Class TrainSmallGPTWithBPE
java.lang.Object
io.github.kirstenali.deepj.examples.TrainSmallGPTWithBPE
Example: tiny GPT training on a small text file using a BPE tokenizer.
Differences from the byte-level example:
- A BPE model is trained from the corpus before GPT training begins.
- The tokenizer is saved to disk so it can be reloaded alongside the model weights.
- GPT vocab size is set from the BPE vocab rather than the fixed 256-byte vocab.
-
Constructor Summary
Constructors -
Method Summary
-
Constructor Details
-
TrainSmallGPTWithBPE
public TrainSmallGPTWithBPE()
-
-
Method Details
-
main
- Throws:
Exception
-