{
 "slug": "Llama_2",
 "title": "Llama 2",
 "type": "model",
 "short_desc": "Meta's July 2023 openly licensed models (7B-70B) with RLHF chat variants; made open weights an official corporate strategy.",
 "categories": [
  "Meta models",
  "Models",
  "Open-weight models",
  "2023 model releases"
 ],
 "infobox": {
  "Developer": "Meta AI (with Microsoft as launch partner)",
  "Family": "Llama",
  "Predecessor": "LLaMA",
  "Successor": "Llama 3",
  "Announced": "July 18, 2023",
  "Architecture": "Decoder-only Transformer; GQA on larger sizes",
  "Parameters": "7B, 13B, 70B (chat and base variants)",
  "Training data": "2 trillion tokens of public data",
  "Post-training": "SFT + RLHF (chat variants)",
  "License / access": "Open weights under the Llama Community License",
  "Paper/report": "arXiv:2307.09288"
 },
 "infobox_links": {
  "Developer": [
   "Meta_AI"
  ],
  "Family": [
   "Llama_(model_family)"
  ],
  "Predecessor": [
   "LLaMA"
  ],
  "Successor": [
   "Llama_3"
  ],
  "Architecture": [
   "Transformer_(architecture)"
  ],
  "Post-training": [
   "Supervised_fine-tuning",
   "Reinforcement_learning_from_human_feedback"
  ]
 },
 "aliases": [],
 "family": "Llama (model family)",
 "org": "Meta AI",
 "registry_status": "stub",
 "words": 305,
 "references": 3,
 "outbound": [
  "Ai2",
  "Anthropic",
  "Code_Llama",
  "Hugging_Face",
  "LLaMA",
  "Llama_(model_family)",
  "Llama_3",
  "MMLU",
  "Meta_AI",
  "Mistral_AI",
  "OpenAI",
  "Reinforcement_learning_from_human_feedback",
  "Supervised_fine-tuning",
  "Transformer_(architecture)"
 ],
 "inbound": [
  "Activation_steering",
  "Amazon_Titan",
  "Attention_sink",
  "Code_Llama",
  "Comma_(models)",
  "Common_Pile",
  "DBRX",
  "Databricks",
  "DeepSeek_LLM",
  "Falcon_(models)",
  "FlashAttention",
  "Gemma",
  "LLaMA",
  "Llama_(model_family)",
  "Llama_3",
  "MPT_(models)",
  "Meta_AI",
  "Mistral_7B",
  "Mixtral_8x7B",
  "MosaicML",
  "Multi-head_attention",
  "Multi-head_latent_attention",
  "Qwen_1",
  "Self-attention",
  "State-space_model",
  "Supervised_fine-tuning",
  "The_Pile",
  "Vicuna"
 ],
 "url": "wiki/Llama_2.html",
 "built": "2026-07-24",
 "evals": {
  "provenance": {
   "source_dataset": "https://huggingface.co/datasets/open-llm-leaderboard/contents",
   "produced_with": "lm-evaluation-harness `leaderboard` task group",
   "harness": "https://github.com/EleutherAI/lm-evaluation-harness",
   "status": "leaderboard retired March 2025; result set frozen",
   "metrics": [
    "IFEval",
    "BBH",
    "MATH Lvl 5",
    "GPQA",
    "MUSR",
    "MMLU-PRO"
   ],
   "values": "normalized 0-100 scores as published; raw values remain in the source dataset",
   "matching": "registry links.hf repo id plus explicit suffix variants; flagged submissions excluded",
   "fetched": "2026-07-19",
   "rows_total": 4576
  },
  "checkpoints": [
   {
    "checkpoint": "meta-llama/Llama-2-7b-hf",
    "url": "https://huggingface.co/meta-llama/Llama-2-7b-hf",
    "revision": "01c7f73d771dfac7d292323805ebc428287df4f9",
    "precision": "float16",
    "params_b": 6.738,
    "submitted": "2024-06-12",
    "average": 8.81,
    "scores": {
     "IFEval": 25.19,
     "BBH": 10.35,
     "MATH Lvl 5": 1.74,
     "GPQA": 2.24,
     "MUSR": 3.76,
     "MMLU-PRO": 9.57
    }
   },
   {
    "checkpoint": "meta-llama/Llama-2-7b-chat-hf",
    "url": "https://huggingface.co/meta-llama/Llama-2-7b-chat-hf",
    "revision": "f5db02db724555f92da89c216ac04704f23d4590",
    "precision": "float16",
    "params_b": 6.738,
    "submitted": "2024-08-30",
    "average": 9.61,
    "scores": {
     "IFEval": 39.86,
     "BBH": 4.46,
     "MATH Lvl 5": 1.96,
     "GPQA": 0.45,
     "MUSR": 3.28,
     "MMLU-PRO": 7.64
    }
   }
  ]
 }
}