proomt

Search

Search posts, papers, and topics

All posts

Apple Machine Learning Research1 min readpaperadvanced

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

Read on apple.com

Summary

Apple researchers propose Trajectory‑Shaped Discrete Flow Matching (TS‑DFM), a training‑time distillation method that replaces blind stochastic jumps in discrete flow‑matching with an energy‑based compass to select higher‑quality intermediate tokens. On a 170 M‑parameter language model, the 8‑step student outperforms the 1 024‑step teacher by 32 % perplexity while being 128× faster, beating basel…

  • In discrete flow‑matching, the teacher’s multi‑step trajectory can be the bottleneck because early poor decisions propagate forward.
  • TS‑DFM introduces a lightweight energy evaluator at each intermediate step during training, guiding the trajectory toward more coherent continuations without affecting inference cost.
  • Empirical results on a 170 M‑parameter LM show the shaped student (8 steps) achieves 32 % lower perplexity than the 1 024‑step teacher and runs 128× faster.
  • Performance gains hold across different source data distributions and three evaluator scales, surpassing baselines trained on significantly more data or larger models.

The work reframes the limiting factor in discrete flow‑matching from model capacity to trajectory quality, offering a practical way to compress many‑step generative processes into a few steps without sacrificing output quality. This could make discrete diffusion‑style language models viable for lat…

8/10
llmmachine learningdistillationgenerative modelsdiscrete flow matching

Related reading

  1. When EOS Tokens Disagree: Understanding Length Inflation in On-Policy Distillation

    On-policy distillation (OPD) can lead to excessively long student responses, a phenomenon called length inflation. This paper identifies "termination-token mismatch" between base students and post-trained teachers as a key source, where models place stopping probability on different EOS tokens. Treating functionally equivalent EOS tokens as a shared semantic stopping action substantially mitigate…

    Hugging Face Daily Papersarxiv.org1 minpaper

  2. Training Specialist Models without Reasoning Trajectories for Domain Expert Distillation

    The paper shows that when specialist LLMs are trained only on QA pairs (no explicit reasoning supervision), their optimization implicitly selects a latent distribution of reasoning trajectories. By treating the distilled student as an agnostic probe—since it inherits only the sampled trajectories—the authors empirically demonstrate a strong correlation (across 27 specialist‑student pairs) between…

    Hugging Face Daily Papersarxiv.org1 minpaper

  4. Dynamically Scaled Activation Steering

    Dynamically Scaled Activation Steering (DSAS) is a method‑agnostic framework that learns per‑token, per‑layer scaling factors to turn existing activation‑steering interventions on only when a model is likely to produce undesired output (e.g., toxic text). The scaling can be optimized jointly with any steering function, improves the toxicity‑utility trade‑off on language models, transfers to text‑…

    Apple Machine Learning Researchapple.com1 minpaper

  5. RetireOPD: Self-Retiring On-Policy Distillation for Agentic Reinforcement Learning

    RetireOPD introduces a self‑retiring on‑policy distillation framework for multi‑turn RL agents. A skill‑conditioned teacher is first trained with environment rewards, then a skill‑free student learns jointly via RL and token‑level distillation. The student automatically drops the teacher once its performance gap stops shrinking and it reaches a target success‑rate fraction, after which training c…

    Hugging Face Daily Papersarxiv.org1 minpaper

  6. Zing-0.5: Toward Playable Worlds with Real-Time Joint Action and Text Control

    Zing‑0.5 is a 5 B autoregressive world model that lets users control generated environments in real time using both keyboard actions and text prompts. The paper introduces unified action‑text conditioning, segment‑level teacher distillation, and a low‑cost streaming inference pipeline that runs at 24 FPS (832×480) for about $0.009 per minute, achieving 81 % overall and 88.5 % consistency on a nav…

    Hugging Face Daily Papersarxiv.org1 minpaper