Instructions to use zpjbtdjm/CNPE-7B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use zpjbtdjm/CNPE-7B with Transformers:

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

pipe = pipeline("text-generation", model="zpjbtdjm/CNPE-7B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("zpjbtdjm/CNPE-7B")
model = AutoModelForCausalLM.from_pretrained("zpjbtdjm/CNPE-7B", 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=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use zpjbtdjm/CNPE-7B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "zpjbtdjm/CNPE-7B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "zpjbtdjm/CNPE-7B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/zpjbtdjm/CNPE-7B

SGLang

How to use zpjbtdjm/CNPE-7B 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 "zpjbtdjm/CNPE-7B" \
    --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": "zpjbtdjm/CNPE-7B",
		"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 "zpjbtdjm/CNPE-7B" \
        --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": "zpjbtdjm/CNPE-7B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use zpjbtdjm/CNPE-7B with Docker Model Runner:
```
docker model run hf.co/zpjbtdjm/CNPE-7B
```

CNPE-7B

CNPE-7B is the model implementation of CNPE (Comparison-Native framework for Paper Evaluation), introduced in From Isolated Scoring to Collaborative Ranking: A Comparison-Native Framework for LLM-Based Paper Evaluation. It is built on Qwen2.5-7B-Instruct and evaluates scientific papers by comparing them against each other rather than assigning absolute scores in isolation.

Training Data

The model was trained on ICLR 2025 submissions collected from OpenReview, following the train-test split of DeepReview to ensure fair comparison, with the average human reviewer score used as ground truth. Qwen2.5-7B-Instruct was chosen as the base model.

Results

Performance on the ICLR 2025 test set, covering acceptance-decision metrics (Accuracy, F1, AUC, Cohen's κ) and ranking metrics (Spearman's ρ, pairwise accuracy, MAP@20, NDCG@20):

Method	Acc.	F1	AUC	κ	Spearman ρ	Pair. Acc.	MAP@20	NDCG@20
DeepReview-14B	0.6845	0.6254	0.6624	0.2510	0.4014	0.6419	0.1478	0.7204
CNPE-7B	0.7192	0.6732	0.7408	0.3464	0.4091	0.6448	0.7076	0.8153

CNPE-7B achieves the best result on every metric. With only 7B parameters, it obtains an average relative improvement of 21.8% over the strong baseline DeepReview-14B.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "zpjbtdjm/CNPE-7B", torch_dtype="bfloat16", device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("zpjbtdjm/CNPE-7B")

For the full pipeline (pair sampling, training, and preference aggregation), see the official code repository: ECNU-Text-Computing/ComparisonReview.

Responsible Use

CNPE-7B is a research system for studying automated paper evaluation. Its judgments are advisory evidence and should not replace human peer review.

Citation

@inproceedings{zheng2026isolated,
  title={From Isolated Scoring to Collaborative Ranking: A Comparison-Native Framework for LLM-Based Paper Evaluation},
  author={Zheng, Pujun and Yao, Jiacheng and Zheng, Jinquan and Gu, Chenyang and He, Guoxiu and Liu, Jiawei and Huang, Yong and Guo, Tianrui and Lu, Wei},
  booktitle={Findings of the Association for Computational Linguistics: ACL 2026},
  pages={23863--23883},
  year={2026}
}

Downloads last month: 21

Safetensors

Model size

8B params

Tensor type

BF16

Model tree for zpjbtdjm/CNPE-7B

Base model

Qwen/Qwen2.5-7B

Finetuned

Qwen/Qwen2.5-7B-Instruct

Finetuned

(3219)

this model