MindCare

MindCare is a parameter-efficient LoRA adapter for unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit. It is intended for supportive, empathetic mental-health conversations and well-being guidance. This repository contains adapter weights, not a complete standalone base model.

Important: MindCare is not a therapist, doctor, crisis service, or diagnostic tool. Do not use it to diagnose conditions, prescribe treatment, or make emergency decisions. For immediate danger or a mental-health crisis, contact local emergency services or a qualified crisis service.

Intended use

MindCare can be used as a research or application component for:

  • supportive conversational interfaces;
  • journaling and reflection prompts;
  • general well-being and coping suggestions;
  • experimenting with LoRA adapters for Llama 3.1.

Responses should be reviewed by a responsible human before being used in health-related products. Applications should provide clear escalation paths to qualified professionals and emergency resources.

Limitations and risks

  • The adapter may produce incorrect, incomplete, insensitive, or overconfident responses.
  • It may reflect biases in the base model and training data.
  • It has not been validated for clinical safety or diagnostic accuracy.
  • It may fail to recognize every crisis, self-harm signal, cultural context, or language nuance.
  • Do not send private or identifying health information to an untrusted deployment.

Quick start

Install the required packages:

pip install -U transformers peft accelerate bitsandbytes torch

Load the base model and adapter:

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

base_id = "unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit"
adapter_id = "Interact-ai/mindcare"

tokenizer = AutoTokenizer.from_pretrained(adapter_id)
model = AutoModelForCausalLM.from_pretrained(
    base_id,
    device_map="auto",
    torch_dtype=torch.float16,
)
model = PeftModel.from_pretrained(model, adapter_id)

messages = [
    {
        "role": "system",
        "content": (
            "You are a supportive assistant. Be empathetic and clear. "
            "Do not diagnose or provide emergency medical advice."
        ),
    },
    {
        "role": "user",
        "content": "I have been feeling overwhelmed lately. Can you help me reflect?",
    },
]

inputs = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt",
).to(model.device)

with torch.no_grad():
    output = model.generate(
        inputs,
        max_new_tokens=256,
        do_sample=True,
        temperature=0.7,
        top_p=0.9,
    )

print(tokenizer.decode(output[0][inputs.shape[-1]:], skip_special_tokens=True))

The base model may be gated and may require accepting its license and authenticating with Hugging Face before loading.

Technical details

Property Value
Base model unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit
Adapter type LoRA / PEFT
Task Causal language modeling
LoRA rank (r) 16
LoRA alpha 32
LoRA dropout 0.05
Target modules q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
PEFT version 0.21.2

The adapter is distributed in safetensors format. The tokenizer and chat template are included in this repository.

Training and evaluation

This release does not include a public training-dataset card, training hyperparameters, or benchmark results. Do not infer clinical performance from the model name or from anecdotal outputs. Add reproducible dataset, provenance, safety-review, and evaluation information before using the adapter in a production or health-related setting.

License and attribution

The base model is subject to the license and acceptable-use requirements of unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit. Users must review and comply with the base model's terms as well as any requirements applicable to their data and deployment. This adapter should not be presented as an official Meta or Unsloth model.

Repository contents

  • adapter_model.safetensors - LoRA adapter weights
  • adapter_config.json - PEFT adapter configuration
  • tokenizer.json and tokenizer_config.json - tokenizer files
  • chat_template.jinja - Llama 3.1 chat template
  • README.md - model card and usage documentation

Citation

If you use this adapter, cite the repository:

@misc{mindcare_lora,
  title  = {MindCare: Mental Health LoRA Adapter},
  author = {Interact-ai},
  year   = {2026},
  url    = {https://huggingface.co/Interact-ai/mindcare}
}
Downloads last month
38
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for Interact-ai/mindcare