GPT-Neo

• 📙Paper: GPT-Neo
• 📚Publisher: other
• 🏠Author Affiliation: EleutherAI
• 🔑Public: ✅
• 🌐Architecture
  • Encoder-Decoder
  • Decoder-Only
• 📏Model Size
  • 125M; 1.3B; 2.7B
• 🗂️Data pre-processing
  • Data Resource
    • The Pile
  • De-duplication: /
  • Filter Strategies
    • /
• 🍉Tokenizer
  • Technology
    • Byte-level Byte-Pair-Encoding (BBPE)
    • SentencePiece
  • Details
    • GPT-2 tokenizer
    • Train a new tokenizer on your own dataset
• 🧪Hyperparameters (GPT-Neo 2.7B)
  • optimizer: Adam
    • betas: 0.9, 0.95
    • eps: 1e-8
  • batch size: /
  • context window: 2,048
  • gradient accumulation steps: /
  • warmup steps: 3,000
  • learning rate: /
  • weight decay: 0.1
  • decay schedule
    • Cosine
    • Linear
    • Polynomial
    • Inverse Square
  • precision floating point: /
• 🏃‍♀️Training
  • model initialization: from scratch
  • training strategies
    • left-to-right
    • fill-in-the-middle
  • trained tokens/steps: /
  • hardware: Training and inference is officially supported on TPU and should work on GPU as well.
  • training time: /