precomputed pile binidx dataset

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

12.05k stars 827 forks source link

precomputed pile binidx dataset #151

Open RichardErkhov opened 1 year ago

RichardErkhov commented 1 year ago

pile binidx dataset with 20b_tokenizer: https://huggingface.co/datasets/RichardErkhov/RWKV-LM_pile_binidx_dataset

wonkyoc commented 1 month ago

Which tokenizer did you use to generate the dataset?

RichardErkhov commented 1 month ago

like done in readme from that time

using https://github.com/Abel2076/json2binidx_tool

python tools/preprocess_data.py --input ./sample.jsonl --output-prefix ./data/sample --vocab ./20B_tokenizer.json --dataset-impl mmap --tokenizer-type HFTokenizer --append-eod