Skip to content
AImpact
IT EN
Training Advanced Also known as: MoD · Mixed Denoising Objectives

Mixture of Denoisers

A pretraining strategy (UL2, Google 2022) that trains a single model on multiple denoising objectives simultaneously: left-to-right language modeling, span prediction (BERT-style masked spans of varying lengths and corruptions), and prefix language modeling. Unifies the strengths of GPT-style and T5-style pretraining. The model learns when to use each mode based on a sentinel token that signals the objective type.

ShareLinkedInX

In practice

A researcher wanting a flexible model for both completion and question answering can use UL2 or a Flan-UL2 checkpoint without choosing between encoder-decoder (T5) and decoder-only (GPT) architectures. In practice, the sentinel token `[S2S]`, `[NLU]`, or `[NLG]` must be prepended to the prompt to activate the correct mode — a detail that significantly impacts performance and is often omitted, causing poor results.

Related terms

Seen in the wild

112 entries mentioning it
  1. Meta releases Llama 4.1: Scout, Maverick, and Behemoth MoE models under Apache 2.0
    Landmark
  2. Google I/O 2026: Gemini Ultra 3, Project Astra goes live on Pixel, 2M context with real-time grounding, Veo 3.2, Imagen 4
    High
  3. Anthropic releases Claude Fable 5: a new model family beyond the 4.x line
    Landmark
  4. Realtime voice AI: sub-second latency and multilingual become the norm
    Medium
  5. Boston Dynamics Atlas Electric: manipulation foundation model trained on 10 million robot hours
    High
  6. Mistral releases Devstral Small: 7B coding model for agentic tasks on consumer GPU
    Medium
  7. MCP at 18 months: the server ecosystem hits critical mass
    High
  8. Robotics foundation model: a new step toward the "GPT of manipulation"
    High
  9. Mistral Small 4: three models (reasoning + vision + coding) fused into one open weight
    High
  10. Groq launches GroqCloud 2.0: LPU Gen3, 2000 tokens/sec, and Frankfurt European data center
    Medium
  11. Nano Banana 2: Google rebuilds its viral image model around consistency and text
    Medium
  12. DeepSeek releases Janus Pro: one model to understand and generate images
    High
  13. Gemini 3 Pro and Flash: Google relaunches the frontier challenge
    High
  14. Alibaba releases Qwen2.5-VL 72B: best open-source multimodal model beats GPT-4o on key benchmarks
    Landmark
  15. OpenAI releases o3-mini: advanced reasoning at a fraction of the cost
    High
  16. MCP ecosystem 2025: Inspector, UI, registry, and cross-vendor adoption
    High
  17. Claude Haiku 4.5: the small model that matches May's Sonnet 4
    Medium
  18. Runway Gen-4: AI video with consistent characters across multiple scenes
    High
  19. Cline: the open-source VS Code coding agent that splits Plan and Act
    Medium
  20. Apollo Research: frontier models 'scheme' in evals — paper published
    High
← All terms