YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
GPT-2 in PyTorch
GPT-style language model built from scratch in PyTorch, trained on a subset of WikiText-103 corpus. Model trained on an NVIDIA A100 GPU — full run completes in ~6 minutes.
Install
python -m venv venv && venv\Scripts\activate
pip install -r requirements.txt
Train
python main.py --gpu --tokenizer tiktoken
Inference
python inference.py --prompt "Once upon a time" --gpu
Pretrained weights load automatically from triton329/gpt2 on HuggingFace.
Example output
Once upon a time . At the early 20th century , though it was seen by the time
in 1864 , it lacks a 1804 Antigua – 5 @-@ class submarine of the hurricane...
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support