Guide-linked quiz · Medium Level

Nous Research Quiz

Explore Nous Research's open assistants, preference fine-tuning, DeMo communication methods, and Psyche distributed-training project.

Related guide pathsNous Research
Question 1 of 8

What is Nous Research best known for in its early rise to prominence?

Preview every question8 questions

Answers and explanations appear as you take the quiz.

  1. What is Nous Research best known for in its early rise to prominence?
    • Fine-tuning open base models into the Hermes assistant series
    • Manufacturing AI chips
    • Running a cloud gaming service
    • Building closed-source frontier models from scratch
  2. Which base models has Nous most commonly fine-tuned?
    • Google's PaLM and Gemini
    • Meta's Llama and Mistral models
    • OpenAI's GPT-4
    • IBM Watson
  3. What problem does Nous Research’s DisTrO project address?
    • Generating realistic images
    • Compressing video files
    • Reducing the communication bandwidth needed for distributed training
    • Detecting spam emails
  4. What is the Psyche network?
    • A social media platform
    • A benchmark for measuring reasoning
    • A type of attention mechanism
    • An effort to train models across geographically distributed internet-connected hardware
  5. What technique does Nous typically use to align its fine-tuned models?
    • Supervised fine-tuning plus preference optimization like DPO on curated data
    • Training entirely from random weights with no base model
    • Only hand-writing rules with no learning
    • Reinforcement learning from raw pixels
  6. What does the name DeMo stand for in Nous Research's distributed-training work?
    • Distributed Model Orchestration
    • Decoupled Momentum Optimization
    • Decoder-Mixture Operation
    • Deterministic Memory Offload
  7. According to the DeMo paper, which training state need not be fully synchronized between accelerators?
    • The text prompts used for evaluation
    • The vocabulary definitions in the tokenizer
    • The complete optimizer states and model parameters
    • The labels used to evaluate image classifiers
  8. Which three roles does Psyche's training system assign?
    • A tokenizer, a rendering engine, and a search index
    • An underwriter, an auditor, and a policyholder
    • A browser, an operating system, and a GPU driver
    • A coordinator, training clients, and a data provider

Keep testing yourself

More quizzes picked for your level.

View all quizzes