README.md

metadata

title: Tiny Transformer Trainer
emoji: 🧠
colorFrom: purple
colorTo: blue
sdk: docker
app_port: 7860
pinned: false

Tiny Transformer Trainer 🧠

Train small GPT-style transformers from your own text data — directly in the browser.

How to Use

1. Upload data:

   • .txt — plain text files
   • .csv — Q&A pairs with named columns
   • .json — Q&A pairs as a list of objects

2. Set column names (for CSV/JSON): specify which columns hold questions/prompts and answers/responses.

3. Configure model: pick hidden size, layers, heads, vocab size, and max sequence length.

4. Train: click 🚀 Start Training and watch live logs.

5. Download: grab the .zip with model weights + tokenizer.

Output Files

File	Description
model/	Trained GPT2LMHeadModel weights + config
tokenizer/	Trained PreTrainedTokenizerFast vocab + config
README.md	Usage code snippet

Local Usage

from transformers import GPT2LMHeadModel, PreTrainedTokenizerFast

model = GPT2LMHeadModel.from_pretrained("trained_model_output/model")
tokenizer = PreTrainedTokenizerFast.from_pretrained("trained_model_output/tokenizer")