English => News => Topic started by: Redaktion on February 04, 2025, 00:50:08
Title: Way to run DeepSeek's 671B AI model without expensive GPUs discovered
Post by: Redaktion on February 04, 2025, 00:50:08
Hugging Face engineer Matthew Carrigan recently revealed on X a method to locally run DeepSeek's advanced R1 model with 8-bit quantization, eliminating the need for expensive GPUs, for a reported cost of $6,000. The key? Having a lot of memory as opposed to vast computing power reserves.