nanogpt_test

Dataset - tiny shakespeare, character-level

Tiny shakespeare, of the good old char-rnn fame :) Treated on character-level.

• Tokenization performed on Character level
• Vocab size 65. Following are the unique tokens
  • !$&',-.3:;?ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz
• Number of total tokens - 1115394
  • trained on 1,003,854 tokens (90%)
  • validation is performed on 111,540 tokens (10%)

The Huggingface Spaces Gradio App

This model is used for the following Huggingface Spaces Gradio App.

The app is available here

The takes following as input

1. Seed Text (Prompt) - This provided as input text to the GPT model, based on which it generates further contents. If no data is provided, the only a space (" ") is provided as input
2. Max tokens to generate - This controls the numbers of character tokens it will generate. The default value is 100.
3. Temperature - This accepts value between 0 to 1. Higher value introduces more randomness in the next token generation. Default value is set to 0.7.
4. Select Top N in each step - This is optional field. If no value is provided (or <= 0), all available tokens are considered for next token prediction based on SoftMax probability. However, if a number is set then only that many top characters will be considered for next token prediction.