agnews

[NAACL 2021] This is the code for our paper `Fine-Tuning Pre-trained Language Model with Weak Supervision: A Contrastive-Regularized Self-Training Approach'.

[ACL'23 Findings] This is the code repo for our ACL'23 Findings paper "ReGen: Zero-Shot Text Classification via Training Data Generation with Progressive Dense Retrieval".

[AAAI 2023] This is the code for our paper `Neighborhood-Regularized Self-Training for Learning with Few Labels'.

A project demonstrating the use of Large Language Models (LLMs) for text classification using the RoBERTa model.

Recurrent Capsule Network for Text Classification

Fine-tuning doesn't require a server cluster! This project uses QLoRA to optimize BERT for news classification. By combining 4-bit quantization and LoRA, I slashed VRAM usage by 75% and trained only 2% of parameters. The result? A pro-level classifier in a tiny 20MB adapter file—high performance, minimal hardware.

Comparative analysis of classical NLP, sentence embeddings and word embeddings across AG News, DBpedia and IMDb 50K

AG News Text Classification: Classical NLP vs sentence and word embeddings with datasets of size 50, 200, 500, 2000, and 10000

News Topic Classifier: Fine-tuned BERT model for lightning-fast news categorization

FedSLIP is a federated learning framework for zero-shot personalized parameter efficient fine-tuning using dual-track LoRA, local sparse identity masks, FedProx stabilization, and low communication cost.