Instructions to use MichaelAnthony/gemma4-e2b-Snowfox-MLX with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use MichaelAnthony/gemma4-e2b-Snowfox-MLX with MLX:

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

from mlx_vlm import load, generate
from mlx_vlm.prompt_utils import apply_chat_template
from mlx_vlm.utils import load_config

# Load the model
model, processor = load("MichaelAnthony/gemma4-e2b-Snowfox-MLX")
config = load_config("MichaelAnthony/gemma4-e2b-Snowfox-MLX")

# Prepare input
image = ["http://images.cocodataset.org/val2017/000000039769.jpg"]
prompt = "Describe this image."

# Apply chat template
formatted_prompt = apply_chat_template(
    processor, config, prompt, num_images=1
)

# Generate output
output = generate(model, processor, formatted_prompt, image)
print(output)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use MichaelAnthony/gemma4-e2b-Snowfox-MLX with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "MichaelAnthony/gemma4-e2b-Snowfox-MLX"

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "MichaelAnthony/gemma4-e2b-Snowfox-MLX"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent

How to use MichaelAnthony/gemma4-e2b-Snowfox-MLX 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 "MichaelAnthony/gemma4-e2b-Snowfox-MLX"

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 MichaelAnthony/gemma4-e2b-Snowfox-MLX

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use MichaelAnthony/gemma4-e2b-Snowfox-MLX with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "MichaelAnthony/gemma4-e2b-Snowfox-MLX"

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 "MichaelAnthony/gemma4-e2b-Snowfox-MLX" \
  --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"

MichaelAnthony commited on 12 days ago

Commit

95ef2e1

verified ·

1 Parent(s): 96b3495

Update README: point to standard 4-bit/6-bit quantized variants

Browse files

Files changed (1) hide show

README.md +11 -39

README.md CHANGED Viewed

@@ -76,50 +76,22 @@ For image prompting, add `--image /path/to/image.png` to the generation command.
 Use current MLX-VLM documentation for image, audio, video, and chat-template
 options.
-## Quantized variants (oMLX oQ)
-This FP16 package is the canonical source for building oMLX oQ quantizations
-on Apple Silicon. The build kit is included in this directory:
-- `build_omlx_variants.py` — builds `oQ4/oQ6/oQ8` and `oQ4e/oQ6e/oQ8e`.
-- `omlx_build_matrix.json` — the variant matrix the script mirrors.
-### Available options
-| Variant | Family | oQ level | Notes |
 | --- | --- | --- | --- |
-| `gemma4-e2b-Snowfox-MLX-FP16` | fp16 | — | shipped in this repo |
-| `gemma4-e2b-Snowfox-MLX-oQ4` | oQ | 4 | GGUF `Q4_K_M` analogue |
-| `gemma4-e2b-Snowfox-MLX-oQ6` | oQ | 6 | GGUF `Q6_K` analogue |
-| `gemma4-e2b-Snowfox-MLX-oQ8` | oQ | 8 | near-lossless |
-| `gemma4-e2b-Snowfox-MLX-oQ4e` | oQe | 4 | enhanced (sensitivity-calibrated) |
-| `gemma4-e2b-Snowfox-MLX-oQ6e` | oQe | 6 | enhanced |
-| `gemma4-e2b-Snowfox-MLX-oQ8e` | oQe | 8 | enhanced |
-The `oQ*` variants are build-to-order on Apple Silicon (see below); only the
-FP16 weights are stored in this repository.
-```bash
-# On Apple Silicon with the oMLX DMG installed:
-/Applications/oMLX.app/Contents/MacOS/python3 build_omlx_variants.py \
-  --source . --output-root .. --no-smoke-test --dry-run
-# Real run (quantizes + smoke-tests each variant):
-/Applications/oMLX.app/Contents/MacOS/python3 build_omlx_variants.py \
-  --source . --output-root ..
-```
-Produces sibling directories `../gemma4-e2b-Snowfox-MLX-oQ4`, `-oQ6`, `-oQ8`,
-`-oQ4e`, `-oQ6e`, `-oQ8e`. The `oQe` variants are the enhanced (sensitivity-
-calibrated) family; `oQ4`/`oQ6` are the oMLX analogues of GGUF `Q4_K_M`/`Q6_K`.
-> **Unverified for Gemma 4.** oMLX oQ was originally validated against
-> `qwen3_5` (Hedgehog). Gemma 4 E2B's vision + audio towers have not been
-> exercised through `quantize_oq_streaming`. If it cannot load the full
-> multimodal model, retry with `--text-only` (language backbone only) or check
-> the installed oMLX version's supported architectures. The smoke test uses
-> `mlx_vlm` (not `mlx_lm`) because this is a multimodal model; skip it with
-> `--no-smoke-test` if the installed `mlx-vlm` API differs.
 ## License

 Use current MLX-VLM documentation for image, audio, video, and chat-template
 options.
+## Quantized variants
+Standard MLX-VLM affine quantizations of SnowFox are published as separate
+repositories and are loadable directly by `mlx_vlm.generate`:
+| Variant | Quantization | Size | Notes |
 | --- | --- | --- | --- |
+| [`gemma4-e2b-Snowfox-MLX-4bit`](https://huggingface.co/MichaelAnthony/gemma4-e2b-Snowfox-MLX-4bit) | 4-bit affine, group 64 | ~6.85 GB | GGUF `Q4_K_M` analogue |
+| [`gemma4-e2b-Snowfox-MLX-6bit`](https://huggingface.co/MichaelAnthony/gemma4-e2b-Snowfox-MLX-6bit) | 6-bit affine, group 64 | ~7.43 GB | GGUF `Q6_K` analogue |
+These use MLX-VLM's standard Linear-only affine quantization (embeddings, norms,
+and convolutions stay FP16) and match `convert --quantize` / `--q-bits 6`
+output byte-for-byte.
+The earlier oMLX oQ ("oQ4/oQ6/oQ8") build-to-order plan was never published;
+use the standard 4-bit/6-bit packages above instead.
 ## License