GEMMA3N Mental Health Fine-tuned Model


Base Model: unsloth/gemma-3n-e2b-it-unsloth-bnb-4bit Developed by: belal212 License: Apache-2.0 Language: English Frameworks Used: Unsloth, Hugging Face Transformers, TRL


Model Overview

This model is a fine-tuned variant of Gemma3n, specifically adapted for mental health support conversations. Training was optimized using Unsloth's accelerated fine-tuning backend, enabling faster and memory-efficient training. The model incorporates LoRA for parameter-efficient fine-tuning.


Training Data

The model was trained on a combined dataset of mental health conversations sourced from four public Hugging Face datasets:

  • ShivomH/Mental-Health-Conversations
  • YvvonM/mental_health_data
  • usham/mental-health-companion-new
  • Harshallama/mental_health_alpaca_format

All datasets were unified into a consistent instruction-based format. The combined dataset contains approximately 4.37 million samples and was saved to a single file named therapist_dataset.csv.


Fine-Tuning Configuration

The model was fine-tuned using LoRA, with adaptation applied to the language layers, attention modules, and MLP modules. Vision layers were not fine-tuned. LoRA configuration used moderate rank and dropout, with no bias term adaptation. The random seed was set for reproducibility.

Supervised Fine-Tuning (SFT) was conducted using TRL's SFTTrainer. The dataset field used for text input was formatted_text. Training used a small per-device batch size with gradient accumulation, an 8-bit AdamW optimizer, linear learning rate scheduling, and evaluation every few hundred steps. The training process saved checkpoints periodically and used a fixed seed.


Training Environment

  • CUDA-enabled GPU: NVIDIA RTX A6000
  • Frameworks: PyTorch, Hugging Face Datasets, Transformers, TRL, Unsloth

Intended Use

This model is designed for use in:

  • Mental health chatbots
  • Supportive dialogue agents
  • Psychological well-being applications
  • Research and experimentation in empathetic NLP

⚠️ Disclaimer

This model is for research and educational purposes only. It is not a substitute for professional mental health care or medical treatment. .

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support