Momento fixed-shape Gemma 4 E2B decoder (Core AI, chunk16)

The exact decoder files the Momento app (github.com/louis6962/Momento, branch CoreAI-transition) downloads and hash-checks on first run. Fixed 4,096-slot host-managed KV; functions main (query-1) and prefill (query-16). Exported from google/gemma-4-E2B-it-qat-q4_0-unquantized @ 6befbaca7398925921802abd1f277b495b78b738. Tokenizer, vision and PLE tables come from mlboydaisuke/gemma-4-E2B-CoreAI @ b2643e50.

Directory	Loaded by
gemma4-fixed-pf16-d353c18c7b1dbcdf.aimodel	Apple silicon Mac running the iPhone app (Designed for iPad); Core AI specializes and caches it
gemma4-fixed-pf16-d353c18c7b1dbcdf.h18p.aimodelc	iPhone 17-class (h18p), compiled ahead of time; never load on a Mac

SHA256SUMS lists every file; the app refuses any mismatch. Export recipe and limits: macos/ModelExport/FixedGemma/README.md in the app repository. Debug validation asset, not a qualified release model.

Gemma 4 is licensed under Apache-2.0 (https://ai.google.dev/gemma/docs/gemma_4_license). These files are modified derivatives (re-exported and compiled) of Gemma 4 E2B.