Trajectory as the Teacher: Few-Step Discrete Flow Matching via Energy-Navigated Distillation

Apple ML Research··

Discrete flow matching generates text by iteratively transforming noise tokens into coherent language, but may require hundreds of forward passes. Distillation uses the multi-step trajectory to train a student to reproduce the process in a few steps. When the student underperforms, the usual explanation is insufficient capacity. We argue the opposite: the trajectory is the bottleneck, not the student. Each training trajectory is built through a chain of blind stochastic jumps with no evaluation ...

Read full article →

Related Articles

OpenAI’s o1 correctly diagnosed 67% of ER patients vs. 50-55% by triage doctors
donsupreme · Hacker News · 4mo ago
Accelerating Gemma 4: faster inference with multi-token prediction drafters
amrrs · Hacker News · 4mo ago
A couple million lines of Haskell: Production engineering at Mercury
unignorant · Hacker News · 4mo ago
Using “underdrawings” for accurate text and numbers
samcollins · Hacker News · 4mo ago
ProgramBench: Can language models rebuild programs from scratch?
jonbaer · Hacker News · 4mo ago