Instructions to use mlx-community/GLM-5.2-DQ4plus-q8 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use mlx-community/GLM-5.2-DQ4plus-q8 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("mlx-community/GLM-5.2-DQ4plus-q8")

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)

Transformers

How to use mlx-community/GLM-5.2-DQ4plus-q8 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="mlx-community/GLM-5.2-DQ4plus-q8")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("mlx-community/GLM-5.2-DQ4plus-q8")
model = AutoModelForCausalLM.from_pretrained("mlx-community/GLM-5.2-DQ4plus-q8")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

vLLM

How to use mlx-community/GLM-5.2-DQ4plus-q8 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "mlx-community/GLM-5.2-DQ4plus-q8"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mlx-community/GLM-5.2-DQ4plus-q8",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/mlx-community/GLM-5.2-DQ4plus-q8

SGLang

How to use mlx-community/GLM-5.2-DQ4plus-q8 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "mlx-community/GLM-5.2-DQ4plus-q8" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mlx-community/GLM-5.2-DQ4plus-q8",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "mlx-community/GLM-5.2-DQ4plus-q8" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mlx-community/GLM-5.2-DQ4plus-q8",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

How to use mlx-community/GLM-5.2-DQ4plus-q8 with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "mlx-community/GLM-5.2-DQ4plus-q8"

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": "mlx-community/GLM-5.2-DQ4plus-q8"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use mlx-community/GLM-5.2-DQ4plus-q8 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 "mlx-community/GLM-5.2-DQ4plus-q8"

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 mlx-community/GLM-5.2-DQ4plus-q8

Run Hermes

hermes

OpenClaw new

How to use mlx-community/GLM-5.2-DQ4plus-q8 with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "mlx-community/GLM-5.2-DQ4plus-q8"

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 "mlx-community/GLM-5.2-DQ4plus-q8" \
  --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"

MLX LM

How to use mlx-community/GLM-5.2-DQ4plus-q8 with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "mlx-community/GLM-5.2-DQ4plus-q8"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "mlx-community/GLM-5.2-DQ4plus-q8"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "mlx-community/GLM-5.2-DQ4plus-q8",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Docker Model Runner
How to use mlx-community/GLM-5.2-DQ4plus-q8 with Docker Model Runner:
```
docker model run hf.co/mlx-community/GLM-5.2-DQ4plus-q8
```

config declares num_nextn_predict_layers=1 but MTP draft tensors were dropped (disables speculative decoding)

by WillMitchell - opened 7 days ago

Discussion

WillMitchell

7 days ago

While benchmarking this quant on a 512GB M3 Ultra we found an inconsistency worth flagging:

config.json declares "num_nextn_predict_layers": 1 (the GLM-5.2 MTP draft layer)
but model.safetensors.index.json contains zero tensors for that layer (no model.layers.78.*, no nextn/mtp/eh_proj/shared_head keys)

So the MTP block was dropped during conversion while the config still advertises it. Consequences:

Runtimes that implement GLM-5.2 Lightning MTP / native speculative decoding (e.g. oMLX 0.5.0) cannot use it with this quant, silently, since the config says the layer exists.
Loaders that trust the config may probe for weights that are not there.

For context, we surveyed all 29 GLM-5.2 MLX conversions on the Hub: 25 drop the MTP tensors the same way (this repo, GLM-5.2-4bit, GLM-5.2-mxfp4, the pipenetwork and inferencerlabs conversions); only 4 keep them (Jundot/GLM-5.2-oQ4e-mtp, the avlp12 Alis-MLX-Dynamic series). The MTP layer is a full transformer block (~791 tensors incl. 256 experts), so keeping it costs real size, dropping it is a legitimate choice, but then the config should say so.

Suggested fix, either:

set "num_nextn_predict_layers": 0 in config.json (as Jundot/GLM-5.2-oQ4 does explicitly for its non-MTP variant), or
re-convert including the MTP block for those who want speculative decoding.

Otherwise this quant measured very well for us: correctness parity with unsloth UD-Q4_K_M GGUF across a 29-sample agentic eval suite (every completed sample passed), and ~845 tok/s prefill on oMLX 0.5.0 with the native DSA kernels. Happy to share more data if useful.

bibproj

MLX Community org 7 days ago

@WillMitchell Hi Will. Thank you for this valuable information and the feedback.

I've set "num_nextn_predict_layers": 0 in config.json for now. This is the quickest way to fix the problem.

The best is for me to re-convert and include the MTP block to allow for speculative decoding.

Happy to share more data if useful.

Always keen to learn, so please share more! 😊

@pcuenq Hi Pedro. Good feedback here for improving our MLX quants for GLM-5.2 and PR #1410

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment