over-show
/

gemma-4-e2b-it-text-only-4bit

Text Generation

4-bit precision

Model card Files Files and versions

Gemma 4 E2B IT Text-Only 4-bit MLX Repack

Text-only repack of mlx-community/gemma-4-e2b-it-4bit for Overshow local inference via MLX Swift.

This artefact keeps the Gemma 4 language model tensors and tokenizer files, strips the language_model. tensor prefix expected by the pinned MLX Swift text loader, and drops unused audio and vision tower tensors.

Source

Source repository: mlx-community/gemma-4-e2b-it-4bit
Source revision: 99d9a53ff828d365a8ecae538e45f80a08d612cd
Repack script: scripts/repack-gemma4-text-only.py in over-show/app

Bundle Shape

Required files:

config.json
model.safetensors
model.safetensors.index.json
tokenizer.json
tokenizer_config.json
generation_config.json

Validation

Local validation passed before publishing:

Tensor plan: 1,234 tensors kept, 1,415 dropped
Weights: 2,512.5 MiB, saving 902.7 MiB against the source MLX checkpoint
scripts/validate-mlx-helper.py: 5/5 commands passed
swift test --filter Gemma4LoadSmokeTest: 2/2 tests passed

Downloads last month: 678

Safetensors

Model size

0.7B params

Tensor type

U32

·

BF16

·

MLX

Hardware compatibility

Log In to add your hardware

4-bit

Model tree for over-show/gemma-4-e2b-it-text-only-4bit

Base model

mlx-community/gemma-4-e2b-it-4bit

Quantized

(2)

this model

Collection including over-show/gemma-4-e2b-it-text-only-4bit

gemma 4

1 item • Updated 20 days ago