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=0.1, cosine scheduler
  • Optimizer: adamw_8bit
  • Early Stopping: Patience=3, threshold=0.01
• Evaluation Metrics (Exact Matches):
  • CEFR Classifier Accuracy: 0.250
  • Precision (Macro): 0.130
  • Recall (Macro): 0.250
  • F1-Score (Macro): 0.153
• Evaluation Metrics (Within ±1 Level):
  • CEFR Classifier Accuracy: 0.733
  • Precision (Macro): 0.701
  • Recall (Macro): 0.733
  • F1-Score (Macro): 0.687
• Other Metrics:
  • Perplexity: 14.218
  • Diversity (Unique Sentences): 0.933
  • Inference Time (ms): 2208.386
  • Model Size (GB): 4.8
  • Robustness (F1): 0.145
• Confusion Matrix (Exact Matches):
  • CSV: confusion_matrix.csv
  • Image: confusion_matrix.png
• Per-Class Confusion Metrics (Exact Matches):
  • A1: TP=0, FP=2, FN=10, TN=48
  • A2: TP=0, FP=0, FN=10, TN=50
  • B1: TP=10, FP=29, FN=0, TN=21
  • B2: TP=2, FP=7, FN=8, TN=43
  • C1: TP=3, FP=7, FN=7, TN=43
  • C2: TP=0, FP=0, FN=10, TN=50
• Usage:
  python

  1from transformers import AutoModelForCausalLM, AutoTokenizer 2 3model = AutoModelForCausalLM.from_pretrained("Mr-FineTuner/Test___01_withNewEval") 4tokenizer = AutoTokenizer.from_pretrained("Mr-FineTuner/Test___01_withNewEval") 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))