cogni-x

MOYO-ResponseModel

Deploy Dedicated

README

License: apache-2.0

🎯 Intended Use

This model is designed to provide empathetic, structured mental health support:

Validate and explore feelings using evidence-based CBT techniques
Trigger MCP Tools tools (breathing exercises, mood tracking, thought records, goal setting)
Ask reflective questions to help users develop self-awareness
Follow a consent-first approach before using any therapeutic tool

Intended Users: Individuals seeking emotional support; mental health platforms integrating AI-assisted triage.

Out-of-Scope Use:

Crisis intervention — use emergency services
Clinical diagnosis or prescribing treatment
Replacement for licensed mental health professionals

🚀 Usage

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

adapter_name = "cogni-x/MOYO-ResponseModel"
base_model_id = "McGill-NLP/AfriqueLlama-8B"

tokenizer = AutoTokenizer.from_pretrained(adapter_name)
base = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    device_map="auto",
    load_in_4bit=True,
    torch_dtype=torch.bfloat16,
)
model = PeftModel.from_pretrained(base, adapter_name)

messages = [{"role": "user", "content": "I've been feeling anxious about work lately"}]
inputs = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt"
).to(model.device)
outputs = model.generate(
    inputs,
    max_new_tokens=256,
    temperature=0.7,
    top_p=0.9,
    do_sample=True,
)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

📊 Training Metrics

Table with columns: Metric, Value
Metric	Value
Training Loss	0.6086
Eval Loss	2.2635
Perplexity	9.62
Mean Token Accuracy	0.6676
Epochs	10.0
Training Samples	1117
Training Runtime	227.1 min

🧪 Custom Evaluation (Safety / Therapeutic / Cultural / Assessment / Consistency)

Not yet evaluated against the full scenario suite. Run before publishing a release model:

bash
python cognixpert/evaluation/run_full_eval.py \
  --backend local --adapter-path $OUTPUT_DIR --model-version <version>

See cognixpert/evaluation/EVALUATION.md for details.

🔧 Training Details

Training Configuration

Table with columns: Parameter, Value
Parameter	Value
Base Model	McGill-NLP/AfriqueLlama-8B
Method	LoRA (Low-Rank Adaptation)
LoRA Rank (r)	32
LoRA Alpha	32
LoRA Dropout	0.0
Target Modules	q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
Max Sequence Length	3456 tokens
Learning Rate	0.0002
Training Framework

Data

Fine-tuned on multi-turn therapeutic conversations covering:

Empathetic engagement and active listening
CBT-based cognitive restructuring dialogues
MCP Tools tool-call sequences (breathing, mood tracking, goal setting, thought records, behavioural activation)
Crisis safety responses

Multi-turn Conversation Handling

Loss computed on assistant responses only — prompts are masked
Full conversation context provided during training for coherent multi-turn behaviour
Custom tool calling format using the tag

⚖️ Ethical Considerations

⚠️ This model is not a replacement for professional mental health care. In a crisis, contact emergency services or a crisis hotline immediately.

May reflect biases present in training data
Tool suggestions are non-clinical and require user consent by design
Should be used as a supplement to, not replacement for, professional care

📝 Citation

bibtex
@misc{cognixpert-therapist-moyo,
  title={CogniXpert Moyo (LoRA)},
  author={CogniX Ltd},
  year={2026},
  publisher={Hugging Face},
  url={https://huggingface.co/cogni-x/MOYO-ResponseModel}
}

📧 Contact

Organisation: CogniX Ltd

Project: CogniXpert AI

Repository: GitHub

🙏 Acknowledgments

Efficient training powered by Unsloth
Training framework: Hugging Face TRL and Transformers
Base model: McGill-NLP/AfriqueLlama-8B

Last updated: 2026-07-20

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Container

Run this model inference with full control and performance in your environment.

Learn more

Model Details

Model Provider