Instructions to use intel-ai/GLM-5.3-Flash-MXFP4-AutoRound with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use intel-ai/GLM-5.3-Flash-MXFP4-AutoRound with Transformers:

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

pipe = pipeline("text-generation", model="intel-ai/GLM-5.3-Flash-MXFP4-AutoRound")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("intel-ai/GLM-5.3-Flash-MXFP4-AutoRound")
model = AutoModelForMultimodalLM.from_pretrained("intel-ai/GLM-5.3-Flash-MXFP4-AutoRound", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.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=256)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use intel-ai/GLM-5.3-Flash-MXFP4-AutoRound with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "intel-ai/GLM-5.3-Flash-MXFP4-AutoRound"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "intel-ai/GLM-5.3-Flash-MXFP4-AutoRound",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/intel-ai/GLM-5.3-Flash-MXFP4-AutoRound

SGLang

How to use intel-ai/GLM-5.3-Flash-MXFP4-AutoRound 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 "intel-ai/GLM-5.3-Flash-MXFP4-AutoRound" \
    --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": "intel-ai/GLM-5.3-Flash-MXFP4-AutoRound",
		"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 "intel-ai/GLM-5.3-Flash-MXFP4-AutoRound" \
        --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": "intel-ai/GLM-5.3-Flash-MXFP4-AutoRound",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use intel-ai/GLM-5.3-Flash-MXFP4-AutoRound with Docker Model Runner:
```
docker model run hf.co/intel-ai/GLM-5.3-Flash-MXFP4-AutoRound
```

Configuration Parsing Warning:In UNKNOWN_FILENAME: "quantization_config.config_groups.group_0.format" must be a string

Model Details

This model is an MXFP4 (W4A4, group_size 32) model of zai-org/GLM-5.3-Flash generated by intel/auto-round using RTN. Please follow the license of the original model.

Quantized (MXFP4): routed expert projections (all sparse MoE layers) and the dense-MLP projections of the first three layers.
Kept in BF16: the KDA linear-attention and sparse-MLA projections, the DSA indexer, shared-expert projections, the MoE router, the MTP layer, the vision tower, the embedding and lm_head, normalization and hyper-connection parameters.

How to Use

Environment

## Using the latest vLLM version should work fine.
uv pip install "vllm @ git+https://github.com/vllm-project/vllm@46d2b23ac5047a813ebb082122166e4ae09b5f39"

vLLM Usage

vllm serve intel-ai/GLM-5.3-Flash-MXFP4-AutoRound \
    --kv-cache-dtype fp8 \
    --dtype bfloat16

Evaluation

Benchmark	BF16	MXFP4	Recovery
PIQA	82.92	82.75	99.79%
HellaSwag	69.03	68.25	98.87%
MMLU	86.66	85.88	99.10%
GSM8K	97.35	97.50	100.15%
Average	83.99	83.59	99.53%

Generate the Model

auto-round \
    --model_name zai-org/GLM-5.3-Flash-BF16 \
    --model_free \
    --scheme MXFP4 \
    --ignore_layers "lm_head,embed_tokens,visual,indexer,weights_proj,self_attn,mlp.shared_experts,layers.45,eh_proj,enorm,hnorm,shared_head,norm,A_log,dt_bias,conv1d" \
    --layer_config '{"model.language_model.layers.0.mlp.gate_proj":{"bits":4,"data_type":"mx_fp"},"model.language_model.layers.0.mlp.up_proj":{"bits":4,"data_type":"mx_fp"},"model.language_model.layers.0.mlp.down_proj":{"bits":4,"data_type":"mx_fp"}}' \
    --format llm_compressor \
    --output_dir GLM-5.3-Flash-BF16-MXFP4-Mixed-CT-RTN-AutoRound

Ethical Considerations and Limitations

The model inherits the capabilities, biases, and safety behavior of the base checkpoint. Quantization does not add alignment, and reduced numerical precision can change outputs on edge cases. No additional training data was used.

Caveats and Recommendations

Serve with --kv-cache-dtype fp8, matching the reference checkpoint's FP8 KV-cache setting.
The exported quantization config targets the module names used by vLLM's GLM implementation (language_model.model.*, visual.*, language_model.lm_head, plus RoutedExperts for the fused MoE); serving through a runtime that expects the raw Hugging Face module names requires retargeting the config.
Evaluation covers four academic text benchmarks only; long-context, agentic, thinking-mode and multimodal behavior were not measured.

Disclaimer

This model is a derivative of GLM-5.3-Flash and is provided under the same MIT license. It is not an official Z.ai or NVIDIA release.

Cite

@software{autoround,
  title  = {AutoRound: Advanced Weight-Only Quantization for LLMs},
  author = {Intel},
  url    = {https://github.com/intel/auto-round}
}

Signed-off-by: Kaokao, Lv kaokao.lv@intel.com

Downloads last month: 274

Safetensors

Model size

321B params

Tensor type

F32

BF16

Model tree for intel-ai/GLM-5.3-Flash-MXFP4-AutoRound

Base model

zai-org/GLM-5.3-Flash-BF16

Quantized

(38)

this model