Student-Support-Qwen2.5-7B-Instruct

Qwen2.5-7B-Instruct Fine-Tuned for Mental Health Counseling

Model Overview

This is a fine-tuned version of unsloth/Qwen2.5-7B-Instruct-bnb-4bit, optimized for mental health counseling conversations. The model is trained to provide supportive, empathetic, and informative responses for individuals seeking mental health support.

Dataset

The model is fine-tuned on a combination of:

• Amod/mental_health_counseling_conversations (cleaned version: arafatanam/Mental-Health-Counseling) - 2752 rows
• chillies/student-mental-health-counseling-vn (translated version: arafatanam/Student-Mental-Health-Counseling-10K) - 7500 rows
• Total dataset size: 10,252 rows

Training Details

• Hardware: Kaggle Notebooks (GPU T4 x2)
• Fine-tuning framework: Unsloth with LoRA
• Training settings:
  • max_seq_length = 512
  • batch_size = 8
  • gradient_accumulation_steps = 4
  • num_train_epochs = 2
  • learning_rate = 5e-5
  • optimizer = adamw_8bit
  • lr_scheduler = cosine

Training Results

• Final training loss: 1.3034
• Total steps: 640
• Trainable parameters: 0.58% of the model`
• Validation loss: 1.245
• Evaluation metric (perplexity): 3.42

Usage

This model can be used for:

• Providing mental health counseling support
• Generating empathetic responses to mental health queries
• Assisting in chatbot development for mental health applications