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:
  - HuggingFaceFW/fineweb-edu
  - mlfoundations/dclm-baseline-1.0
  - cerebras/SlimPajama-627B
  - EleutherAI/pile
  - bigcode/starcoderdata
  - oscar-corpus/OSCAR-2301

RWKV-7 World

Use rwkv pip package 0.8.28+ for RWKV-7 inference: https://pypi.org/project/rwkv/

Evals and more information: https://rwkv.com/

For developers: https://github.com/BlinkDL/RWKV-LM

Chat demo: https://github.com/BlinkDL/ChatRWKV/blob/main/API_DEMO_CHAT.py

MMLU eval: https://github.com/BlinkDL/RWKV-LM/blob/main/RWKV-v7/rwkv_mmlu_eval.py

rwkv7-v3-2.9b 54.56% (rwkv6-v2.1-3.1b 32.38%)

rwkv7-v3-1.5b 44.84% (rwkv6-v2.1-1.6b 26.34%)

0.1B = L12 D768 // 0.4B = L24 D1024 // 1.5B = L24 D2048 // ~3B = L32 D2560 // ~7B = L32 D4096

Model Description

RWKV-7 trained on 100+ world languages (80% English, 10% multilang, 10% code).

World-v3 = 3.1T tokens

World-v2.9 = subsampled 2T tokens

World-v2.8 = subsampled 1T tokens

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 never be extra \n\n in response):

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 never be extra \n\n in response):

Question: xxx

Answer:

and

Instruction: xxx

Input: xxx

Response:

!!! There should not be any space after your final ":" or you will upset the tokenizer and see non-English reponse !!!