Sep 4 08:01 EmbeddingGemma adaptive Core ML bucket set — int8 S64/S128/S256/S512/S2048, accepted manifest fd19e3c3f85e6aa9; compiled bundles ship as bytes because coremlcompiler does not reproduce coremldata.bin Aug 27 16:43 model card: standard structure, measured facts, verified license declarations Aug 17 23:37 docs: refresh measured performance (local retest 2026-08-18) 22:38 docs: descriptive model card Ben Graville
20:22 rebuild-verification 2026-08-17: Linux rebuild reproduces weight.bin BIT-EXACT (sha256 f81f60eb…, 0/308,616,576 bytes differ) and model_config.json + tokenizer identical; encoder.mlmodelc container not reproducible off macOS (coremlcompiler) Ben Graville
19:43 embeddinggemma-300m Core ML bundle (ANE encoder) — production embedding model Ben Graville
19:43 initial commit Ben Graville
