{
  "id": "hf-load-choices",
  "title": "Hugging Face model load choices",
  "topic": "huggingface",
  "columns": [
    "Choice",
    "Meaning",
    "Related slug"
  ],
  "rows": [
    [
      "AutoModel* + from_pretrained",
      "Resolve class from config; load weights",
      "hf-transformers-auto-classes"
    ],
    [
      "device_map / meta / shards",
      "Fit large models across devices",
      "hf-transformers-big-model-loading"
    ],
    [
      "KV cache class",
      "Speed decoding; pick cache implementation",
      "hf-transformers-kv-cache"
    ],
    [
      "Quantization path",
      "Load-time vs pre-quantized weights",
      "hf-transformers-quantization-overview"
    ],
    [
      "Attention backend",
      "sdpa / flash / flex / paged tradeoffs",
      "hf-transformers-attention-backends"
    ]
  ],
  "related_slugs": [
    "hf-transformers-auto-classes",
    "hf-transformers-big-model-loading",
    "hf-transformers-kv-cache",
    "hf-transformers-quantization-overview",
    "hf-transformers-attention-backends"
  ],
  "provenance": [
    {
      "title": "iinventedlight Library notes",
      "url": "https://iinventedlight.com/library/notes/",
      "license": "per-note",
      "accessed": "2026-11-09"
    }
  ],
  "as_of": "2026-11-09",
  "license": "CC-BY-4.0 summaries; see related notes for source licenses",
  "markdown": "/library/artifacts/hf-load-choices.table.md"
}