judywq/llama-ft-feedback_score

📝 LLaMA 3.3 70B – Fine-tuned for Argumentative Essay Scoring

This model is a fully fine-tuned version of LLaMA 3.3 70B for the task of automated scoring of argumentative essays. It predicts a holistic score from 0 to 5 (in 0.5 increments) based on the prompt and student response, using TOEFL-style rubrics.

---

📌 Task Overview

• Task: Holistic essay scoring
• Input: Essay prompt + student essay
• Output: A numerical score (e.g., 3.5)
• Score scale: 0–5, in 0.5 increments
• Rubric: Adapted from TOEFL iBT Independent Writing scoring guidelines
• Training data: TOEFL Public Dataset (requires permission from ETS)

---

🔧 Training Configuration

Hyperparameter	Value
Base model	LLaMA 3.3 70B
Fine-tuning method	Full-parameter
Epochs	5
Batch size	32
Learning rate	3e-6
Optimizer	AdamW
Scheduler	Cosine
Training hardware	2×A100

---

📊 Evaluation

• Loss curve is available in training_loss.png
• Additional metrics (e.g., QWK, RMSE, exact match) can be provided upon request

---

🚀 Usage Example

from transformers import AutoTokenizer, AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained("your-org/llama3-70b-argscore")
tokenizer = AutoTokenizer.from_pretrained("your-org/llama3-70b-argscore")

prompt = "Essay prompt: Do you agree or disagree... Essay: ..."
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

📄 Citation

@misc{llama3-argscore, title={LLaMA 3.3 70B Fine-tuned for Argumentative Essay Scoring}, author={Wang, Q, Labib, A, Yuan, Z}, year={2025}, note={\url{https://huggingface.co/judywq/llama-ft-feedback_score}}, }