|
|
--- |
|
|
title: PicoChat |
|
|
emoji: 🤖 |
|
|
colorFrom: blue |
|
|
colorTo: indigo |
|
|
sdk: gradio |
|
|
sdk_version: 5.42.0 |
|
|
app_file: app.py |
|
|
pinned: false |
|
|
license: mit |
|
|
short_description: PicoChat frontend. |
|
|
--- |
|
|
|
|
|
# 🤖 PicoChat |
|
|
|
|
|
> A 335M parameter language model trained from scratch on a MacBook Air M2 in ~7 days. |
|
|
|
|
|
This Space is the interactive frontend for **PicoChat**. The model weights are hosted separately to optimize for bandwidth and storage. |
|
|
|
|
|
🔗 **Model Weights:** [huggingface.co/MGow/PicoChat](https://huggingface.co/MGow/PicoChat) |
|
|
|
|
|
## 📊 Model Stats |
|
|
|
|
|
| Feature | Details | |
|
|
| :--- | :--- | |
|
|
| **Architecture** | Depth-16 GPT-style Transformer | |
|
|
| **Parameters** | ~335M | |
|
|
| **Training Data** | 377M tokens (FineWeb-Edu + Synthetic) | |
|
|
| **Hardware** | MacBook Air M2 (16GB RAM) | |
|
|
| **Training Cost** | 5kWh of electricity | |
|
|
|
|
|
**PicoChat** is a "lab notebook" proof-of-concept for training capable small language models (SLMs) on consumer hardware. It can chat, solve basic math problems, and has a quirky personality. |
|
|
|
