---
language:
- en
- zh
- fr
- es
- de
- pt
- ru
- it
- ja
- ko
- vi
- ar
tags:
- pytorch
- text-generation
- causal-lm
- rwkv
license: apache-2.0
datasets:
- cerebras/SlimPajama-627B
- EleutherAI/pile
- bigcode/starcoderdata
- oscar-corpus/OSCAR-2301
---

# RWKV-5 World (Training in Progress)

## I am now uploading latest training-in-progress checkpts to https://huggingface.co/BlinkDL/temp/tree/main (to avoid bloating git history)

## I am now uploading latest training-in-progress checkpts to https://huggingface.co/BlinkDL/temp/tree/main (to avoid bloating git history)

Use rwkv pip package 0.8.14+ for RWKV-5 inference: https://pypi.org/project/rwkv/

Online Demo: https://huggingface.co/spaces/BlinkDL/ChatRWKV-gradio

GUI: https://github.com/josStorer/RWKV-Runner (see Releases)

How it works: https://twitter.com/BlinkDL_AI/status/1685230712247795713

https://www.rwkv.com/

## Model Description

RWKV-5 trained on 100+ world languages (70% English, 15% multilang, 15% code).

World = Some_Pile + Some_SlimPajama + Some_StarCoder + Some_OSCAR + All_Wikipedia + All_ChatGPT_Data_I_can_find

RWKV-5 training: set --my_testing "r2r4" in latest RWKV-LM v4neo: https://github.com/BlinkDL/RWKV-LM

World v1 = 0.59T tokens

World v2 = 1.12T tokens

Imagine what happens when we use more data :)