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

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use MichaelAnthony/gemma4-e2b-Snowfox-GGUF with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf MichaelAnthony/gemma4-e2b-Snowfox-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf MichaelAnthony/gemma4-e2b-Snowfox-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf MichaelAnthony/gemma4-e2b-Snowfox-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf MichaelAnthony/gemma4-e2b-Snowfox-GGUF:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf MichaelAnthony/gemma4-e2b-Snowfox-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf MichaelAnthony/gemma4-e2b-Snowfox-GGUF:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf MichaelAnthony/gemma4-e2b-Snowfox-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf MichaelAnthony/gemma4-e2b-Snowfox-GGUF:Q4_K_M

Use Docker

docker model run hf.co/MichaelAnthony/gemma4-e2b-Snowfox-GGUF:Q4_K_M

LM Studio
Jan
Ollama
How to use MichaelAnthony/gemma4-e2b-Snowfox-GGUF with Ollama:
```
ollama run hf.co/MichaelAnthony/gemma4-e2b-Snowfox-GGUF:Q4_K_M
```
Unsloth Desktop

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

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf MichaelAnthony/gemma4-e2b-Snowfox-GGUF:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "MichaelAnthony/gemma4-e2b-Snowfox-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use MichaelAnthony/gemma4-e2b-Snowfox-GGUF with Docker Model Runner:
```
docker model run hf.co/MichaelAnthony/gemma4-e2b-Snowfox-GGUF:Q4_K_M
```

Lemonade

How to use MichaelAnthony/gemma4-e2b-Snowfox-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull MichaelAnthony/gemma4-e2b-Snowfox-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.gemma4-e2b-Snowfox-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use MichaelAnthony/gemma4-e2b-Snowfox-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf MichaelAnthony/gemma4-e2b-Snowfox-GGUF:Q4_K_M

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-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

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

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf MichaelAnthony/gemma4-e2b-Snowfox-GGUF:Q4_K_M

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-GGUF:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --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 14 days ago

Commit

fe5b1e9

verified ·

1 Parent(s): 022b917

docs: correct GGUF model card

Browse files

Files changed (1) hide show

README.md +40 -24

README.md CHANGED Viewed

@@ -1,48 +1,64 @@
 ---
 base_model: google/gemma-4-E2B-it-qat-q4_0-unquantized
-license: gemma
 library_name: llama.cpp
 tags:
 - gemma4
 - gguf
 - qat-derived
-- peft
-- rocm
 - snowfox
 ---
-# gemma4-e2b-Snowfox
-`gemma4-e2b-Snowfox` is a validation-selected, language-only SnowFox LoRA fine-tune of Google's Gemma 4 E2B instruction QAT-derived BF16 checkpoint. The vision and audio encoders remain frozen. The adapter was trained with PyTorch ROCm through Unsloth.
-This release is accurately described as **QAT-derived**. The post-LoRA weights were not separately re-QAT calibrated.
-## Downloads
-All GGUF variants are available at the repository root for direct download. Use exactly one LLM quant plus the BF16 multimodal projector when your runtime supports Gemma 4 multimodality.
-| Variant | File | Size | SHA-256 |
-| --- | --- | ---: | --- |
-| Q4_0 | `gemma4-e2b-Snowfox-Q4_0.gguf` | 3,349,516,000 bytes | `105e63469457449dcdc147785c4317b7891a9aafaec2e40bdcbc1b16c6fd8f3b` |
-| Q4_K_M | `gemma4-e2b-Snowfox-Q4_K_M.gguf` | 3,416,120,032 bytes | `f4a38a2a95cefbf91659c355aa80410c9e0b307072c3dc7deca842b2d76d613a` |
-| Q6_K | `gemma4-e2b-Snowfox-Q6_K.gguf` | 3,829,835,488 bytes | `e7d2a0c8cc168db9652799d3bfd8198563dc821bae424b4b8ed1a767bbc51537` |
-| Q8_0 | `gemma4-e2b-Snowfox-Q8_0.gguf` | 4,947,414,752 bytes | `44e7f9dd879f0ae2b18f57f78819e147783b4795738b3d3321f6e20542fa6a68` |
-| Multimodal projector | `gemma4-e2b-Snowfox-BF16-mmproj.gguf` | 986,833,408 bytes | `cbc8946d8e6a85714c7f4868302ccd7bf716301b47a5efefe5850eb5348b0a3e` |
-No older Qwen-based SnowFox artifact is included in this package.
 ## Lineage and validation
-- Base model: [`google/gemma-4-E2B-it-qat-q4_0-unquantized`](https://huggingface.co/google/gemma-4-E2B-it-qat-q4_0-unquantized)
 - Pinned base revision: `6befbaca7398925921802abd1f277b495b78b738`
-- Required-field accuracy: `0.9984962406015038`
-- Strict JSON validity: `1.0`
-- Native tool-call validity: `1.0`
-- Contract success: `0.9936305732484076`
-The deterministic validation curriculum was held out from training and contains no BenchLocal cases or answers. These validation results do not imply unmeasured external benchmark scores.
 ## License
-This derivative release is subject to the [Gemma terms](https://ai.google.dev/gemma/terms) and the base model's access and use requirements.

 ---
 base_model: google/gemma-4-E2B-it-qat-q4_0-unquantized
 library_name: llama.cpp
+license: apache-2.0
 tags:
 - gemma4
 - gguf
+- llama.cpp
 - qat-derived
 - snowfox
 ---
+# Gemma 4 E2B SnowFox GGUF
+This is the GGUF distribution of `gemma4-e2b-Snowfox`: a language-only SnowFox
+LoRA merge based on Google's Gemma 4 E2B instruction QAT-derived checkpoint.
+The image and audio towers were frozen during SnowFox fine-tuning. The post-merge
+weights were exported to the four GGUF quantizations below; they were not
+separately re-QAT calibrated.
+## Download exactly one LLM quantization, plus the shared projector
+| LLM file | Size | SHA-256 |
+| --- | ---: | --- |
+| `gemma4-e2b-Snowfox-Q4_0.gguf` | 3,349,516,000 bytes | `105e63469457449dcdc147785c4317b7891a9aafaec2e40bdcbc1b16c6fd8f3b` |
+| `gemma4-e2b-Snowfox-Q4_K_M.gguf` | 3,416,120,032 bytes | `f4a38a2a95cefbf91659c355aa80410c9e0b307072c3dc7deca842b2d76d613a` |
+| `gemma4-e2b-Snowfox-Q6_K.gguf` | 3,829,835,488 bytes | `e7d2a0c8cc168db9652799d3bfd8198563dc821bae424b4b8ed1a767bbc51537` |
+| `gemma4-e2b-Snowfox-Q8_0.gguf` | 4,947,414,752 bytes | `44e7f9dd879f0ae2b18f57f78819e147783b4795738b3d3321f6e20542fa6a68` |
+For a multimodal-capable llama.cpp-family runtime, also download the shared
+projector:
+| Projector file | Size | SHA-256 |
+| --- | ---: | --- |
+| `gemma4-e2b-Snowfox-BF16-mmproj.gguf` | 986,833,408 bytes | `cbc8946d8e6a85714c7f4868302ccd7bf716301b47a5efefe5850eb5348b0a3e` |
+`Q4_K_M` is the usual balanced default. Use a recent runtime that supports
+Gemma 4 E2B and its multimodal projector. This repository intentionally holds
+all four verified downloadable quantizations in one GGUF-only package.
+Example download:
+```bash
+hf download MichaelAnthony/gemma4-e2b-Snowfox-GGUF \
+  gemma4-e2b-Snowfox-Q4_K_M.gguf \
+  gemma4-e2b-Snowfox-BF16-mmproj.gguf \
+  --local-dir ./gemma4-e2b-Snowfox-GGUF
+```
 ## Lineage and validation
+- Base: [`google/gemma-4-E2B-it-qat-q4_0-unquantized`](https://huggingface.co/google/gemma-4-E2B-it-qat-q4_0-unquantized)
 - Pinned base revision: `6befbaca7398925921802abd1f277b495b78b738`
+- Release claim: QAT-derived; the SnowFox post-LoRA weights were **not** newly QAT-calibrated.
+- Internal release gates: required-field accuracy `0.9984962406015038`, strict JSON validity `1.0`, native tool-call validity `1.0`, and contract success `0.9936305732484076`.
+Those figures are the original held-out SnowFox validation results, not an
+inference benchmark for every GGUF/runtime combination.
 ## License
+Gemma 4 is Apache-2.0. This derivative package uses the Apache-2.0 license
+declared by the pinned base model. See [`LICENSE`](LICENSE) and [`NOTICE.md`](NOTICE.md)
+for the lineage and modification notice.