You can deploy this model and start earning money today!

WangDong2017/
GrammarSeeker-SFT-Qwen2.5-7B

Name: GrammarSeeker-SFT-Qwen2.5-7B
Availability: InStock

safetensors

qwen2

text-classification

Qwen/Qwen2.5-7B-Instruct

finetune

apache-2.0

Views

No views yet

GrammarSeeker-SFT-Qwen2.5-7B

A fine-tuned Qwen2.5-7B-Instruct model specifically designed for grammatical project parsing systems.

🔗 Repository Links

📁 Source Code: GitHub Repository - Contains testing code and development scripts
🤗 Model Hub: Hugging Face Model - Hosts the complete fine-tuned model

🎉 Latest Update (2025-08-11)

✅ Model Successfully Deployed to Hugging Face!

The fine-tuned model is now available for direct use without any additional steps.

📋 Model Information

Base Model: Qwen/Qwen2.5-7B-Instruct
Fine-tuned Model: WangDong2017/GrammarSeeker-SFT-Qwen2.5-7B
Fine-tuning Method: LoRA (Low-Rank Adaptation) during training, now provided as complete model
Task: Binary classification for grammatical project annotation (T/F output)
Performance:
- F1 Score: 0.9797 (97.97%)
- Positive Accuracy: 0.9640 (96.40%)
- Negative Accuracy: 0.9960 (99.60%)
- Test Samples: 1000
- Test Date: 2025-08-11
- Tested Performance: 16 annotations/s (Test completed in ~1 minute on RTX 4090)

🎯 Use Case

This model serves as the core component of a grammatical project parsing system. It is designed to:

Receive structured prompts (as shown in GM-TestData.csv)
Output binary decisions (T/F) for grammatical annotation
Enable automated grammar project marking based on model predictions

🔧 Usage

Installation

pip install transformers peft torch

Testing Performance

bash

# Test the model from Hugging Face
python test_hf_model.py

Latest Test Results (2025-08-11):

✅ Model successfully loaded from HF repository
✅ All 1000 test samples processed
✅ F1 Score: 0.9797 (97.97%)
✅ Test completed in ~1 minute on RTX 4090

Loading the Model

python

from transformers import AutoTokenizer, AutoModelForCausalLM

# Load the complete fine-tuned model directly from HF
model = AutoModelForCausalLM.from_pretrained(
    "WangDong2017/GrammarSeeker-SFT-Qwen2.5-7B",
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True
)

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained("WangDong2017/GrammarSeeker-SFT-Qwen2.5-7B")

🏭 Production Environment Usage

Recommended workflow:

Pre-filtering: Use regular expressions for coarse screening
String matching: Trigger prompt generation based on string matching
Model inference: Send generated prompt to this model
Output processing: Model outputs T/F
Automatic annotation: Generate grammatical project markers based on T/F output

📊 Dataset

GM-TestData.csv: 1000 test samples with prompts and expected answers
Format: prompt1, prompt2, answer (T/F)
Test Results: Successfully validated with 97.97% F1 score

🚀 Deployment & Integration

Hugging Face Integration

Model Hub: WangDong2017/GrammarSeeker-SFT-Qwen2.5-7B
Direct Loading: Available for immediate use
API Access: Can be deployed through HF Inference API

📝 Citation

bibtex

@misc{wang2025CPGEVALMultitieredBenchmark,
  title = {{{CPG-EVAL}}: A Multi-Tiered Benchmark for Evaluating the Chinese Pedagogical Grammar Competence of Large Language Models},
  author = {Wang, Dong},
  year = {2025},
  publisher = {arXiv},
  doi = {10.48550/ARXIV.2504.13261}
}

Note: This model has been successfully tested and deployed. For production use, please ensure proper testing and validation in your specific use case.