You can deploy this model and start earning money today!

ShakhzoDavronov/
xlm-roberta-lora-ner-uz

Views

No views yet

Named Entity Recognition model for Uzbek

Model Description

The XLM-RoBERTa model was fine tuned with LORA (Low-Rank Adaption) to reduce time consumption and efficent use of resource (GPU/CPU).

Base model: xlm-roberta-large (over 570 million parameters)
LORA fine-tuned model: xlm-roberta-lora-ner-uz (around 12 million paramters)

Limitations

The model was fine-tuned dataset collected from news resources. So, it may cause domain adaption (perform well news related data rather than other domain)

Named Entity Categories

B-LOC (Location names)
B-ORG (Organization names)
B-PERSON (Person names)
B-DATE (Date expressions)
B-EVENT (Event names)
B-GPE (Countries, cities, states)

Testing Model

With code provided below user can test model performance:

Example text: "Toshkentda Shavkat Mirziyoyev Xalqaro dzyudo federatsiyasi rahbarini "Do'stlik" ordeni bilan mukofotladi."

python

from peft import PeftModel, PeftConfig
from transformers import AutoModelForTokenClassification, AutoTokenizer

label_names=['O','B-PERSON','B-GPE','B-ORG','B-LOC','B-DATE','B-EVENT']
num_labels=len(label_names)
id_to_label = {i: label for i, label in enumerate(label_names)}
label_to_id = {label:i for i, label in enumerate(label_names)}

peft_model_id = "ShakhzoDavronov/xlm-roberta-lora-ner-uz"
config = PeftConfig.from_pretrained(peft_model_id)
model = AutoModelForTokenClassification.from_pretrained(config.base_model_name_or_path, num_labels=len(label_names),
                                                        id2label=id_to_label, label2id=label_to_id)

tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path)
model = PeftModel.from_pretrained(model, peft_model_id)

from transformers import pipeline
ner_pipeline=pipeline('ner', model=model, tokenizer=tokenizer)

text="""Toshkentda Shavkat Mirziyoyev Xalqaro dzyudo federatsiyasi rahbarini "Do'stlik" ordeni bilan mukofotladi."""
ner=ner_pipeline(text)
for entity in ner:
    print(entity)

Results:

python

{'entity': 'B-LOC', 'score': 0.70824957, 'index': 1, 'word': '▁Toshkent', 'start': 0, 'end': 8}
{'entity': 'B-LOC', 'score': 0.6332058, 'index': 2, 'word': 'da', 'start': 8, 'end': 10}
{'entity': 'B-PERSON', 'score': 0.94302815, 'index': 3, 'word': '▁Shavkat', 'start': 11, 'end': 18}
{'entity': 'B-PERSON', 'score': 0.9404429, 'index': 4, 'word': '▁Mirziyoyev', 'start': 19, 'end': 29}
{'entity': 'B-ORG', 'score': 0.8909233, 'index': 5, 'word': '▁X', 'start': 30, 'end': 31}
{'entity': 'B-ORG', 'score': 0.8900482, 'index': 6, 'word': 'alqaro', 'start': 31, 'end': 37}
{'entity': 'B-ORG', 'score': 0.8870184, 'index': 7, 'word': '▁', 'start': 38, 'end': 39}
{'entity': 'B-ORG', 'score': 0.84329146, 'index': 8, 'word': 'dzy', 'start': 38, 'end': 41}
{'entity': 'B-ORG', 'score': 0.8615051, 'index': 9, 'word': 'udo', 'start': 41, 'end': 44}
{'entity': 'B-ORG', 'score': 0.8913255, 'index': 10, 'word': '▁', 'start': 45, 'end': 46}
{'entity': 'B-ORG', 'score': 0.8640603, 'index': 11, 'word': 'feder', 'start': 45, 'end': 50}
{'entity': 'B-ORG', 'score': 0.8678666, 'index': 12, 'word': 'atsiyasi', 'start': 50, 'end': 58}

Training Metrics

Epoch	Training Loss	Validation Loss	Precision	Recall	F1 Score	Accuracy
1	0.214000	0.204440	0.745021	0.718837	0.731695	0.931259
2	0.197800	0.193916	0.783191	0.697650	0.737950	0.934701
3	0.182500	0.188210	0.765977	0.731226	0.748198	0.935774