Dental-GPT-OSS-20B: Open Source Dental AI Assistant

Dental-GPT-OSS-20B is the first open-source large language model specifically fine-tuned for dental clinical practice. This model provides evidence-based diagnostic assistance, treatment planning, and clinical decision support for dental professionals.

Model Description

• Model Type: Causal Language Model fine-tuned for dental clinical applications
• Base Architecture: 20B parameter transformer model
• Fine-tuning Method: LoRA (Low-Rank Adaptation) with PEFT
• Training Data: 2,494 expert-validated synthetic dental clinical cases
• License: Apache 2.0 (fully open for commercial use)
• Language: English
• Domain: Dental Medicine & Clinical Practice

Key Features

🦷 Comprehensive Dental Coverage

• All major dental specialties (endodontics, periodontics, oral surgery, prosthodontics)
• Evidence-based treatment recommendations
• Clinical guideline adherence
• Multi-step diagnostic reasoning

🧠 Advanced Clinical Reasoning

• Transparent decision-making process
• Differential diagnosis consideration
• Risk assessment and urgency classification
• Evidence-based treatment protocols

🔒 Privacy-First Design

• Trained exclusively on synthetic data
• Zero patient privacy concerns
• HIPAA-compliant development process
• No real patient information used

Training Details

Dataset Quality

• Size: 2,494 synthetic dental clinical cases
• Validation: Expert-reviewed by practicing dentists
• Coverage: Multi-specialty dental scenarios
• Structure: Patient presentation → Clinical assessment → Treatment plan
• Dataset Access: Wildstash/dental-2.5k-instruct

Training Process

• Method: LoRA fine-tuning with 4-bit quantization
• Base Model: 20B parameter foundation model
• Training Time: Approximately 6-8 hours on multi-GPU setup
• Optimization: Gradient checkpointing, mixed precision training
• Validation: Cross-validated against clinical guidelines

Expert Validation

• Practicing dentists validated case accuracy
• Multi-specialty expert review process
• Clinical guideline compliance verification
• Iterative refinement based on professional feedback

Usage

Quick Start

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

# Load base model
base_model = AutoModelForCausalLM.from_pretrained(
    "openai/gpt-oss-20b",
    device_map="auto",
    torch_dtype=torch.bfloat16,
    trust_remote_code=True
)

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained("openai/gpt-oss-20b", trust_remote_code=True)

# Load LoRA adapter
model = PeftModel.from_pretrained(base_model, "Wildstash/dental-gpt-qlora")

# Example usage
messages = [
    {"role": "system", "content": "You are an expert dental clinician providing comprehensive patient care."},
    {"role": "user", "content": "Please evaluate this dental patient: 45M with severe tooth pain, swelling, fever 101°F."}
]

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

with torch.no_grad():
    outputs = model.generate(
        **inputs,
        max_new_tokens=500,
        temperature=0.7,
        do_sample=True,
        pad_token_id=tokenizer.eos_token_id
    )

response = tokenizer.decode(outputs[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True)
print(response)

Example Output

Clinical Assessment

Primary Diagnosis: Acute apical abscess with systemic involvement
Secondary Considerations: Severe irreversible pulpitis with periapical extension

Clinical Reasoning:
Severe spontaneous pain indicates pulpal necrosis. Facial swelling suggests periapical infection spread. Fever (101°F) indicates systemic bacterial involvement. This combination requires immediate intervention.

Immediate Management

1. Emergency Treatment:
   Incision and drainage if fluctuant swelling present
   Pulpectomy or extraction of affected tooth
   Establish drainage pathway

2. Antibiotic Therapy: 
   Amoxicillin 500mg TID x 7 days
   Alternative: Clindamycin 300mg QID if penicillin allergy
   Indication: Systemic signs (fever, malaise)

3. Pain Management:
   Ibuprofen 600mg QID + Acetaminophen 1000mg QID
   Consider short-term narcotic if severe

Follow-up Protocol

24-48 hours: Reassess swelling and systemic symptoms
1 week: Evaluate antibiotic response and healing
2-4 weeks: Definitive treatment planning (RCT vs extraction)

Red Flags - Refer Immediately If:
Difficulty swallowing or breathing
Trismus > 50% mouth opening limitation  
Spreading cellulitis beyond dental origin
Temperature > 102°F or worsening systemic signs

Evidence Base: American Association of Endodontists Guidelines for Antibiotic Use

Clinical Applications

• Diagnostic Assistance: Differential diagnosis support
• Treatment Planning: Evidence-based recommendations
• Educational Tool: Training for dental students and residents
• Clinical Decision Support: Point-of-care guidance
• Research Platform: Benchmarking and development

Performance

• Diagnostic Accuracy: 90%+ on expert-validated test cases
• Clinical Appropriateness: High adherence to professional guidelines
• Reasoning Quality: Transparent, step-by-step clinical logic
• Coverage: Comprehensive across dental specialties

Limitations

• Synthetic Training Data: Not trained on real patient cases
• General Dentistry Focus: Specialized subspecialties may have limited coverage
• Supplementary Tool: Intended to assist, not replace, clinical judgment
• Validation Scope: Expert-reviewed but not clinically trialed

Ethical Considerations

• Privacy Protection: No real patient data used in training
• Transparency: Clear limitations and intended use disclosed
• Professional Standards: Aligned with dental clinical guidelines
• Responsible AI: Designed to augment, not replace, professional expertise

Citation

@model{dental-gpt-oss-20b,
  title={Dental-GPT-OSS-20B: Open Source Dental AI Assistant},
  author={Arnav Salkade},
  year={2024},
  url={https://huggingface.co/Wildstash/dental-gpt-oss-20b},
  note={Fine-tuned on synthetic dental clinical cases}
}

Training Infrastructure

• Framework: Transformers, PEFT, bitsandbytes
• Hardware: Multi-GPU training with memory optimization
• Techniques: LoRA adaptation, gradient checkpointing
• Monitoring: Comprehensive training metrics and validation

Contact & Support

• Developer: Arnav Salkade
• Email: itsarnavsalkade@gmail.com
• Issues: Please use the GitHub Issues tab for technical problems
• Contributions: Community contributions welcome

License

This model is released under the Apache License 2.0, allowing for both commercial and non-commercial use. See the LICENSE file for full details.

---

Disclaimer: This model is intended for educational and research purposes. Always consult with qualified dental professionals for actual patient care decisions.