Instructions to use cfontes/GLM-5.2-Fable5-R2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use cfontes/GLM-5.2-Fable5-R2 with Transformers:

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

pipe = pipeline("text-generation", model="cfontes/GLM-5.2-Fable5-R2")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForMultimodalLM

tokenizer = AutoTokenizer.from_pretrained("cfontes/GLM-5.2-Fable5-R2")
model = AutoModelForMultimodalLM.from_pretrained("cfontes/GLM-5.2-Fable5-R2")
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/GLM-5.2-Fable5-R2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "cfontes/GLM-5.2-Fable5-R2"
# 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/GLM-5.2-Fable5-R2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/cfontes/GLM-5.2-Fable5-R2

SGLang

How to use cfontes/GLM-5.2-Fable5-R2 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/GLM-5.2-Fable5-R2" \
    --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/GLM-5.2-Fable5-R2",
		"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/GLM-5.2-Fable5-R2" \
        --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/GLM-5.2-Fable5-R2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use cfontes/GLM-5.2-Fable5-R2 with Docker Model Runner:
```
docker model run hf.co/cfontes/GLM-5.2-Fable5-R2
```

GLM-5.2-Fable5-R2

LoRA fine-tune of GLM-5.2 (Test 3a ablated base) using Fable 5 distillation data.

Training Details

Parameter	Value
Base model	GLM-5.2 Test 3a (alignment-ablated)
Method	LoRA (rank=64, alpha=128)
Target layers	60-77 (top 18 of 78)
Trainable params	98M (0.013% of 753B)
Learning rate	2e-5, cosine schedule
Steps	610 (1 epoch)
Training data	4,876 Fable 5 distillation examples
Hardware	8x H200 (141GB each)

Benchmark Results (v2 harness, accurate grading)

Benchmark	Fable5-R2	Test 3a (base)	Delta
GSM8K	99%	91%	+8pp
HumanEval	84.2%	67.7%	+16.5pp
HellaSwag	80.5%	78%	+2.5pp
MMLU-Pro	84%	82%	+2pp
GPQA	8%	-	-
SimpleQA	32%	-	-
AdvBench refusal	91%	1%	+90pp (regression)
Borderline refusal	0%	0%	-

Known Issues

High refusal rate (91% AdvBench): LoRA re-activated alignment circuits ablated in Test 3a
Visible chain-of-thought contamination in responses
Not for production use without additional alignment work

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("cfontes/GLM-5.2-Fable5-R2", torch_dtype="auto", device_map="auto", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained("cfontes/GLM-5.2-Fable5-R2", trust_remote_code=True)

Downloads last month: -

Safetensors

Model size

743B params

Tensor type

F32

BF16

F8_E4M3

Model tree for cfontes/GLM-5.2-Fable5-R2

Base model

zai-org/GLM-5.2

Finetuned

cfontes/GLM-5.2-Ablated-4L

Adapter

(1)

this model