Matt Oswalt
Codex
  • Linux
    • File Descriptors
    • Networking
      • eBPF
      • Sockets
  • LLM
    • Resources
    • Inference Stack
    • Apps & Libraries
    • Model Evaluation
    • Memory
    • Glossary
  • Machine Learning
    • Deep Learning
    • Machine Learning
    • Glossary
  • Math
    • Glossary
  • Rust
    • Common Traits
    • Ownership
  • Video
    • GoPro
  • Cheat Sheets
Matt Oswalt
  • Blogs
    • All Categories
    • Rust
    • General Programming
    • Systems
    • Machine Learning
    • Personal
  • Codex
  • Bookclub
  • Portfolio
  • Sponsor Me!
  • Github
  • Twitter
  • Twitch
  • LinkedIn
  • YouTube
  • Facebook
  • Bluesky
  • RSS

Search

Loading search index…

No recent searches

No results for "Query here"

  • to select
  • to navigate
  • to close

Search by FlexSearch

  • Linux
    • File Descriptors
    • Networking
      • eBPF
      • Sockets
  • LLM
    • Resources
    • Inference Stack
    • Apps & Libraries
    • Model Evaluation
    • Memory
    • Glossary
  • Machine Learning
    • Deep Learning
    • Machine Learning
    • Glossary
  • Math
    • Glossary
  • Rust
    • Common Traits
    • Ownership
  • Video
    • GoPro
  • Cheat Sheets

This Glossary

  • Context window
  • Dense model
  • Distilled model
  • Instruct model
  • KV cache
  • MoE (Mixture of Experts)
  • Open-weight model
  • Parameter
  • Reasoning model
  • RLHF (Reinforcement Learning from Human Feedback)
  • t/s (tokens per second)
  • Token

Instruct model

← Back to Glossary

A base model fine-tuned with instruction-following data and/or RLHF to make it useful for conversation and following directions. Contrasted with a raw base/pretrained model which just predicts next tokens.

  • Chat and conversational models - Hugging Face
Referenced in
  • RLHF (Reinforcement Learning from Human Feedback)
Prev
GGUF
Next
IQ (importance-matrix quantization)
    • © 2010 - 2026 Matt Oswalt · Powered by Hugo & Hyas.