Instructions to use cyankiwi/GLM-5.3-AWQ-INT4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use cyankiwi/GLM-5.3-AWQ-INT4 with Transformers:

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

pipe = pipeline("text-generation", model="cyankiwi/GLM-5.3-AWQ-INT4")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("cyankiwi/GLM-5.3-AWQ-INT4")
model = AutoModelForCausalLM.from_pretrained("cyankiwi/GLM-5.3-AWQ-INT4", 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 cyankiwi/GLM-5.3-AWQ-INT4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "cyankiwi/GLM-5.3-AWQ-INT4"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "cyankiwi/GLM-5.3-AWQ-INT4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/cyankiwi/GLM-5.3-AWQ-INT4

SGLang

How to use cyankiwi/GLM-5.3-AWQ-INT4 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 "cyankiwi/GLM-5.3-AWQ-INT4" \
    --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": "cyankiwi/GLM-5.3-AWQ-INT4",
		"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 "cyankiwi/GLM-5.3-AWQ-INT4" \
        --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": "cyankiwi/GLM-5.3-AWQ-INT4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use cyankiwi/GLM-5.3-AWQ-INT4 with Docker Model Runner:
```
docker model run hf.co/cyankiwi/GLM-5.3-AWQ-INT4
```

Larger INT4 footprint vs 5.2; is it possible to match 5.2's quantization coverage

by themoatai - opened 19 days ago

Discussion

themoatai

19 days ago

Thanks for the GLM-5.3-AWQ-INT4 release! We run your GLM-5.2-AWQ-INT4 in production and hoped to drop 5.3 in behind the same stack.

On disk, 5.3 is ~48 GB larger than 5.2, and it looks like a few layers are kept in BF16 rather than INT4 (we're seeing this around layers 3 and 77, plus the layer-78 MTP head). On our 128 GB nodes that extra mass leaves no KV headroom, so we can't reach a usable context length that 5.2 handled.

Is the BF16 retention intentional? Would you be able to produce a 5.3 quant at the same coverage/footprint as 5.2 (i.e. those layers quantized to INT4)?

Thanks again for the work on these.

ctcanbol

10 days ago

Is the BF16 retention intentional? Would you be able to produce a 5.3 quant at the same coverage/footprint as 5.2 (i.e. those layers quantized to INT4)?

I'm also curious about that...

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