What are the patch lengths and the initial values of the generated tokens during generation?

Hello, thank you very much for your work! If I input the prompt "The capital of France is" into the BLT model, how can I know the patch length for the next generated patch, and how can I initialize the corresponding token embeddings?

The issue names only the BLT model and the example prompt, with no file, test, or entry point to inspect. Start by locating the generation path and its documentation, then clarify where patch lengths and initial token embeddings are exposed; done means the documented procedure answers both questions for the given prompt.

Written by the indexing model from the issue text.