Instructions to use dancinlab/supergemma4-e4b-abliterated-MLX-4bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use dancinlab/supergemma4-e4b-abliterated-MLX-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("dancinlab/supergemma4-e4b-abliterated-MLX-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
LM Studio

Pi new

How to use dancinlab/supergemma4-e4b-abliterated-MLX-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 "dancinlab/supergemma4-e4b-abliterated-MLX-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": "dancinlab/supergemma4-e4b-abliterated-MLX-4bit"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use dancinlab/supergemma4-e4b-abliterated-MLX-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 "dancinlab/supergemma4-e4b-abliterated-MLX-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 dancinlab/supergemma4-e4b-abliterated-MLX-4bit

Run Hermes

hermes

MLX LM

How to use dancinlab/supergemma4-e4b-abliterated-MLX-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 "dancinlab/supergemma4-e4b-abliterated-MLX-4bit"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "dancinlab/supergemma4-e4b-abliterated-MLX-4bit"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "dancinlab/supergemma4-e4b-abliterated-MLX-4bit",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

dancinlife commited on 6 days ago

Commit

2ffc029

verified ·

1 Parent(s): 0200469

SEO: English-only prompt samples

Browse files

Files changed (1) hide show

README.md +3 -3

README.md CHANGED Viewed

@@ -40,7 +40,7 @@ pip install -U mlx-lm   # needs mlx-lm >= 0.31.3 (native gemma4 arch)
 # 4-bit — recommended for 16 GB / 24 GB Macs
 mlx_lm.generate --model dancinlab/supergemma4-e4b-abliterated-MLX-4bit \
-  --prompt "안녕? 너는 누구야?" --max-tokens 60
 # interactive chat
 mlx_lm.chat --model dancinlab/supergemma4-e4b-abliterated-MLX-4bit
@@ -54,8 +54,8 @@ mlx_lm.chat --model dancinlab/supergemma4-e4b-abliterated-MLX-4bit
 | `-MLX-8bit`     | 7.4 GB | 9.1 GB | ~6  | 32 GB+ Mac, higher fidelity |
 | `-MLX-bf16`     | 14 GB  | 8.6 GB | ~3  | reference, full precision |
-Verified: both `안녕? 너는 누구야?` (coherent Korean) and `2+2=` (→ 4) on
-stock `mlx-lm==0.31.3`. **Text-only** — the upstream abliterated safetensors
 contain no vision/audio tower weights, so multimodal MLX is upstream-blocked,
 not a tooling limitation.

 # 4-bit — recommended for 16 GB / 24 GB Macs
 mlx_lm.generate --model dancinlab/supergemma4-e4b-abliterated-MLX-4bit \
+  --prompt "Who are you?" --max-tokens 60
 # interactive chat
 mlx_lm.chat --model dancinlab/supergemma4-e4b-abliterated-MLX-4bit
 | `-MLX-8bit`     | 7.4 GB | 9.1 GB | ~6  | 32 GB+ Mac, higher fidelity |
 | `-MLX-bf16`     | 14 GB  | 8.6 GB | ~3  | reference, full precision |
+Verified on stock `mlx-lm==0.31.3`: coherent multilingual output (English +
+Korean) and correct arithmetic (`2+2=` → 4). **Text-only** — the upstream abliterated safetensors
 contain no vision/audio tower weights, so multimodal MLX is upstream-blocked,
 not a tooling limitation.