Instructions to use long10024070/GLM-5.2-HYBRID-FP8-MXFP4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use long10024070/GLM-5.2-HYBRID-FP8-MXFP4 with Transformers:

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

pipe = pipeline("text-generation", model="long10024070/GLM-5.2-HYBRID-FP8-MXFP4")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("long10024070/GLM-5.2-HYBRID-FP8-MXFP4")
model = AutoModelForCausalLM.from_pretrained("long10024070/GLM-5.2-HYBRID-FP8-MXFP4", device_map="auto")
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

vLLM

How to use long10024070/GLM-5.2-HYBRID-FP8-MXFP4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "long10024070/GLM-5.2-HYBRID-FP8-MXFP4"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "long10024070/GLM-5.2-HYBRID-FP8-MXFP4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/long10024070/GLM-5.2-HYBRID-FP8-MXFP4

SGLang

How to use long10024070/GLM-5.2-HYBRID-FP8-MXFP4 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 "long10024070/GLM-5.2-HYBRID-FP8-MXFP4" \
    --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": "long10024070/GLM-5.2-HYBRID-FP8-MXFP4",
		"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 "long10024070/GLM-5.2-HYBRID-FP8-MXFP4" \
        --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": "long10024070/GLM-5.2-HYBRID-FP8-MXFP4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use long10024070/GLM-5.2-HYBRID-FP8-MXFP4 with Docker Model Runner:
```
docker model run hf.co/long10024070/GLM-5.2-HYBRID-FP8-MXFP4
```

GLM-5.2 Hybrid FP8/MXFP4

This deterministic checkpoint uses block-FP8 non-MoE weights and the block-FP8 MTP layer 78 from zai-org/GLM-5.2-FP8. Routed and shared MoE expert tensors for layers 3 through 77 are copied byte-for-byte from amd/GLM-5.2-MXFP4. No tensor is dequantized or requantized during assembly.

The checkpoint declares hybrid-fp8-mxfp4 and requires the matching conditional SGLang loader patch shipped with its reproducibility assets. Original FP8 and Quark checkpoints do not use that patch path.

Downloads last month: 24

Safetensors

Model size

390B params

Tensor type

BF16

F8_E4M3

F32

Model tree for long10024070/GLM-5.2-HYBRID-FP8-MXFP4

Base model

zai-org/GLM-5.2

Quantized

amd/GLM-5.2-MXFP4

Quantized

(1)

this model