Applications of Large Language Models

I am using Hugging Face libraries to perform model fine-tuning, evaluation and inference for various downstream tasks like Question-Answering, Named Entity Recognition and Text Classification.

Medical QA

It contains all the necessary data and notebooks to fine-tune Question Answering task using Large Language Model (LLM) on the medical transcription dataset.

CyberSecurityNER

It contains all the necessary data and notebooks to fine-tune Named Entity Recognition task using Large Language Model (LLM) on the cyber security dataset.

NepNER

In-depth analysis on Named Entity Recognition in Nepali Language using various Large Language Models. In this project, LLM is trained to extract abusive terms (Profanity, Violence and General) from the given texts written in Nepal language. The Nepali dataset is taken from the comment section of Nepali News Youtube channels.

Dataset Statistics

Dataset	Count
Train	2323
Test	280
Validation	330

Sample sentence

Entity	Label
भालुनी	B-PROFANITY
सावित्री	O
कुकुरनी	B-PROFANITY
मिले	O
को	O
रहेछ	O
आजा	O
प्रक्षया	O
थाहा	O
भयो	O
निर्माल	O
बहिनी	O
लाई	O
बलत्कार	B-VIOLENCE
गर्न	O
लगाउने	O
यनि	O
भलु	B-PROFANITY
हरु	O
रहेछ	O
।	O

Evaluation

Sakonii/distilbert-base-nepali

Metric	Score
test_precision	24.08%
test_recall	29.61%
test_f1	26.56%
test_accuracy	84.48%

xlm-roberta-large

Metric	Score
test_precision	38.26%
test_recall	44.50%
test_f1	41.15%
test_accuracy	86.15%

Inference

Input sentence = त्यो साला खाते जलेर मरे हुन्छ ।

Word	Predictions
साला खाते	GENERAL
जलेर मरे	VIOLENCE

Deployment

NepSA

It contains all the necessary data and notebooks to fine-tune Abusive Sentiment Analysis Task in Nepali Langauge using various Large Language Models (LLM).