Back to Home
AI on Medium··Industry Media

Running a 284B Parameter MoE LLM at 0.00 GB Active VRAM: The Power of Layer-Streaming

中文摘要

2840亿参数MoE大模型无需显存运行。

English Summary

284B MoE LLM runs with 0.00 GB active VRAM.

Original Excerpt

How I bypassed enterprise hardware clusters… Continue reading on Medium »