byte-deep-hybrid

details

bytebox is a model designed to change public perspective on byte-level language models. It's a deep model pretrained on roughly 1B tokens that demonstrates instruction-following skills with only 60m parameters. Training and inference codes are included.

To the author's knowledge, this is the first-ever decoder-only instruction pretrained byte-level transformer openly available to date.

python infer_bytefast60m.py --architecture-script bytefalcon_fast60m.py \
  --model /bytefast-60m/ \
  --tokenizer /byte-tokenizer/ \
  --instruction "Fix the sentence" \
  --text "this sentence need fixing you know i love it" \
  --max-new-tokens 128 \
  --temperature 0

Prompt

What is Python?

Output

Python is a high-level, general-purpose programming language. Its design philosophy emphasizes code readability, and its syntax allows programmers to express concepts in fewer lines of code than would be possible in languages such as C++ or Java. The language provides confidence towards the use of programming languages and their debugging problems."

Downloads last month
1
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Dataset used to train appvoid/bytebox

Collection including appvoid/bytebox