SimpleStories Model Family

The SimpleStories models are a tiny model family created for interpretability research, trained on the SimpleStories dataset.

Usage

pip install simple_stories_train

from transformers import AutoTokenizer
import torch

from simple_stories_train.models.llama import Llama
from simple_stories_train.models.model_configs import MODEL_CONFIGS

# Select the model size you want to use
model_size = "35M"  # Options: "35M", "30M", "11M", "5M", "1.25M"

# Load model configuration
model_config = MODEL_CONFIGS[model_size]

# Load appropriate model
model_path = f"SimpleStories/SimpleStories-{model_size}"
model = Llama.from_pretrained(model_path, model_config)
device = torch.device("cuda" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu")
model.to(device)
model.eval()

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_path)

# Define your prompt
prompt = "The curious cat looked at the"

inputs = tokenizer(prompt, return_tensors="pt")
input_ids = inputs.input_ids.to(device)

# Generate text
with torch.no_grad():
    output_ids = model.generate(
        idx=input_ids,
        max_new_tokens=50,
        temperature=0.0,
        top_k=40,
        eos_token_id=tokenizer.eos_token_id
    )

# Decode output
output_text = tokenizer.decode(output_ids[0], skip_special_tokens=True)
print(f"Generated text:\n{output_text}")

Model Variants

Model Name	n_params	n_layers	d_model	n_heads	n_ctx	d_vocab
SimpleStories-35M	35 million	12	512	8	512	4096
SimpleStories-30M	30 million	10	512	8	512	4096
SimpleStories-11M	11 million	6	384	6	512	4096
SimpleStories-5M	5 million	6	256	4	512	4096
SimpleStories-1.25M	1.25 million	4	128	4	512	4096

Performance Comparison

Model-evaluated generation quality metrics:

Tokenizer

We use a custom WordPiece tokenizer with a small vocabulary size of 4096. We conducted morphological analysis and coverage gain analysis on the dataset to build a small tokenizer without compromising on the quality of generation.

Dataset

The SimpleStories dataset is a collection of short stories generated by state-of-the-art language models. It features:

Story annotation with high-level concepts: theme, topic, style, etc.
Higher semantic and syntactic diversity through seeded story generation
Generated by 2024 models
Several NLP-metrics pre-computed to aid filtering
ASCII-only guarantee for the English dataset

Read the dataset paper on arXiv.

SimpleStories
/

SimpleStories-35M