{
  "id": "gpt2-medium",
  "name": "gpt2-medium",
  "identifiers": {
    "huggingface": [
      "openai-community/gpt2-medium"
    ]
  },
  "developer": "OpenAI",
  "release_date": {
    "value": "2019-04",
    "source": "https://web.archive.org/web/20240527012120/https://openai.com/index/gpt-2-6-month-follow-up/",
    "status": "partial",
    "note": "The follow-up post's own retrospective timeline says 'April 2019 Released medium parameter (355M) model.' The originally-announced size was described elsewhere as '345M' in the initial post (as archived) -- unresolved naming discrepancy between OpenAI's own posts; 355M matches this repo's actual parameter count (config: 24 layers, 1024 hidden = ~355M)."
  },
  "weights_status": "open",
  "availability": {
    "value": "available",
    "checked": "2026-09-24",
    "source": "https://huggingface.co/openai-community/gpt2-medium"
  },
  "license": {
    "value": "mit",
    "source": "https://huggingface.co/openai-community/gpt2-medium/blob/6dcaa7a952f72f9298047fd5137cd6e4f05f41da/README.md",
    "status": "partial",
    "note": "From HF card metadata (uploader-declared); confirm against the license text."
  },
  "architecture": {
    "family": "decoder_only",
    "note": "family read from config 'architectures': ['GPT2LMHeadModel'].",
    "n_layers": {
      "value": 24,
      "source": "https://huggingface.co/openai-community/gpt2-medium/raw/6dcaa7a952f72f9298047fd5137cd6e4f05f41da/config.json",
      "status": "recorded"
    },
    "hidden_size": {
      "value": 1024,
      "source": "https://huggingface.co/openai-community/gpt2-medium/raw/6dcaa7a952f72f9298047fd5137cd6e4f05f41da/config.json",
      "status": "recorded"
    },
    "n_heads": {
      "value": 16,
      "source": "https://huggingface.co/openai-community/gpt2-medium/raw/6dcaa7a952f72f9298047fd5137cd6e4f05f41da/config.json",
      "status": "recorded"
    },
    "vocab_size": {
      "value": 50257,
      "source": "https://huggingface.co/openai-community/gpt2-medium/raw/6dcaa7a952f72f9298047fd5137cd6e4f05f41da/config.json",
      "status": "recorded"
    },
    "context_length": {
      "value": 1024,
      "source": "https://huggingface.co/openai-community/gpt2-medium/raw/6dcaa7a952f72f9298047fd5137cd6e4f05f41da/config.json",
      "status": "recorded"
    },
    "positional_encoding": {
      "value": "learned_absolute",
      "source": "https://cdn.openai.com/research-covers/language-unsupervised/language_understanding_paper.pdf",
      "status": "recorded",
      "note": "GPT-2 paper: model 'largely follows the details of the OpenAI GPT model'; GPT-1 paper: 'We used learned position embeddings instead of the sinusoidal version'."
    },
    "activation": {
      "value": "GELU (tanh approximation, config 'gelu_new')",
      "source": "https://cdn.openai.com/research-covers/language-unsupervised/language_understanding_paper.pdf",
      "status": "recorded"
    }
  },
  "training_data": {
    "value": "WebText: web pages linked from Reddit posts with karma >= 3. Same corpus as the 1.5B (XL) model.",
    "source": "https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf",
    "status": "recorded"
  },
  "techniques": [],
  "primary_sources": [
    "https://huggingface.co/openai-community/gpt2-medium",
    "https://huggingface.co/openai-community/gpt2-medium/raw/6dcaa7a952f72f9298047fd5137cd6e4f05f41da/config.json"
  ],
  "record_history": [
    {
      "date": "2026-09-24",
      "change": "ingested as candidate from HF (openai-community/gpt2-medium@6dcaa7a952f72f9298047fd5137cd6e4f05f41da)",
      "by": "ingest_hf.py"
    },
    {
      "date": "2026-09-24",
      "change": "preparer: developer, release date, positional encoding, activation, trained_on set",
      "by": "claude (preparer, Sonnet 5)"
    },
    {
      "date": "2026-09-24",
      "change": "preparer cleanup: removed 2 stale flag(s), filled training_data",
      "by": "claude (preparer, Sonnet 5)"
    },
    {
      "date": "2026-09-24",
      "change": "reviewed and promoted from staging (1 edge(s) accepted)",
      "by": "Wilson Pruitt"
    }
  ]
}