Runback EmbeddingGemma-300m CoreML (int8, seq-512, 768-dim, ANE)
The exact embedding pack Runback's Setup window installs (PRD 31). Converted
from unsloth/embeddinggemma-300m (weight-identical mirror of
google/embeddinggemma-300m). Tokenizer is the Gemma SentencePiece tokenizer
with added_tokens pruned 6,415 → 173 (<unusedN> placeholders dropped) for
a 37x encode speedup in swift-transformers. Governed by the Gemma Terms of
Use — see EmbeddingGemma-NOTICE.md.
IO contract in embeddinggemma_config.json (seq-len 512 — 4x the public
seq-128 export; do not mix vectors with other exports).
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support