metadata

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)

Use rwkv pip package 0.8.21+ for RWKV-5 inference: https://pypi.org/project/rwkv/

Online 1.5B Demo: https://huggingface.co/spaces/BlinkDL/RWKV-Gradio-1

Online 3B Demo: https://huggingface.co/spaces/BlinkDL/RWKV-Gradio-2

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 :)

Recommended fine-tuning format (use \n for newlines):

User: xxxxxxxxxxxxxxx

Assistant: xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx

User: xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx

Assistant: xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx

A good chat prompt (better replace \n\n in xxx to \n, such that there will be no newlines in xxx):

User: hi

Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it.

User: xxx

Assistant:

QA prompt (better replace \n\n in xxx to \n, such that there will be no newlines in xxx):

Question: xxx

Answer:

and

Instruction: xxx

Input: xxx

Response: