From bcc6d6b20f03b9acad792b2ae94f6870e326db12 Mon Sep 17 00:00:00 2001 From: Alex Date: Thu, 16 Nov 2023 10:59:48 +0500 Subject: [PATCH] Fix README.md (#149) --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index 3945ab8..6096921 100644 --- a/README.md +++ b/README.md @@ -16,7 +16,7 @@ Loading LoRA checkpoints in [Blealtan's format](https://github.com/Blealtan/RWKV If you use `rwkv.cpp` for anything serious, please [test all available formats for perplexity and latency](rwkv%2Fmeasure_pexplexity.py) on a representative dataset, and decide which trade-off is best for you. -In general, `RWKV v5` models are 2 times slower than `RWKV v4` models, and require from 1.5 times (sequence length = 1) to 6 times (sequence length = 64) more memory. +In general, **`RWKV v5` models are as fast as `RWKV v4` models**, with minor differencies in latency and memory consumption, and with having way higher quality than `v4`. Therefore, it is recommended to use `RWKV v5`. Below table is for reference only. Measurements were made on 4C/8T x86 CPU with AVX2, 4 threads. The models are `RWKV v4 Pile 169M`, `RWKV v4 Pile 1.5B`.