transformers model for NER

Cgrdj · October 18, 2023, 8:17am

Hello,

I' m currently training a NER model with bert as base mode.I already done the manual annotation part using the bert-vocab and transformers_tokenizers files.I then trained the model using these annotations and a spacy procedure with config file.
Now i would like to use ner.correct and ner.teach model using my trained model.
I could start the correct method that way : prodigy ner.correct ner_logs ./model/output/model-best ./data.jsonl --label INTRACTABLE_STR,OS,URL -F transformers_tokenizers.py
the problem is i cant provide the method with a correct tokenization,how could i do that?
thanks

ryanwesslen · October 18, 2023, 7:29pm

hi @Cgrdj!

Thanks for your question and welcome to the Prodigy community

The problem is the transformers_tokenizers.py recipe only has the manual recipe; it doesn't include a correct or teach recipe, so you'll need to write a custom recipe that combines that with how the built-in correct or teach recipes are.

To view examples of how the ner.correct and ner.teach recipes are structured, you can use the existing prodigy-recipes repo examples like ner.correct to learn how to do this.

github.com

explosion/prodigy-recipes/blob/master/ner/ner_correct.py

import copy
from typing import List, Optional
import spacy
from spacy.training import Example
import prodigy
from prodigy.components.loaders import JSONL
from prodigy.components.preprocess import add_tokens, split_sentences
from prodigy.util import split_string, set_hashes

def make_tasks(nlp, stream, labels):
    """Add a 'spans' key to each example, with predicted entities."""
    # Process the stream using spaCy's nlp.pipe, which yields doc objects.
    # If as_tuples=True is set, you can pass in (text, context) tuples.
    texts = ((eg["text"], eg) for eg in stream)
    for doc, eg in nlp.pipe(texts, as_tuples=True):
        task = copy.deepcopy(eg)
        spans = []
        for ent in doc.ents:
            # Ignore if the predicted entity is not in the selected labels.
            if labels and ent.label_ not in labels:

This file has been truncated. show original

For example, use the manual recipe but add a model in the loop with the make_tasks like in ner.correct but you'll need to modify the tokenization like in bert.ner.manual to modify for BertWordPieceTokenizer tokenizer.

Also - it's important to know that the prodigy-recipes are similar but not identical to the actual built-in recipes. If you want to view those, you can instead run prodigy stats, find the Location, then look for the recipes/ner.py script. There you can find the actual built-in ner.correct recipe.

Hope this helps!

Cgrdj · October 19, 2023, 3:09pm

Hello,

Yeah I finally figured out i had to create my own recipes. I' m currently working on it ,would be nice to provide support for transformers in prodigy directly.

Thank you for your explanation!

koaning · October 24, 2023, 11:39am

Hi there!

As luck would have it, I'm currently working on a Prodigy plugin that can train Huggingface models directly on top of your Prodigy annotations. You can see the preview here:

It may not cover every use-case out there, but if nothing else it may be a source of inspiration for some more custom recipes that you might write yourself.

Cgrdj · October 30, 2023, 9:19am

Very interesting work indeed!

Cgrdj · October 31, 2023, 6:15pm

your answer is really interesting, do you know where i can find the code for bert.ner.manual?

ryanwesslen · October 31, 2023, 9:41pm

github.com

explosion/prodigy-recipes/blob/master/other/transformers_tokenizers.py

"""This recipe requires Prodigy v1.10+."""
from typing import List, Optional, Union, Iterable, Dict, Any
from tokenizers import BertWordPieceTokenizer
from prodigy.components.loaders import get_stream
from prodigy.util import get_labels
import prodigy


@prodigy.recipe(
    "bert.ner.manual",
    # fmt: off
    dataset=("Dataset to save annotations to", "positional", None, str),
    source=("Data to annotate (file path or '-' to read from standard input)", "positional", None, str),
    loader=("Loader (guessed from file extension if not set)", "option", "lo", str),
    label=("Comma-separated label(s) to annotate or text file with one label per line", "option", "l", get_labels),
    tokenizer_vocab=("Tokenizer vocab file", "option", "tv", str),
    lowercase=("Set lowercase=True for tokenizer", "flag", "LC", bool),
    hide_special=("Hide SEP and CLS tokens visually", "flag", "HS", bool),
    hide_wp_prefix=("Hide wordpieces prefix like ##", "flag", "HW", bool)
    # fmt: on

This file has been truncated. show original

Topic		Replies	Views
Training BERT on prodigy transformers , relations	3	796	February 2, 2023
Can't find recipe or command 'bert.ner.manual' usage , ner , solved , transformers	4	587	September 23, 2022
config.cfg for bert.ner.manual usage , ner , transformers	5	812	September 30, 2022
BERT support for prodigy train ner usage , ner , spacy , solved	2	1022	June 30, 2021
How does model-in-the-loop in ner.correct work? usage , ner , solved	2	340	November 13, 2022

transformers model for NER

Related topics