PEFT boosts LLM hate speech detection in Roman Urdu to over 93% F1

A new research paper explores the effectiveness of Parameter-Efficient Fine-Tuning (PEFT) methods, specifically Low-Rank Adaptation (LoRA), for hate speech detection in Roman Urdu. The study found that while zero-shot inference on models like Mistral, LLaMA, Falcon, and BERT yielded moderate results (F1 score of 0.56), fine-tuning with PEFT significantly improved performance to over 0.93 F1 score. This approach demonstrates strong computational efficiency, making it a viable solution for processing low-resource languages. AI

IMPACT Demonstrates a highly efficient method for adapting LLMs to low-resource languages, potentially improving AI's utility in diverse linguistic contexts.

RANK_REASON Academic paper detailing a new method for LLM adaptation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Bert
• Falcon
• Hate Speech Detection
• large-language models
• llama
• Low Rank Adaptation
• Mistral AI
• Parameter-Efficient Fine-Tuning
• PURUTT
• Roman Urdu
• South Asians
• Urdu

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →