Instructions to use cfontes/Aesop-v1-GPTQ-4bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use cfontes/Aesop-v1-GPTQ-4bit with Transformers:

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

pipe = pipeline("text-generation", model="cfontes/Aesop-v1-GPTQ-4bit")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("cfontes/Aesop-v1-GPTQ-4bit")
model = AutoModelForCausalLM.from_pretrained("cfontes/Aesop-v1-GPTQ-4bit")
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 cfontes/Aesop-v1-GPTQ-4bit with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "cfontes/Aesop-v1-GPTQ-4bit"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "cfontes/Aesop-v1-GPTQ-4bit",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/cfontes/Aesop-v1-GPTQ-4bit

SGLang

How to use cfontes/Aesop-v1-GPTQ-4bit 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 "cfontes/Aesop-v1-GPTQ-4bit" \
    --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": "cfontes/Aesop-v1-GPTQ-4bit",
		"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 "cfontes/Aesop-v1-GPTQ-4bit" \
        --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": "cfontes/Aesop-v1-GPTQ-4bit",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use cfontes/Aesop-v1-GPTQ-4bit with Docker Model Runner:
```
docker model run hf.co/cfontes/Aesop-v1-GPTQ-4bit
```

Aesop v1 — GPTQ 4-bit

A 4-bit GPTQ (compressed-tensors) quantization of cfontes/Aesop-v1, a reasoning-focused, safety-aligned adaptation of zai-org/GLM-5.2.

This variant shrinks the full-precision checkpoint from 680 GB to **383 GB** by quantizing the Mixture-of-Experts MLP weights to 4-bit while keeping attention, lm_head, and other precision-sensitive modules in higher precision.

Quantization details

Method: GPTQ (calibrated), 4-bit, group-wise
Format: compressed-tensors (pack-quantized), glm_moe_dsa architecture
Scope: MoE expert MLP weights quantized to INT4; attention (re:.*self_attn.*), lm_head, and modules_to_not_convert kept in higher precision to preserve coherence
Shards: 80 safetensor files
Serving: SGLang (compressed-tensors WNA16 4-bit path)

Note on lower bit-widths: naive round-to-nearest (RTN) MoE quantization below FP8 produced incoherent output for this architecture, and SGLang's compressed-tensors path only supports 4-bit and 8-bit WNA16. This calibrated GPTQ 4-bit build is the recommended sub-FP8 variant.

Model Details

Quantized from: cfontes/Aesop-v1
Original base: zai-org/GLM-5.2
Architecture: Mixture-of-Experts (glm_moe_dsa), 78 transformer blocks
Parameters: ~671B total / ~37B active per token

Usage (SGLang)

python -m sglang.launch_server \
  --model-path cfontes/Aesop-v1-GPTQ-4bit \
  --tp 8 \
  --trust-remote-code

Serve with a generous output budget (max_tokens ≥ 2048): the model produces long chain-of-thought before its final answer.

License

Released under the GLM license, inheriting all terms from the base model zai-org/GLM-5.2. See the license link.

Quantized from cfontes/Aesop-v1. GPTQ 4-bit, expert-focused, compressed-tensors format.

Downloads last month: -

Safetensors

Model size

749B params

Tensor type

BF16

I64

F32

I32

Model tree for cfontes/Aesop-v1-GPTQ-4bit

Base model

zai-org/GLM-5.2

Finetuned

cfontes/Aesop-v1

Quantized

(1)

this model