Tokenizer for fine tuning RWKV-v5 world model

BlinkDL / RWKV-LM

RWKV is an RNN with transformer-level LLM performance. It can be directly trained like a GPT (parallelizable). So it's combining the best of RNN and transformer - great performance, fast inference, saves VRAM, fast training, "infinite" ctx_len, and free sentence embedding.

Apache License 2.0

11.99k stars 825 forks source link

Tokenizer for fine tuning RWKV-v5 world model #230

Open mathewchris96 opened 4 months ago

mathewchris96 commented 4 months ago

Can someone help with what tokenizer to use here:- tokenizer = PreTrainedTokenizerFast(tokenizer_file=f'{repo_dir}/20B_tokenizer.json') for RWKV v5 world mode. I'm trying to finetune the model here.

BlinkDL commented 3 months ago

see https://huggingface.co/RWKV/v5-Eagle-7B-HF for a HF-compatible tokenizer