AINewsnow

Model soups failed until we fixed one line: our 26M grammar tagger from 11% to 1.9% false fixes

Our keyboard's grammar model is a 26.2M-parameter tagger we trained from scratch: one label per word (KEEP / DELETE / REPLACE_x / APPEND_x), no text generation, Core ML on the phone. Over two weeks it went from F0.5 0.693 with 11.1% false positives to 0.803 with 1.9% , and from fixing 22% to 77% of…

Read the full story at DEV Community — Machine Learning ↗

Timeline · 1 report

  1. 2026-10-28 16:52 · DEV Community — Machine Learning
    Model soups failed until we fixed one line: our 26M grammar tagger from 11% to 1.9% false fixes

More stories

  1. Gemini 4 Carbon: Our next frontier model — Google DeepMind Blog
  2. NVIDIA Nemotron 3.5 Transcribe and Granary v2: Intelligent Multilingual Transcription You Can Prompt and Adapt — Hugging Face Blog
  3. PyTorch 2.15 Release Blog — PyTorch Blog
  4. Voice AI in vogue: Nvidia releases new open transcription model — TechCrunch AI
  5. Startups, Did You Know That Agentforce Coworker Can Do This? — Salesforce AI Blog
  6. Building an Agentic AI Platform on Amazon Bedrock AgentCore — AWS Machine Learning Blog
  7. As AI agents reshape the web, Google opens Gemini and Search to more apps — TechCrunch AI
  8. Distributed training with Ray Train DLCs and EFA on EKS — AWS Machine Learning Blog

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