Fine-Tuned LLaMA-3-8B CEFR Model

• Base Model: unsloth/llama-3-8b-instruct-bnb-4bit
• Fine-Tuning: LoRA with SMOTE-balanced dataset
• Training Details:
  • Dataset: CEFR-level sentences with SMOTE and undersampling for balance
  • LoRA Parameters: r=32, lora_alpha=32, lora_dropout=0.5
  • Training Args: learning_rate=2e-5, batch_size=8, epochs=0.01, cosine scheduler
  • Optimizer: adamw_8bit
  • Early Stopping: Patience=3, threshold=0.01
• Evaluation Metrics (Exact Matches):
  • CEFR Classifier Accuracy: 0.233
  • Precision (Macro): 0.212
  • Recall (Macro): 0.233
  • F1-Score (Macro): 0.202
• Evaluation Metrics (Within ±1 Level):
  • CEFR Classifier Accuracy: 0.667
  • Precision (Macro): 0.767
  • Recall (Macro): 0.667
  • F1-Score (Macro): 0.633
• Other Metrics:
  • Perplexity: 3.080
  • Diversity (Unique Sentences): 1.000
  • Inference Time (ms): 6948.521
  • Model Size (GB): 4.8
  • Robustness (F1): 0.192
• Confusion Matrix (Exact Matches):
  • CSV: confusion_matrix_exact.csv
  • Image: confusion_matrix_exact.png
• Confusion Matrix (Within ±1 Level):
  • CSV: confusion_matrix_within1.csv
  • Image: confusion_matrix_within1.png
• Per-Class Confusion Metrics (Exact Matches):
  • A1: TP=7, FP=10, FN=3, TN=40
  • A2: TP=3, FP=21, FN=7, TN=29
  • B1: TP=0, FP=7, FN=10, TN=43
  • B2: TP=2, FP=4, FN=8, TN=46
  • C1: TP=2, FP=3, FN=8, TN=47
  • C2: TP=0, FP=1, FN=10, TN=49
• Per-Class Confusion Metrics (Within ±1 Level):
  • A1: TP=10, FP=3, FN=0, TN=47
  • A2: TP=10, FP=10, FN=0, TN=40
  • B1: TP=8, FP=4, FN=2, TN=46
  • B2: TP=6, FP=3, FN=4, TN=47
  • C1: TP=4, FP=0, FN=6, TN=50
  • C2: TP=2, FP=0, FN=8, TN=50
• Usage:
  python

  1from transformers import AutoModelForCausalLM, AutoTokenizer 2 3model = AutoModelForCausalLM.from_pretrained("Mr-FineTuner/Test_02_llama_trainPercen_myValidator") 4tokenizer = AutoTokenizer.from_pretrained("Mr-FineTuner/Test_02_llama_trainPercen_myValidator") 5 6# Example inference 7prompt = "<|user|>Generate a CEFR B1 level sentence.<|end|>" 8inputs = tokenizer(prompt, return_tensors="pt") 9outputs = model.generate(**inputs, max_length=50) 10print(tokenizer.decode(outputs[0], skip_special_tokens=True))