llama.cpp updates test-save-load-state for token input and empty prompts.

Core summary

One sentence to understand this update

llama.cpp release b9501 refactors `test-save-load-state` to accept token input and generates random tokens for models without tokenizers when no prompt is provided.

Impact & opportunity

What this could mean

This update improves testing flexibility for `llama.cpp` users, particularly for models lacking tokenizers or requiring specific token inputs.