From a5f9751a2e4482c4a4df37ce1b8b8a9a0cc09282 Mon Sep 17 00:00:00 2001 From: randaller Date: Sun, 19 Mar 2023 19:13:39 +0300 Subject: [PATCH] Update README.md --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index 51bfee6..2c61aae 100644 --- a/README.md +++ b/README.md @@ -25,7 +25,7 @@ One may run with 32 Gb of RAM, but inference will be slow (with the speed of you I am running PyArrow version on a [12700k/128 Gb RAM/NVIDIA 3070ti 8Gb/fast huge nvme with 256 Gb swap for 65B model] and getting one token from 30B model in a few seconds. -For example, **30B model uses around 70 Gb of RAM**. 7B model fits into 18 Gb. 13B model uses 48 Gb. +For example, **PyArrow 30B model uses around 70 Gb of RAM**. 7B model fits into 18 Gb. 13B model uses 48 Gb. If you do not have nvidia videocard, you may use another repo for cpu-only inference: https://github.com/randaller/llama-cpu or [HF 🤗 version](https://github.com/randaller/llama-chat#hugging-face--version-inference--training).