bert-base-ner¶

Use Case and High-Level Description¶

bert-base-ner is a fine-tuned BERT model that is ready to use for Named Entity Recognition and achieves state-of-the-art performance for the NER task. It has been trained to recognize four types of entities: location (LOC), organizations (ORG), person (PER) and Miscellaneous (MISC).

Specifically, this model is a bert-base-cased model that was fine-tuned on the English version of the standard CoNLL-2003 Named Entity Recognition dataset. For details about the original model, check out BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding, HuggingFace’s Transformers: State-of-the-art Natural Language Processing papers and repository.

Tokenization occurs using the BERT tokenizer (see the demo code for implementation details) and the enclosed vocab.txt dictionary file.

Metric

Value

GOps

22.3874

MParams

107.4319

Source framework

PyTorch*

Accuracy¶

The quality metric was calculated on CONLL-2003 Named Entity Recognition dataset (dev set). Input sequences were padded to 128 symbols.

Metric

Value

F1

94.45%

Input¶

Original model¶

1. Token IDs, name: input_ids, shape: 1, 128. Token IDs is sequence of integer values that is representing the tokenized input sentence. The sequence structure is as follows ([CLS] and [SEP] should be replaced by corresponding token IDs as specified by the dictionary): [CLS] + tokenized text + [SEP] + 0 (for padding to sequence length 128]

2. Input mask, name: attention_mask, shape: 1, 128. Input mask is a sequence of integer values representing the mask of valid values in the input. The values of this input are equal to:

• 1 at positions corresponding to the [CLS] + tokenized text + [SEP] part of the input_ids (i.e. all positions except those containing the padding), and

• 0 at all other positions

3. Token types, name: token_type_ids, shape: 1, 128. Token types is sequence of integer values representing the segmentation of the input_ids. The values are equal to 0 at all other positions (all text belongs to single segment)

• [CLS] is a special symbol added in front of the text.

• [SEP] is a special separator added at the end of the text.

Converted model¶

Converted model has the same inputs like in original.

Output¶

Original model¶

Token classifier, name: output, shape: 1, 128, 9 floating point-valued logit scores vectors that represents probability of belonging each token to 9 classes:

Abbreviation

Description

O

Outside of a named entity

B-MIS

Beginning of a miscellaneous entity right after another miscellaneous entity

I-MIS

Miscellaneous entity

B-PER

Beginning of a person’s name right after another person’s name

I-PER

Person’s name

B-ORG

Beginning of an organisation right after another organisation

I-ORG

Organisation

B-LOC

Beginning of a location right after another location

I-LOC

Location

Converted model¶

Converted model has the same output like original.

You can download models and if necessary convert them into Inference Engine format using the Model Downloader and other automation tools as shown in the examples below.

python3 <omz_dir>/tools/downloader/downloader.py --name <model_name>

An example of using the Model Converter:

python3 <omz_dir>/tools/downloader/converter.py --name <model_name>

Use Case and High-Level Description¶

bert-base-ner is a fine-tuned BERT model that is ready to use for Named Entity Recognition and achieves state-of-the-art performance for the NER task. It has been trained to recognize four types of entities: location (LOC), organizations (ORG), person (PER) and Miscellaneous (MISC).

Specifically, this model is a bert-base-cased model that was fine-tuned on the English version of the standard CoNLL-2003 Named Entity Recognition dataset. For details about the original model, check out BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding, HuggingFace’s Transformers: State-of-the-art Natural Language Processing papers and repository.

Tokenization occurs using the BERT tokenizer (see the demo code for implementation details) and the enclosed vocab.txt dictionary file.

Metric

Value

GOps

22.3874

MParams

107.4319

Source framework

PyTorch*

Accuracy¶

The quality metric was calculated on CONLL-2003 Named Entity Recognition dataset (dev set). Input sequences were padded to 128 symbols.

Metric

Value

F1

94.45%

Input¶

Original model¶

1. Token IDs, name: input_ids, shape: 1, 128. Token IDs is sequence of integer values that is representing the tokenized input sentence. The sequence structure is as follows ([CLS] and [SEP] should be replaced by corresponding token IDs as specified by the dictionary): [CLS] + tokenized text + [SEP] + 0 (for padding to sequence length 128]

2. Input mask, name: attention_mask, shape: 1, 128. Input mask is a sequence of integer values representing the mask of valid values in the input. The values of this input are equal to:

• 1 at positions corresponding to the [CLS] + tokenized text + [SEP] part of the input_ids (i.e. all positions except those containing the padding), and

• 0 at all other positions

3. Token types, name: token_type_ids, shape: 1, 128. Token types is sequence of integer values representing the segmentation of the input_ids. The values are equal to 0 at all other positions (all text belongs to single segment)

• [CLS] is a special symbol added in front of the text.

• [SEP] is a special separator added at the end of the text.

Converted model¶

Converted model has the same inputs like in original.

Output¶

Original model¶

Token classifier, name: output, shape: 1, 128, 9 floating point-valued logit scores vectors that represents probability of belonging each token to 9 classes:

Abbreviation

Description

O

Outside of a named entity

B-MIS

Beginning of a miscellaneous entity right after another miscellaneous entity

I-MIS

Miscellaneous entity

B-PER

Beginning of a person’s name right after another person’s name

I-PER

Person’s name

B-ORG

Beginning of an organisation right after another organisation

I-ORG

Organisation

B-LOC

Beginning of a location right after another location

I-LOC

Location

Converted model¶

Converted model has the same output like original.

You can download models and if necessary convert them into Inference Engine format using the Model Downloader and other automation tools as shown in the examples below.

python3 <omz_dir>/tools/downloader/downloader.py --name <model_name>
python3 <omz_dir>/tools/downloader/converter.py --name <model_name>