single mention · inference & serving
MTP-to-EAGLE-3 draft-model fine-tuning
Converts a pretrained MTP layer into an EAGLE-3-style draft model while freezing the target model and updating the draft layer and feature-fusion projection.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See decoding strategy for the methods filed beside it.
Also called Draft Model Fine-Tuning.
Filed under inference & serving › decoding strategy