AINewsnow

Prefix tuning changes no weights: 128 numbers steering a frozen model

LoRA computes W_eff = W + BA . It changes a weight matrix by a low-rank delta, and at serve time you can add the delta in and forget it was ever separate. Prefix tuning never forms a delta. It leaves every weight bit-for-bit where it was and prepends P learned key rows and P learned value rows to t…

Read the full story at DEV Community — Machine Learning ↗

Timeline · 1 report

  1. 2026-11-09 16:57 · DEV Community — Machine Learning
    Prefix tuning changes no weights: 128 numbers steering a frozen model

More stories

  1. Nvidia opens board probe into CEO Jensen Huang's affair with suspected Chinese spy — The Verge AI
  2. Meta Expands AI Age Estimation Globally to Extend Instagram Teen Account Protections — Meta Newsroom
  3. Meta launches Muse Ember, a small multimodal model aimed at cutting AI agent costs — TechCrunch AI
  4. Google adds TPU 8tx and TPU 8i Max to eighth-generation AI chip lineup — Google: The Keyword
  5. Google Adds Maps Platform Development to AI Studio — Google AI Blog
  6. US AI safety center CAISSI asked to report to White House, raising independence concerns — AP News
  7. MIT engineers unveil FLEX, an AI-assisted learning platform that helps robots reason beyond literal commands — MIT CSAIL News
  8. MIT CSAIL unveils WorldParticle, a single AI model simulating many physical substances — MIT CSAIL News

Get the daily brief of stories like this at 6:30 every morning →