a small language model trained on tinystories.

ollama run navthings/tale

curl http://localhost:11434/api/chat \
  -d '{
    "model": "navthings/tale",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='navthings/tale',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'navthings/tale',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Models

View all →

Name

1 model

Size / Usage

Context

Input

tale:latest

97MB · 384 context window · Text · 1 month ago

tale:latest

97MB

384

Text

Readme

tale

github

a small language model trained on tinystories.

tale is a ~50m parameter llama-style model built from scratch, trained on apple silicon and exported to gguf for local inference.

model

architecture: llamaforcausallm
parameters: ~50m
tokenizer: gpt-2 bpe
context: 384 tokens
training data: tinystories
training device: apple silicon (mps)
format: gguf
inference: llama.cpp / ollama

run

ollama run navthings/tale

or run the gguf directly with llama.cpp:

./llama-cli -m tale-step2000-f16.gguf -p "once upon a time "

why

tale is an experiment, the evolution to lilstory

i built it because my little brother kept on asking me for a bedtime story

it’s intentionally small.

status

early release.

the current model is trained for 2,000 steps and can generate simple tinystories-style text, but it is still very much a work in progress.

licence

mit