{
  "algorithm": "deepseek-engram-normalization/v1",
  "algorithm_reference": "https://github.com/deepseek-ai/Engram/blob/main/engram_demo_v1.py",
  "binary": "NGRAM_CANONICAL_IDS.uint16",
  "binary_sha256": "6359312eb0b5e892828430d2c9d049021f452580f724265f2f21003565a732a3",
  "binary_size_bytes": 131072,
  "canonical_ids_contiguous": true,
  "canonical_vocabulary_size": 39789,
  "created_at": "2026-08-15T05:26:32.143022+00:00",
  "dtype": "uint16",
  "empty_normalization_fallback": "decoded_token",
  "endianness": "little",
  "entry_count": 65536,
  "manifest_sha256": "0d502c64584cf435bfff6f244ff1dc7d879ca120cd8b060c7616d6dc453c4101",
  "maximum_canonical_id": 39788,
  "minimum_canonical_id": 0,
  "normalization_steps": [
    "NFKC",
    "NFD",
    "StripAccents",
    "Lowercase",
    "CollapseAsciiWhitespace",
    "PreserveSingleSpace",
    "Strip"
  ],
  "replacement_character_fallback": "raw_tokenizer_vocab_token",
  "schema": "metis.ngram-canonical-ids/v1",
  "tokenizer_sha256": "2304b0e1c7b1ff18a3797c40d035bc781652485db4f1fc89381bfde44bbfc76c",
  "vocabulary_size": 65536
}