Instructions to use CosmicRaisins/GLM-5.2-AWQ-INT4-15pct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use CosmicRaisins/GLM-5.2-AWQ-INT4-15pct with Transformers:

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

pipe = pipeline("text-generation", model="CosmicRaisins/GLM-5.2-AWQ-INT4-15pct")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("CosmicRaisins/GLM-5.2-AWQ-INT4-15pct")
model = AutoModelForCausalLM.from_pretrained("CosmicRaisins/GLM-5.2-AWQ-INT4-15pct")
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 CosmicRaisins/GLM-5.2-AWQ-INT4-15pct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "CosmicRaisins/GLM-5.2-AWQ-INT4-15pct"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "CosmicRaisins/GLM-5.2-AWQ-INT4-15pct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/CosmicRaisins/GLM-5.2-AWQ-INT4-15pct

SGLang

How to use CosmicRaisins/GLM-5.2-AWQ-INT4-15pct 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 "CosmicRaisins/GLM-5.2-AWQ-INT4-15pct" \
    --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": "CosmicRaisins/GLM-5.2-AWQ-INT4-15pct",
		"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 "CosmicRaisins/GLM-5.2-AWQ-INT4-15pct" \
        --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": "CosmicRaisins/GLM-5.2-AWQ-INT4-15pct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use CosmicRaisins/GLM-5.2-AWQ-INT4-15pct with Docker Model Runner:
```
docker model run hf.co/CosmicRaisins/GLM-5.2-AWQ-INT4-15pct
```

GLM-5.2-AWQ-INT4-15pct

15% expert-pruned cyankiwi/GLM-5.2-AWQ-INT4 (256→218 experts/layer), to free KV-cache headroom for 256k context on memory-constrained clusters (built for 4× GB10 / DGX Spark). The prune is data-free. ~636B total / ~40B active, AWQ compressed-tensors W4A16, ~378 GB.

Quality is coherence-checked, not benchmarked. Evaluate before production use; for guaranteed quality use the unpruned cyankiwi/GLM-5.2-AWQ-INT4 or zai-org/GLM-5.2.

Method (data-free)

GLM/DeepSeek routers carry a learned e_score_correction_bias per expert: a high bias means the router had to boost that expert to select it (least favored). awq_surgery.py drops the 38 highest-bias experts per layer, keeps the 218 lowest, re-indexes survivors, and row-slices the router. No calibration data, no forward passes. Both num_experts and n_routed_experts become 218. This is not REAP (REAP needs calibration data and was infeasible on this hardware).

Serving

A 256k 4-bit MoE for multi-node TP, not a single-GPU model. Needs sm_121 Triton sparse-MLA kernels (native _flashmla_C is Hopper-only). Stack and bootstrap: github.com/CosmicRaisins/glm-5.2-gb10. Runtime: TP=4, --kv-cache-dtype fp8_ds_mla, --reasoning-parser glm45 --tool-call-parser glm47 --enable-auto-tool-choice, cudagraph FULL, gpu-memory-utilization 0.93.

For MTP speculative decode, pair it with the matching draft built for this target: CosmicRaisins/GLM-5.2-MTP-INT4-aligned.

License

MIT, inherited. Retain upstream notices on redistribution. GLM-5.2 © Z.ai (MIT); GLM-5.2-AWQ-INT4 © cyankiwi (MIT). The data-free prune is the only modification. Not affiliated with Z.ai or cyankiwi.

Downloads last month: 226

Safetensors

Model size

675B params

Tensor type

BF16

I32

I64

F32

Model tree for CosmicRaisins/GLM-5.2-AWQ-INT4-15pct

Base model

zai-org/GLM-5.2

Quantized

cyankiwi/GLM-5.2-AWQ-INT4

Quantized

(2)

this model

Quantizations

1 model