Instructions to use Micklavin/gemma-4-12B-it-4bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Micklavin/gemma-4-12B-it-4bit with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("Micklavin/gemma-4-12B-it-4bit")

prompt = "Write a story about Einstein"
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True
)

text = generate(model, tokenizer, prompt=prompt, verbose=True)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use Micklavin/gemma-4-12B-it-4bit with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Micklavin/gemma-4-12B-it-4bit"

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "Micklavin/gemma-4-12B-it-4bit"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

MLX LM

How to use Micklavin/gemma-4-12B-it-4bit with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "Micklavin/gemma-4-12B-it-4bit"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "Micklavin/gemma-4-12B-it-4bit"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "Micklavin/gemma-4-12B-it-4bit",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Hermes Agent

How to use Micklavin/gemma-4-12B-it-4bit with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Micklavin/gemma-4-12B-it-4bit"

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default Micklavin/gemma-4-12B-it-4bit

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use Micklavin/gemma-4-12B-it-4bit with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Micklavin/gemma-4-12B-it-4bit"

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "Micklavin/gemma-4-12B-it-4bit" \
  --custom-provider-id mlx-lm \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Access Gemma on Hugging Face

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

To access Gemma on Hugging Face, you must review and agree to Google's Gemma Terms of Use and Prohibited Use Policy. Requests are processed immediately.

Gemma 4 12B IT — 4-bit MLX

A 4-bit quantization of google/gemma-4-12B-it for MLX on Apple Silicon.

Produced for Triad, a local three-model token-fusion experiment. These are the exact weights that project was built and validated against, published so its results can be reproduced.

Size on disk: 6.3 GB (down from ~24.8 GB at bf16)

⚠️ License notice — read before using

Gemma is not released under an open-source license. It is governed by the Gemma Terms of Use and the Gemma Prohibited Use Policy, which impose restrictions that Apache-2.0 and similar licenses do not.

By downloading or using this model you agree to those terms, exactly as if you had obtained it from Google directly. Quantization creates a derivative — it does not relicense anything. If you redistribute this model or anything derived from it, you are responsible for passing these terms and restrictions on to your recipients.

Some copies of Gemma quantizations circulating on the Hub are mislabelled apache-2.0, including the upstream metadata this conversion inherited. That label was incorrect and has been corrected here. Do not rely on a license tag alone — read the terms.

Requirements

Apple Silicon
A build of mlx-lm that supports gemma4_unified — the PyPI release does not

⚠️ pip install mlx-lm is not sufficient for this model

This model's config declares model_type: gemma4_unified. mlx-lm resolves that through its MODEL_REMAPPING table (gemma4_unified → gemma4), but that entry is not present in the PyPI release — verified against the mlx-lm 0.31.3 wheel, whose MODEL_REMAPPING has no gemma4_unified key and which ships no models/gemma4_unified.py. Installing from PyPI and calling load() on this repo will fail to resolve the architecture.

Install from the git commit this model was built and verified against:
pip install "mlx-lm @ git+https://github.com/ml-explore/mlx-lm.git@cf10f962b7a20e63a6df43dbf0faf06070153d40"
Once a PyPI release includes gemma4_unified in MODEL_REMAPPING, plain pip install mlx-lm will work and this note can be ignored. Check before assuming either way.

No packages beyond mlx-lm are required at load time. (mlx-optiq is used during conversion — see Quantization provenance — but is not needed to load this repo: verified by blocking the optiq import and loading successfully.)

Usage

from mlx_lm import generate, load

model, tokenizer = load("Micklavin/gemma-4-12B-it-4bit")

prompt = tokenizer.apply_chat_template(
    [{"role": "user", "content": "What is the chemical symbol for gold?"}],
    tokenize=False,
    add_generation_prompt=True,
)
print(generate(model, tokenizer, prompt=prompt, max_tokens=256))

Thinking mode

Gemma 4's chat template branches on enable_thinking and emits a <|channel>thought block. Triad disables it so all ensemble members answer in the same phase:

prompt = tokenizer.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True, enable_thinking=False
)

Note that with thinking disabled this template still opens a thought channel where other models' templates close theirs. If you filter model output, account for that.

Quantization provenance

Converted with mlx_lm.convert:

convert(
    hf_path="google/gemma-4-12B-it",
    mlx_path="models/gemma-4bit",
    quantize=True,
    q_bits=4,
    q_group_size=64,
    dtype="bfloat16",
)

Resulting config: {"group_size": 64, "bits": 4, "mode": "affine"}, model_type: gemma4_unified. Per mlx-lm's remapping comment, this is the encoder-free multimodal variant with vision and audio weights stripped by sanitize() — so this is a text-only conversion.

Component	Version
mlx	0.32.0
mlx-lm	0.31.3
transformers	5.12.1

The reproduction script is scripts/quantize_models.py.

Evaluation

None beyond a smoke test. No benchmark comparison against the bf16 original was run, so the quantization's quality cost is unmeasured. Treat it as an untested 4-bit conversion rather than a validated one.

License

Gemma Terms of Use. See the notice above — this is a custom license with use restrictions, not an open-source one.

Downloads last month: 8

Safetensors

Model size

2B params

Tensor type

BF16

U32

MLX

Hardware compatibility

4-bit

Model tree for Micklavin/gemma-4-12B-it-4bit

Base model

google/gemma-4-12B

Finetuned

google/gemma-4-12B-it

Quantized

(299)

this model

Micklavin
/

gemma-4-12B-it-4bit

Access Gemma on Hugging Face

Gemma 4 12B IT — 4-bit MLX

⚠️ License notice — read before using

Requirements

⚠️ `pip install mlx-lm` is not sufficient for this model

Usage

Thinking mode

Quantization provenance

Evaluation

License

Model tree for Micklavin/gemma-4-12B-it-4bit

Access Gemma on Hugging Face

Gemma 4 12B IT — 4-bit MLX

⚠️ License notice — read before using

Requirements

⚠️ pip install mlx-lm is not sufficient for this model

Usage

Thinking mode

Quantization provenance

Evaluation

License

Model tree for Micklavin/gemma-4-12B-it-4bit

⚠️ `pip install mlx-lm` is not sufficient for this model