Researchers have developed Riazi-8B, a new large language model specifically designed for mathematical reasoning in the Urdu language. This model addresses the limitations of existing English-centric LLMs, which perform poorly on low-resource languages like Urdu. Riazi-8B was created through a two-step process involving pre-training on Urdu Wikipedia and fine-tuning with Urdu Chain-of-Thought data derived from GSM8K. Evaluations on the MGSM-Urdu benchmark show that Riazi-8B significantly improves answer correctness, reasoning quality, and Urdu generation compared to other Urdu instruction-tuned models. AI
IMPACT Extends mathematical reasoning capabilities of LLMs to low-resource languages, potentially benefiting Urdu-speaking users and informing future multilingual AI development.
RANK_REASON The cluster describes a new research paper detailing the creation and evaluation of a specialized LLM for a low-resource language.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
