You can deploy this model and start earning money today!

dreamer1cc/
hovor-gemma-3-4b-it-coreml-int4

Name: hovor-gemma-3-4b-it-coreml-int4
Availability: InStock

Views

No views yet

Gemma 3 4B IT — CoreML int4 (Apple Neural Engine)

Converted from google/gemma-3-4b-it for on-device inference on iOS 18+ / macOS 15+ via Apple Neural Engine (ANE).

Used in Hovor iOS quality comparison benchmarks.

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

Gemma 3 uses the <start_of_turn> / <end_of_turn> turn marker format:

<bos><start_of_turn>user
{user message}<end_of_turn>
<start_of_turn>model
{model response}<end_of_turn>

With generation prompt (inference):

<bos><start_of_turn>user
{user message}<end_of_turn>
<start_of_turn>model

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