
+ |---|---|
45
+ | Upstream repo | [`zuzab/pico-extractor`](https://huggingface.co/zuzab/pico-extractor) |
46
+ | Upstream commit | `ed4df86e482d9c53d953791a71c15caa5a687d96` |
47
+ | License | `apache-2.0` (declared by Mineris, the upstream owner) |
48
+ | Conversion notebook | `notebook/notebooks/converter/convert_pico_extractor.ipynb` (mineris repo) |
49
+ | Mineris commit | `9817fe7bbeb2a4b23fd683144a28a5843af94af9` |
50
+ | Version | `v1.0.0` |
51
+ | dtype | fp16 |
52
+ | Tensors / size | 199 / 215.3 MB |
53
+
54
+ ## Conversion parity (fp16 mirror vs fp32 upstream, PyTorch)
55
+
56
+ | Metric | Value |
57
+ |---|---|
58
+ | Texts | 18 |
59
+ | Tokenizer id mismatches (`tokenizer.json` via `tokenizers` vs upstream tokenizer) | 0 |
60
+ | Token argmax label agreement, per text — min / mean | 1.000000 / 1.000000 |
61
+ | Logits, non-padding tokens — max abs diff / median relative diff | 9.82e-03 / 5.09e-04 |
62
+ | Threshold (min token argmax label agreement) | 0.995 |
63
+
64
+ MLX-runtime parity is tracked separately in the mineris `notebook/notebooks/benchmark/` notebooks.
65
+
66
+ ## Versioning
67
+
68
+ Releases are semver git tags on this repo (`vMAJOR.MINOR.PATCH`); consumers pin a tag. MAJOR =
69
+ output-incompatible change, MINOR = compatible re-conversion, PATCH = card/metadata only.
70
+
71
+ ## CHANGELOG
72
+
73
+ - `v1.0.0` — initial conversion from `zuzab/pico-extractor@ed4df86`.
config.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "gradient_checkpointing": false,
12
+ "hidden_act": "gelu",
13
+ "hidden_dropout_prob": 0.1,
14
+ "hidden_size": 768,
15
+ "id2label": {
16
+ "0": "O",
17
+ "1": "B-P",
18
+ "2": "B-I",
19
+ "3": "B-C",
20
+ "4": "B-O",
21
+ "5": "I-P",
22
+ "6": "I-I",
23
+ "7": "I-C",
24
+ "8": "I-O"
25
+ },
26
+ "initializer_range": 0.02,
27
+ "intermediate_size": 3072,
28
+ "is_decoder": false,
29
+ "label2id": {
30
+ "B-C": 3,
31
+ "B-I": 2,
32
+ "B-O": 4,
33
+ "B-P": 1,
34
+ "I-C": 7,
35
+ "I-I": 6,
36
+ "I-O": 8,
37
+ "I-P": 5,
38
+ "O": 0
39
+ },
40
+ "layer_norm_eps": 1e-12,
41
+ "max_position_embeddings": 512,
42
+ "model_type": "bert",
43
+ "num_attention_heads": 12,
44
+ "num_hidden_layers": 12,
45
+ "pad_token_id": 0,
46
+ "position_embedding_type": "absolute",
47
+ "tie_word_embeddings": true,
48
+ "transformers_version": "5.5.0",
49
+ "type_vocab_size": 2,
50
+ "use_cache": false,
51
+ "vocab_size": 28895,
52
+ "torch_dtype": "float16"
53
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7ec14a22bad28f1e13c6f6348b33a13fd9601b808f78aa6dcd6b8e001ce2deff
3
+ size 215321290
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "backend": "tokenizers",
3
+ "cls_token": "[CLS]",
4
+ "do_basic_tokenize": true,
5
+ "do_lower_case": true,
6
+ "is_local": true,
7
+ "local_files_only": false,
8
+ "mask_token": "[MASK]",
9
+ "max_length": 512,
10
+ "model_max_length": 1000000000000000019884624838656,
11
+ "never_split": null,
12
+ "pad_token": "[PAD]",
13
+ "sep_token": "[SEP]",
14
+ "stride": 0,
15
+ "strip_accents": null,
16
+ "tokenize_chinese_chars": true,
17
+ "tokenizer_class": "BertTokenizer",
18
+ "truncation_side": "right",
19
+ "truncation_strategy": "longest_first",
20
+ "unk_token": "[UNK]"
21
+ }