{
 "slug": "DeepSeek_LLM",
 "title": "DeepSeek LLM",
 "type": "model",
 "short_desc": "DeepSeek's first general-purpose open-weight language model series (7B and 67B), released in late 2023 and accompanied by a scaling-laws study.",
 "categories": [
  "DeepSeek models",
  "Models",
  "2023 model releases",
  "Open-weight models"
 ],
 "infobox": {
  "Developer": "DeepSeek",
  "Family": "DeepSeek",
  "Announced": "November 2023",
  "Architecture": "Dense decoder-only Transformer, broadly following the LLaMA design (pre-norm, RMSNorm, SwiGLU, rotary position embeddings); the 67B variant uses grouped-query attention",
  "Parameters": "7B and 67B",
  "Context window": "4,096 tokens",
  "Post-training": "Supervised fine-tuning and DPO for the chat variants",
  "License / access": "Open weights under a DeepSeek model license permitting commercial use; code under MIT",
  "Paper/report": "arXiv:2401.02954 (January 2024)"
 },
 "infobox_links": {
  "Developer": [
   "DeepSeek"
  ],
  "Family": [
   "DeepSeek_(model_family)"
  ],
  "Architecture": [
   "Transformer_(architecture)",
   "LLaMA"
  ],
  "Post-training": [
   "Instruction_tuning",
   "Direct_preference_optimization"
  ],
  "License / access": [
   "DeepSeek"
  ]
 },
 "aliases": [],
 "family": "DeepSeek (model family)",
 "org": "DeepSeek",
 "registry_status": "stub",
 "words": 348,
 "references": 3,
 "outbound": [
  "Chinchilla",
  "DeepSeek",
  "DeepSeek-R1",
  "DeepSeek-V2",
  "DeepSeek-V3",
  "DeepSeek_(model_family)",
  "Direct_preference_optimization",
  "GPT-3.5",
  "Instruction_tuning",
  "LLaMA",
  "Llama_2",
  "MMLU",
  "Mixture_of_experts",
  "Qwen_1",
  "RMSNorm",
  "Reinforcement_learning_from_human_feedback",
  "Scaling_laws",
  "Transformer_(architecture)"
 ],
 "inbound": [
  "Comma_(models)",
  "DeepSeek",
  "DeepSeek-Coder-V2",
  "DeepSeek-V2",
  "DeepSeek_(model_family)",
  "InternLM",
  "Yi_(model_family)",
  "iFlytek_Spark"
 ],
 "url": "wiki/DeepSeek_LLM.html",
 "built": "2026-07-24"
}