This repository provides a tiny 16M parameters language model for debugging and testing purposes. This is created by tuning sbintuitions/tiny-lm with oasset1 datasets in Japanese and English.

How to use

python

from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
 
model = AutoModelForCausalLM.from_pretrained("sbintuitions/tiny-lm-chat", torch_dtype="auto")
tokenizer = AutoTokenizer.from_pretrained("sbintuitions/tiny-lm-chat", use_fast=False)
generator = pipeline("text-generation", model=model, tokenizer=tokenizer)

prompt = tokenizer.apply_chat_template([{"role": "user", "content": "Hello!"}], add_generation_prompt=True, tokenize=False)
print(generator(prompt, max_length=30, do_sample=True, top_k=100))

Model architecture

A 4-layer, 512-hidden-size transformer-based language model.

Training

The model was first pre-trained on English Wikipedia and Japanese Wikipedia to optimize a traditional language modelling objective for 25B tokens. And then it was fine-tuned on oasst1 datasets in Japanese and English for 15 epochs.

License

MIT License