15 times Faster than Llama 2: Introducing DeciLM - NAS-Generated LLM with Variable GQA
Explore DeciLM 6B, a high-efficiency large language model that outpaces Llama 2 7B by 15 times. The model was generated using Deci's proprietary Neural Architecture Search-powered technology, AutoNAC. Delve into this powerful model's architecture, efficiency and performance.
Source Community Community License Agreement Large Language Models Neural Architecture Search Grouped Query Attention Multi Head Attention
Source: deci.ai