embedding-space-multitoken-prediction

Accelerate LLM decoding by predicting multiple future tokens simultaneously using mask-token probing in embedding space, without retraining or auxiliary models.

Independent GitHub snapshot · Automated classification · Original documentation governs