milmmt-46-4b-4bit-mlx

MiLMMT-46-4B — 4-bit MLX

A 4-bit MLX conversion of xiaomi-research/MiLMMT-46-4B-v0.1, a multilingual machine-translation model built on Google Gemma 3 (4B). This build is used for on-device English ↔ Simplified Chinese translation in the Duet app.

License: Gemma Terms of Use. Gemma is provided under and subject to the Gemma Terms of Use found at ai.google.dev/gemma/terms. Use is also subject to the Gemma Prohibited Use Policy. See NOTICE and LICENSE.

Provenance

Modification (marking of changes)

The source bf16 weights were quantized to 4-bit and converted to MLX format with:

mlx-lm convert --hf-path <MiLMMT-46-4B bf16> --mlx-path model -q --q-bits 4 --q-group-size 64

Tokenizer, config.json, and generation_config.json are redistributed to run the converted model. No other behavioural changes were made. These files are therefore modified from the original and marked as such here and in NOTICE, as required by the Gemma Terms of Use §3.1.

Repository layout

README.md                     # this card (Gemma license notice)
NOTICE                        # required Gemma notice + provenance + modification record
LICENSE                       # Gemma Terms of Use (link + key conditions)
model/                        # the model payload the app downloads
  model.safetensors
  model.safetensors.index.json
  config.json
  generation_config.json
  tokenizer.json
  tokenizer_config.json
  README.md
  model_manifest.json         # per-file SHA-256 + expected file list (integrity)

Use restrictions (pass-through)

By downloading or using these weights you agree to the Gemma Terms of Use and the Gemma Prohibited Use Policy, and you must pass these restrictions on to any further recipients.

Non-affiliation

"Gemma" is a model and trademark of Google LLC; MiLMMT is a model of Xiaomi. This is an independent redistribution and is not endorsed by or affiliated with Google or Xiaomi. All rights in the underlying model remain with their owners.