7 ms·
Could these quantized models make MTP (Multi-Token Prediction) significantly faster when used as drafters for larger regular Gemma 4 models?
by somewhatrandom9 3mo ago
Could these quantized models make MTP (Multi-Token Prediction) significantly faster when used as drafters for larger regular Gemma 4 models?
- dist-epoch 3mo agoGoogle already released specialized drafters for Gemma 4.
- Havoc 3mo agoThe E2B ones? Or what do you mean by specialized drafters?
- girvo 3mo agoThe “-assistant” models released by Google are specialised tiny MTP draft models :) 31b-it-assistant is what enables MTP
- int_19h 3mo agoThey have -assistant in the name, so e.g.: https://huggingface.co/google/gemma-4-31B-it-assistant https://huggingface.co/google/gemma-4-31B-it-assistant
- Havoc 3mo agoThanks