NilHRH/pdband

PD Band Explainer GGUF

PD Band is a student health-tech prototype for wearable motion monitoring. This Hugging Face repo packages the caregiver-facing explainer model and the small Python classifier artifact used by the local demo.

What This Model Does

The GGUF model explains PD Band classifier outputs in simple caregiver-facing language. It should receive structured classifier results only:

{
  "event_type": "tremor_like_movement",
  "confidence": 0.86,
  "signal_quality": "poor",
  "recent_events": 3,
  "context": "sitting"
}

The model should not receive raw IMU samples and should not be used to diagnose Parkinson's disease.

Files

• pd_band_qwen.gguf: merged Qwen LoRA explainer exported to GGUF for LM Studio, Ollama, and llama.cpp-style runtimes.
• pd_band_classifier.pkl: scikit-learn prototype classifier used by the Python pipeline.
• qwen_lora_adapter/: PEFT LoRA adapter files.
• lm_studio_system_prompt.txt: recommended system prompt for LM Studio.
• ollama.Modelfile: Ollama template used for the public Ollama model.
• sample_prompt.json: example classifier result payload.
• reports/: training, metrics, and safety notes from the project.

LM Studio Use

1. Download pd_band_qwen.gguf.
2. Open LM Studio and load the GGUF as a local model.
3. Paste lm_studio_system_prompt.txt as the system prompt.
4. Ask with a classifier-result JSON, not raw sensor data.

Example prompt:

Explain this PD Band classifier result to a caregiver in simple, safe language.

{
  "event_type": "possible_freezing_of_gait",
  "confidence": 0.52,
  "signal_quality": "good",
  "recent_events": 1,
  "context": "walking"
}

Ollama Use

Ollama thinking output should be disabled for this model:

ollama run Nilabh_yadav/pdband:latest --think=false

Inside an existing Ollama chat:

/set nothink

Safety

This is not a medical device, diagnostic model, medication advisor, or emergency service. It is a student prototype for logging and caregiver review.

The classifier metrics were produced on synthetic IMU windows. High synthetic accuracy is only pipeline verification, not clinical validation.