Is it possible to release the trained speculators for Gemma4 and Qwen3? This would make efforts to replicate the paper easier, and would also be useful for people who are trying to inference the models locally. Since a lot of inference frameworks already support dflash and eagle, those models in particular could be gotten to work with minimum effort.
Thank you.
Is it possible to release the trained speculators for Gemma4 and Qwen3? This would make efforts to replicate the paper easier, and would also be useful for people who are trying to inference the models locally. Since a lot of inference frameworks already support dflash and eagle, those models in particular could be gotten to work with minimum effort.
Thank you.