AutoResearch-tinystories-depth8

AutoResearch-tinystories-depth8 is a 172.0M parameter decoder-only Transformer trained from scratch on TinyStories (karpathy/tinystories-gpt4-clean).

This model is part of the AutoResearch project, which focuses on training, evaluating, and releasing efficient language models with reproducible research workflows.

Overview

This is a 8-layer decoder-only Transformer trained on the TinyStories (karpathy/tinystories-gpt4-clean) dataset for 0.2 hours of wall-clock training time. The model achieves a validation bits-per-byte (val_bpb) of 0.896881 (perplexity: 1.8620) on the held-out validation set.

References

Papers

NanoGPT / NanoChat architecture patterns

Datasets

Training: TinyStories (karpathy/tinystories-gpt4-clean)
Tokenizer: climbmix-400b-shuffle

Related Projects

karpathy/autoresearch

WANDB Run

https://wandb.ai/dustinsdelivery-delivery-ai/autoresearch/runs/auyqhcl5

Highlights

Trained from scratch
172.0M parameters
Trained on 22.0M tokens (42 steps)
8-layer decoder-only Transformer with sliding window attention
RoPE positional encoding, RMSNorm, ReLU² activation
MuonAdamW optimizer (Muon for matrices, AdamW for embeddings)
Mixture of Experts (8 routed + 1 shared, top-2 routing)
Hugging Face Transformers compatible

Model Architecture

Property	Value
Architecture	Decoder-only Transformer
Parameters	171,999,760 (172.0M)
Layers	8
Hidden Size	512
Attention Heads	4
KV Heads	4
Head Dimension	128
Feed Forward Size	1024 (MoE: 8 experts, 1 shared, top-2)
Context Length	2048
Vocabulary Size	16,384
Positional Encoding	RoPE
Activation	ReLU²
Normalization	RMSNorm
Window Pattern	SSSL
Weight Tying	No

Training

This model was trained from scratch for 0.2 hours (620s) of wall-clock training time.

Training Configuration

Setting	Value
Optimizer	MuonAdamW (Muon + AdamW)
Precision	torch.bfloat16
Learning Rate	0.04 (matrix) / 0.6 (embedding)
Weight Decay	0.2
Batch Size	4 × 2048 = 8,192 tokens/step
Gradient Accumulation	64 steps
Total Batch Size	524,288 tokens
Context Length	2048
Vocabulary	16,384 tokens (BPE)
LR Scheduler	Linear warmdown (50%)
Activation Checkpointing	Enabled

Hardware

GPU: NVIDIA GeForce RTX 4060 Ti
VRAM: 16.0 GB
Peak VRAM Used: 4.0 GB
MFU: 25.80%
Framework: PyTorch 2.9.1+cu128

Dataset

Name: TinyStories (karpathy/tinystories-gpt4-clean)
Language: English

Preprocessing

Data is packed into fixed-length sequences of 2048 tokens using the nanochat-compatible BPE tokenizer (16,384 vocabulary, 9 special tokens). No additional filtering or deduplication is applied beyond what is in the source dataset.

Intended Use

This model is intended for:

Educational purposes and research
Text generation experiments
Studying small language model training dynamics

Not recommended for:

Production use or safety-critical applications
Tasks requiring factual accuracy

Evaluation

Results

Metric	Score
Validation BPB	0.896881
Perplexity	1.8620
Peak VRAM	4.0 GB
MFU	25.80%

Example Generations

Example 1

Prompt

Once upon a time,

Generation

Once upon a time, there was a little girl named Lucy. She had a new friend named Sue, and her long. Sue was happy to see more than anything, but it was too hot. Sue was sad because she could not make her friends feel

Example 2

Prompt

A lonely dragon

Generation

A lonely dragon bird saw the bird and bird, trying to help. The bird became the bird was happy and said, "I want to play there. I can't help you."
The bird and the bird learned to be friendship with the bird. They learned

Example 3

Prompt

The opposite of boy is

Generation

The opposite of boy is doing?"
Lily and Ben nodded. They hugged each other and wished they were busy with each other and smiled. They had their mom and had a valuable lesson that day to their mom and mom and mom said they would forgive

Example 4

Prompt

The opposite of queen is

Generation

The opposite of queen is kind and sharing animals. They are friends and a big animals. They are happy.
The people at the line and and, and animals. They are very happy. They moms and say sorry. They say sorry for their mum and their parents

Example 5

Prompt

My name is

Generation

My name is M T Lucy. She is very sad and slow. She wants her hands and makes Anna happy. She wants to show her mom the dog, and mom. She says, "What is that?" Anna says, "I am sorry

Usage

import torch
import pickle
import json
from train import GPT, GPTConfig, Tokenizer

# Load config
with open('config.json', 'r') as f:
    config_dict = json.load(f)
config = GPTConfig(**{k: v for k, v in config_dict.items() if k in GPTConfig.__dataclass_fields__})

# Load model
model = GPT(config)
state_dict = torch.load('model.pt', map_location='cpu')['state_dict']
model.load_state_dict(state_dict)
model.eval()

# Load tokenizer
with open('tokenizer.pkl', 'rb') as f:
    tokenizer = pickle.load(f)

# Generate
prompt = 'Once upon a time, '
input_ids = tokenizer.encode(prompt)
x = torch.tensor([input_ids], dtype=torch.long)
with torch.no_grad():
    for _ in range(50):
        logits = model(x)
        probs = torch.softmax(logits[:, -1, :] / 0.8, dim=-1)
        next_token = torch.multinomial(probs, num_samples=1)
        input_ids.append(next_token.item())
        x = torch.tensor([input_ids], dtype=torch.long)
print(tokenizer.decode(input_ids))

Repository Structure

model.pt                  # Model weights
config.json               # Model architecture config
dataset.txt               # Dataset name used for training
token_bytes.pt            # Token byte mappings
tokenizer.pkl             # Trained BPE tokenizer
tokenizer_config.json     # Tokenizer configuration
training_metrics.json     # Training metrics
README.md                 # This file

Limitations

Small model size limits language understanding and coherence
Trained on a single dataset (TinyStories) — limited domain
Fixed time budget training — not fully trained to convergence
No RLHF or safety alignment

Ethical Considerations

This is a research artifact, not a production model
The training data consists of synthetic stories (GPT-4 generated)
No harmful content filtering was applied
Intended for research and educational use only

Citation

@misc{autoresearch_tinystories_depth8,
  title={AutoResearch-tinystories-depth8},
  author={Dustin Loring},
  year={2026},
  howpublished={\url{https://huggingface.co/quik-models/rose-dust-100}}
}}

Version History

Version	Date	Notes
v1.0	2026-07-30	Initial release

Acknowledgements

Built with the AutoResearch training framework.

Thanks to:

Hugging Face
PyTorch
The creators of the TinyStories dataset
The open-source AI research community

License

This model is released under the MIT License unless otherwise specified.

Downloads last month: 20