We tried to use the huggingface transformers library to recreate the TinyStories models on Consumer GPU. Output model has 9 million parameters.

Tweaked code of springtangent (https://github.com/springtangent/tinystoriestrainer/blob/main/tinystories_train.py)

Code credit - springtangent

------ EXAMPLE USAGE ---

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("segestic/Tinystories-0.1-9m")

model = AutoModelForCausalLM.from_pretrained("segestic/Tinystories-0.1-9m")

prompt = "Once upon a time there was"

input_ids = tokenizer.encode(prompt, return_tensors="pt")

output = model.generate(input_ids, max_length = 1000, num_beams=1)

output_text = tokenizer.decode(output[0], skip_special_tokens=True)

print(output_text)