{
  "id": "zephyr-7b-beta",
  "name": "Zephyr 7B β",
  "identifiers": {
    "huggingface": [
      "HuggingFaceH4/zephyr-7b-beta"
    ]
  },
  "developer": "Hugging Face (H4 team)",
  "release_date": {
    "value": "2023-10-25",
    "source": "https://arxiv.org/abs/2310.16944",
    "status": "partial",
    "note": "Date of the paper's arXiv v1; the weights' own publication date was not confirmed."
  },
  "weights_status": "open",
  "availability": {
    "value": "available",
    "checked": "2026-09-24",
    "source": "https://huggingface.co/HuggingFaceH4/zephyr-7b-beta"
  },
  "license": {
    "value": "MIT",
    "source": "https://huggingface.co/HuggingFaceH4/zephyr-7b-beta",
    "status": "recorded"
  },
  "architecture": {
    "family": "decoder_only",
    "n_layers": {
      "value": 32,
      "source": "https://huggingface.co/HuggingFaceH4/zephyr-7b-beta/raw/892b3d7a7b1cf10c7a701c60881cd93df615734c/config.json",
      "status": "recorded"
    },
    "hidden_size": {
      "value": 4096,
      "source": "https://huggingface.co/HuggingFaceH4/zephyr-7b-beta/raw/892b3d7a7b1cf10c7a701c60881cd93df615734c/config.json",
      "status": "recorded"
    },
    "n_heads": {
      "value": 32,
      "source": "https://huggingface.co/HuggingFaceH4/zephyr-7b-beta/raw/892b3d7a7b1cf10c7a701c60881cd93df615734c/config.json",
      "status": "recorded"
    },
    "vocab_size": {
      "value": 32000,
      "source": "https://huggingface.co/HuggingFaceH4/zephyr-7b-beta/raw/892b3d7a7b1cf10c7a701c60881cd93df615734c/config.json",
      "status": "recorded"
    },
    "positional_encoding": {
      "value": "rotary (RoPE)",
      "source": "https://huggingface.co/HuggingFaceH4/zephyr-7b-beta/raw/892b3d7a7b1cf10c7a701c60881cd93df615734c/config.json",
      "status": "recorded"
    },
    "context_length": {
      "value": 32768,
      "source": "https://huggingface.co/HuggingFaceH4/zephyr-7b-beta/raw/892b3d7a7b1cf10c7a701c60881cd93df615734c/config.json",
      "status": "recorded",
      "note": "Config max_position_embeddings, inherited from Mistral 7B's config; the Mistral paper gives 8192 as the design context length."
    },
    "n_kv_heads": {
      "value": 8,
      "source": "https://huggingface.co/HuggingFaceH4/zephyr-7b-beta/raw/892b3d7a7b1cf10c7a701c60881cd93df615734c/config.json",
      "status": "recorded"
    },
    "sliding_window": {
      "value": 4096,
      "source": "https://huggingface.co/HuggingFaceH4/zephyr-7b-beta/raw/892b3d7a7b1cf10c7a701c60881cd93df615734c/config.json",
      "status": "recorded"
    }
  },
  "training_data": {
    "value": "Filtered UltraChat (synthetic dialogues 'generated by ChatGPT') for distilled SFT; UltraFeedback (64k prompts with model completions 'ranked by GPT-4') for distilled DPO.",
    "source": "https://huggingface.co/HuggingFaceH4/zephyr-7b-beta",
    "status": "recorded"
  },
  "techniques": [
    "instruction-tuning",
    "dpo"
  ],
  "primary_sources": [
    "https://arxiv.org/abs/2310.16944",
    "https://huggingface.co/HuggingFaceH4/zephyr-7b-beta",
    "https://huggingface.co/HuggingFaceH4/zephyr-7b-beta/raw/892b3d7a7b1cf10c7a701c60881cd93df615734c/config.json"
  ],
  "record_history": [
    {
      "date": "2026-09-24",
      "change": "created from primary sources (Phase 1 seed, batch 1)",
      "by": "wilson-pruitt + claude"
    },
    {
      "date": "2026-09-24",
      "change": "note on context_length provenance",
      "by": "wilson-pruitt + claude"
    },
    {
      "date": "2026-09-24",
      "change": "availability checked and recorded",
      "by": "wilson-pruitt + claude"
    },
    {
      "date": "2026-09-24",
      "change": "added fine_tuned_from edge to mistral-7b-sft-beta (SFT stage), per that card",
      "by": "claude (Sonnet 5), tranche-2 follow-up"
    }
  ]
}