dreamer1cc/hovor-gemma-4-e2b-coreml-int4

Gemma 4 E2B — CoreML int4 (Apple Neural Engine)

Converted from google/gemma-4-E2B-it for on-device inference on iOS 18+ / macOS 15+ via Apple Neural Engine (ANE).

Files

File	Purpose
model.mlpackage/	CoreML stateful transformer (int4 palettized, CPU+ANE)
model_config.json	Metadata for the Swift inference engine
tokenizer.json, tokenizer_config.json, tokenizer.model	Tokenizer files
special_tokens_map.json	Special token definitions

Chat template

Gemma 4 uses new turn markers (different from Gemma 3):

<bos><|turn>user
{user message}<turn|>
<|turn>model
{model response}<turn|>

With generation prompt (inference):

<bos><|turn>user
{user message}<turn|>
<|turn>model

Attribution

Built with Gemma. Subject to the Gemma Terms of Use.