llama-modchip runs 284B DeepSeek model on 64GB RAM without GPU

Shows large MoE models can run on commodity CPU RAM, lowering local inference barriers.

An open project called llama-modchip runs a 284B-parameter DeepSeek model on a 64GB CPU-only system, showing off-device inference for large models without a GPU.

Name

DeepSeek 284B

Params

284B

Domain

general

Read it at Hacker News → 💬 37 on HN

Coverage: 2 sources, in order

1. 08 Nov 10:22 Hacker News Read the article →
2. 08 Nov 12:49 news.lavx.hu Read the article →

← Front page