{
 "slug": "Instruction_tuning",
 "title": "Instruction tuning",
 "type": "technique",
 "short_desc": "Supervised fine-tuning of pretrained language models on instruction-response pairs, teaching them to follow natural-language tasks zero-shot.",
 "categories": [
  "Post-training",
  "Training methods"
 ],
 "infobox": {
  "Introduced": "FLAN (Wei et al.) and T0 (Sanh et al.), 2021; SFT stage of InstructGPT, 2022",
  "Field": "Post-training",
  "Used in": "Essentially every deployed assistant model",
  "Key papers": "arXiv:2109.01652, arXiv:2110.08207, arXiv:2203.02155"
 },
 "infobox_links": {
  "Introduced": [
   "Flan-T5",
   "InstructGPT"
  ]
 },
 "aliases": [
  "Self-Instruct",
  "T0"
 ],
 "family": "",
 "org": "",
 "registry_status": "stub",
 "words": 288,
 "references": 3,
 "outbound": [
  "Alpaca",
  "Anthropic",
  "Constitutional_AI",
  "EleutherAI",
  "Flan-T5",
  "Hugging_Face",
  "InstructGPT",
  "LLaMA",
  "Large_language_model",
  "Llama_(model_family)",
  "MMLU",
  "Meta_AI",
  "Mistral_AI",
  "Next-token_prediction",
  "OpenAI",
  "Reinforcement_learning_from_human_feedback",
  "Supervised_fine-tuning",
  "Transformer_(architecture)"
 ],
 "inbound": [
  "AFM-4.5B",
  "AI21_Labs",
  "Activation_verbalizer",
  "Apple_foundation_models",
  "Arcee_AI",
  "Aya",
  "BLIP",
  "ChatGLM",
  "Code_Llama",
  "Codestral",
  "Cohere",
  "Comma_(models)",
  "Constitutional_AI",
  "DBRX",
  "DeepSeek-Coder-V2",
  "DeepSeek_LLM",
  "ERNIE_Bot",
  "Flan-T5",
  "GLM-4",
  "GLM-4.5",
  "GLM-4.6",
  "GPT-1",
  "GPT-4.5",
  "GPT-4_Turbo",
  "GPT-4o_mini",
  "GPT-J",
  "GPT-Neo",
  "GPT-NeoX-20B",
  "Gemini_1.0",
  "Grok-2",
  "IBM_Granite",
  "InternLM",
  "InternVL",
  "Jurassic_(models)",
  "Knowledge_distillation",
  "LLaMA",
  "LLaVA",
  "LaMDA",
  "Large_language_model",
  "Ling_(models)",
  "Llama_3.1",
  "Llama_3.2",
  "Llama_3.3",
  "MPT_(models)",
  "Mechanistic_interpretability",
  "MiniMax-Text-01",
  "Mistral_AI",
  "Mistral_Large",
  "Mistral_Medium_3",
  "Mistral_Small",
  "Mixture_of_experts",
  "Molmo",
  "Nemotron",
  "Nous_Research",
  "PaLM",
  "PaLM_2",
  "Pixtral",
  "Pretraining",
  "Qwen2",
  "Qwen2.5",
  "Qwen2.5-Max",
  "Qwen2.5-VL",
  "Qwen3",
  "Qwen3-Coder",
  "Qwen3-Coder-Next",
  "Qwen3-Max",
  "Qwen_1",
  "Seed-OSS",
  "SmolLM",
  "Sparse_autoencoder",
  "Step-2",
  "Supervised_fine-tuning",
  "T5",
  "Thinking_Machines_Lab",
  "Transformer_(architecture)",
  "Trinity_(model_family)",
  "Trinity_Large",
  "Trinity_Large_Thinking",
  "Tulu_3",
  "ULMFiT",
  "Vicuna",
  "YaLM-100B",
  "Yi_(model_family)",
  "iFlytek_Spark"
 ],
 "url": "wiki/Instruction_tuning.html",
 "built": "2026-07-24"
}
