Model Card for Model MentalChat-16K

Model Details

This model is a fine-tuned version of Llama-3.2-1B-Instruct, optimized for empathetic and supportive conversations in the mental health domain. It was trained on the ShenLab/MentalChat16K dataset, which includes over 16,000 counseling-style Q&A examples, combining real clinical paraphrases and synthetic mental health dialogues. The model is designed to understand and respond to emotionally nuanced prompts related to stress, anxiety, relationships, and personal well-being.

Model Description

• Language(s) (NLP): English
• License: MIT
• Finetuned from model: unsloth/Llama-3.2-1B-Instruct
• Dataset: ShenLab/MentalChat16K

Uses

This model is intended for research and experimentation in AI-driven mental health support. Key use cases include:

• Mental health chatbot prototypes
• Empathy-focused dialogue agents
• Benchmarking LLMs on emotional intelligence and counseling-style prompts
• Educational or training tools in psychology or mental health communication

This model is NOT intended for clinical diagnosis, therapy, or real-time intervention. It must not replace licensed mental health professionals.

Bias, Risks, and Limitations

• Biases:

  • The real interview data is biased toward caregivers (mostly White, female, U.S.-based), which may affect the model’s cultural and demographic generalizability.
  • The synthetic dialogues are generated by GPT-3.5, which may introduce linguistic and cultural biases from its pretraining.

• Limitations:

  • The base model, Qwen2.5-0.5B-Instruct, is a small model (0.5B parameters), limiting depth of reasoning and nuanced understanding.
  • Not suitable for handling acute mental health crises or emergency counseling.
  • Responses may lack therapeutic rigor or miss subtle psychological cues.
  • May produce hallucinated or inaccurate mental health advice.

How to Get Started with the Model

Use the code below to get started with the model.

from huggingface_hub import login
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel


tokenizer = AutoTokenizer.from_pretrained("unsloth/Llama-3.2-1B-Instruct",)
base_model = AutoModelForCausalLM.from_pretrained(
    "unsloth/Llama-3.2-1B-Instruct",
    device_map={"": 0}
)

model = PeftModel.from_pretrained(base_model,"khazarai/MentalChat-16K")

system = """You are a helpful mental health counselling assistant, please answer the mental health questions based on the patient's description.
The assistant gives helpful, comprehensive, and appropriate answers to the user's questions.
"""

question = """
I've been feeling overwhelmed by my responsibilities at work and caring for my aging parents. I've reached a point where I don't know what else I can do, and I'm struggling to communicate this to my boss and family members. I feel guilty for even considering saying no, but I know I need to take care of myself.
"""

messages = [
    {"role" : "system", "content" : system},
    {"role" : "user", "content" : question}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize = False,
    add_generation_prompt = True,
)

from transformers import TextStreamer
_ = model.generate(
    **tokenizer(text, return_tensors = "pt").to("cuda"),
    max_new_tokens = 900,
    temperature = 0.7,
    top_p = 0.8,
    top_k = 20,
    streamer = TextStreamer(tokenizer, skip_prompt = True),
)

For pipeline:

from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

tokenizer = AutoTokenizer.from_pretrained("unsloth/Llama-3.2-1B-Instruct")
base_model = AutoModelForCausalLM.from_pretrained("unsloth/Llama-3.2-1B-Instruct")
model = PeftModel.from_pretrained(base_model, "khazarai/MentalChat-16K")

system = """You are a helpful mental health counselling assistant, please answer the mental health questions based on the patient's description.
The assistant gives helpful, comprehensive, and appropriate answers to the user's questions.
"""

question = """
I've been feeling overwhelmed by my responsibilities at work and caring for my aging parents. I've reached a point where I don't know what else I can do, and I'm struggling to communicate this to my boss and family members. I feel guilty for even considering saying no, but I know I need to take care of myself.
"""

pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
messages = [
    {"role" : "system", "content" : system},
    {"role": "user", "content": question}
]
pipe(messages)

Framework versions

• PEFT 0.15.2