3 ms·
The tokenizer in llama.cpp probably needs fixing then or it has some other bug.
by bugglebeetle 2y ago
The tokenizer in llama.cpp probably needs fixing then or it has some other bug.
- 0x7cfe 2y agoDefinitely. I tried gemma2:27B model with phrases like "translate the following sentence to language X" and it even failed to understand the task and spat out completely irrelevant things, like math formulas. OTOH, smaller model did it perfectly.