# First-party vocabulary receipt — gemma4:26b n_vocab = 262,144

Read 2026-08-21 ~10:5xZ from the inference box's own server journal: the
llama-server load print for the model blob attributed to gemma4:26b.

Blob attribution (from the runtime's own manifests, same method as the
published CONTENTION-EVIDENCE file): gemma4:26b -> model-layer digest
7121486771cbfe21...

Journal lines (instants stated in UTC; the load is 2026-08-21T10:13:11Z —
this session's re-seat of the model at num_ctx 32768):

  starting llama-server ... --model .../blobs/sha256-7121486771cbfe218851513210c40b35dbdee93ab1ef43fe36283c883980f0df --port 45085 ... -c 32768 ...
  print_info: n_vocab               = 262144

Also read first-party the same day via /api/show:
  general.parameter_count = 25,805,936,462  (= 25.8B total)
  gemma4.expert_count = 128, gemma4.expert_used_count = 8
  gemma4.context_length = 262144

NOTE for hostile readers: n_vocab (262,144) and context_length (262,144)
are numerically identical on this model — both 2^18 — and are receipted by
SEPARATE fields/log lines above; neither figure was derived from the other.
The /api/show model_info block on ollama 0.32.13 exposes no vocab field;
the load-log print is the first-party source.