SHAKTI: A 2.5 Billion Parameter Small Language Model Optimized for Edge AI and Low-Resource Environments

15 Oct 2024  ·  Syed Abdul Gaffar Shakhadri, Kruthika KR, Rakshit Aralimatti ·

We introduce Shakti, a 2.5 billion parameter language model specifically optimized for resource-constrained environments such as edge devices, including smartphones, wearables, and IoT systems. Shakti combines high-performance NLP with optimized efficiency and precision, making it ideal for real-time AI applications where computational resources and memory are limited. With support for vernacular languages and domain-specific tasks, Shakti excels in industries such as healthcare, finance, and customer service. Benchmark evaluations demonstrate that Shakti performs competitively against larger models while maintaining low latency and on-device efficiency, positioning it as a leading solution for edge AI.

PDF Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Question Answering BBH Shakti-LLM (2.5B) Accuracy 58.2 # 1
Question Answering BoolQ Shakti-LLM (2.5B) Accuracy 61.1 # 54
Question Answering HellaSwag Shakti-LLM (2.5B) Accuracy 52.4 # 1
Question Answering MedQA Shakti-LLM (2.5B) Accuracy 60.3 # 12
Question Answering MMLU qwen-LLM 7B Accuracy 71.8 # 1
Question Answering PIQA Shakti-LLM (2.5B) Accuracy 86.2 # 8
Question Answering TriviaQA Shakti-LLM (2.5B) EM 58.2 # 39
Question Answering TruthfulQA Shakti-LLM (2.5B) Accuracy 68.4 # 1

Methods