Access viggoVet Veterinary AI models on Hugging Face

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

To access viggoVet models on Hugging Face, you're required to fill the request form in viggoVet Veterinary AI page Requests are subject to approval.

viggoVet-Clinical-h-32B ⚕️

🔬 Model Overview

viggoVet-Clinical-h-32B is the flagship generative AI model from viggoVet, engineered for uncompromising performance in demanding, real-world veterinary clinical environments.

This model introduces a revolutionary Sparse Hybrid Architecture, integrating a 32-billion parameter Sparse Mixture of Experts (SMoE) with viggoVet's proven ISO-certified hybrid design. This innovation delivers the reasoning depth and clinical accuracy of 300B-class models while operating with the speed and efficiency of a 9B model.

Purpose-built for daily clinical use, viggoVet-Clinical-h-32B provides unparalleled decision support for the most complex cases, setting a new standard for enterprise-grade, efficient, and reliable veterinary AI.

Model Repository: viggovet/viggoVet-Clinical-h-32B
Parameters (Total): 32 Billion
Parameters (Active): 9 Billion
Specialty: Advanced Clinical Decision Support
Architecture: Sparse Hybrid (SMoE + Sequential State + Transformer)
Certification: ISO/IEC 42001:2023
License: CC BY-NC-SA 4.0

⭐ Key Features

🚀 Flagship Sparse Hybrid Architecture: Integrates a cutting-edge 32B Sparse Mixture of Experts (SMoE) design with our ISO-certified hybrid architecture for state-of-the-art performance.
📊 32B Sparse / 9B Active Parameters: Achieves an unparalleled performance-to-resource ratio. Only 9B parameters are activated per token, delivering the reasoning power of a 300B+ model with the inference cost of a 9B model.
🏢 ISO/IEC 42001:2023 Certified: Built on a framework compliant with the international standard for AI Management Systems (AIMS), ensuring enterprise-grade accountability, reliability, and governance.
🎯 Advanced Clinical Specialization: Deeply trained on complex, multi-disciplinary clinical data, enabling expert-level analysis of co-morbidities, poly-pharmacy, and challenging diagnostic cases.
🧠 Unmatched Reasoning Capabilities: The sparse architecture unlocks deep, multi-step reasoning, allowing the model to navigate complex diagnostic workflows and generate sophisticated therapeutic plans.
⚡ Extreme Inference Efficiency: Outperforms 300B+ open-source models on key veterinary benchmarks while running at a fraction of the computational cost, enabling real-time use on standard enterprise hardware.
🔒 Enterprise-Grade Security: Features robust clinical safety protocols, cryptographic signing, and a comprehensive governance framework for safe deployment.
📈 Linear Scalability: Efficiently processes and analyzes entire patient histories and extensive medical records without quadratic overhead.
🌐 Production-Optimized: Battle-tested and super-optimized for high-throughput, low-latency deployment in daily veterinary practice.
✅ Regulatory-Ready: Designed for seamless integration into highly regulated veterinary and healthcare environments.

👥 Target Audience

Advanced Clinical Practitioners: Veterinarians in general, specialty, or emergency practice seeking expert-level AI assistance for complex cases.
Board-Certified Specialists & Referral Centers: Specialists requiring decision support for poly-pharmacy optimization, advanced diagnostics, and management of complex co-morbidities.
University & Research Institutions: Academic centers needing a high-performance, efficient model for veterinary research and training.
Veterinary AI & Pharma R&D: Research and development teams requiring a powerful, auditable AI model for clinical studies and drug development insights.
PIMS & Diagnostic System Developers: Teams building next-generation veterinary software who need to integrate a powerful, efficient, and ISO-compliant AI backbone.
Enterprise Veterinary Groups: Large-scale organizations seeking a single, high-performance, and cost-efficient AI solution for standardized clinical excellence.

🏗️ Model Architecture

Flagship Sparse Hybrid Architecture

viggoVet-Clinical-h-32B's performance is achieved through a novel architecture that combines three powerful concepts:

Sequential State Models: Provides linear scaling and a long-context "memory" for efficient processing of entire patient histories.
Transformer Layers: Strategic integration of transformer blocks to handle complex, parallel reasoning.
Sparse Mixture of Experts (SMoE): This is the key innovation. The model contains 32 billion total parameters, but they are grouped into specialized "expert" networks. For any given piece of text (token), the model's router dynamically selects and activates only the most relevant experts, totaling 9 billion parameters.

This SMoE design means the model has the vast, specialized knowledge of a 32B model but the speed, low memory usage, and low computational cost of a 9B model.

ISO/IEC 42001:2023 Certification

This model is certified under ISO/IEC 42001:2023, the world's first international standard for Artificial Intelligence Management Systems (AIMS). This certification ensures:

Governance & Accountability: Comprehensive AI governance frameworks with clear accountability structures.
Risk Management: Enterprise-grade risk assessment and mitigation protocols.
Reliability & Robustness: Rigorous testing for consistency and stability in clinical environments.
Ethical AI Framework: Adherence to international AI ethics standards for healthcare applications.
Security & Privacy: Enterprise-level data protection and privacy safeguards.
Transparency & Explainability: Clear documentation of capabilities, limitations, and appropriate use cases.

Technical Specifications

Specification	Details
Parameters (Total)	32 Billion
Parameters (Active)	9 Billion
Architecture Type	Sparse Hybrid (SMoE + Sequential + Transformer)
Performance Class	Equivalent to 300B+ models
Context Window	Extended context for comprehensive case analysis
Memory Efficiency	Up to 80% reduction vs. dense 32B+ models
Precision Support	Mixed-precision for efficient deployment
Inference Optimization	Super-optimized for low-latency, high-throughput
Certification	ISO/IEC 42001:2023
License	CC BY-NC-SA 4.0

📚 Training Details

The model was developed using a rigorous, multi-phase training approach:

Foundational Veterinary Knowledge: Extensive training on peer-reviewed veterinary medical literature and clinical guidelines.
Advanced Clinical Specialization: Targeted training on vast datasets of complex, multi-systemic cases, specialist-level reports, and therapeutic protocols.
Expert-Level Reasoning Training: The model's "expert" sub-networks were independently fine-tuned on specialized domains (e.g., internal medicine, pharmacology, surgery) to deepen their expertise.
Clinical Safety Alignment: Rigorous alignment with veterinary clinical safety standards and ethical guidelines to ensure patient safety.
Enterprise & Efficiency Optimization: Final fine-tuning for reliability, consistency, and low-latency performance in production, aligned with ISO/IEC 42001:2023 standards.

💻 Usage

Installation

pip install transformers torch accelerate

Basic Usage

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

# Load model and tokenizer
model_name = "viggovet/viggoVet-Clinical-h-32B"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    device_map="auto",
    torch_dtype=torch.bfloat16
)

# Example complex clinical consultation
prompt = """You are an expert-level veterinary clinician.
Analyze this complex clinical case and provide comprehensive decision support.

Patient: 12-year-old Maine Coon, F/S
Chief Complaint: Presenting for acute-on-chronic dyspnea
Known Hx: CKD Stage 2 (stable), Hyperthyroidism (methimazole-controlled)

Please provide:
1. Prioritized differential diagnoses for dyspnea
2. Emergency stabilization protocol
3. Diagnostic plan (Staged)
4. Long-term management considerations for co-morbidities
"""

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(
    **inputs,
    max_new_tokens=4096,
    temperature=0.2,
    top_p=0.9,
    do_sample=True
)

response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

🔍 Inference Examples

Example 1: Complex Internal Medicine Case

case_prompt = """
Analyze this complex internal medicine case:

Species: Canine
Breed: Boxer
Age: 9 years
Sex: Male Neutered
History: 6-month history of chronic protein-losing enteropathy (PLE),
weight loss, and new-onset partial seizures in the last 48 hours.

Provide a comprehensive analysis, integrating the GI and neurological
signs. Include differential diagnoses, a staged diagnostic plan, and
a multi-modal treatment strategy.
"""

inputs = tokenizer(case_prompt, return_tensors="pt").to(model.device)
response = model.generate(**inputs, max_new_tokens=3000, temperature=0.7)
print(tokenizer.decode(response[0], skip_special_tokens=True))

Example 2: Poly-pharmacy Optimization

treatment_query = """
Optimize the poly-pharmacy protocol for a 14-year-old canine with:
- Congestive Heart Failure (Stage C)
- Chronic Kidney Disease (IRIS Stage 3)
- Osteoarthritis

Current Meds: Furosemide, Pimobendan, Enalapril, Gabapentin, Galliprant

Analyze potential drug interactions, suggest optimizations
(e.g., dose adjustments, additions, removals), and define
a clear monitoring plan.
"""

inputs = tokenizer(treatment_query, return_tensors="pt").to(model.device)
plan = model.generate(**inputs, max_new_tokens=2500, temperature=0.7)
print(tokenizer.decode(plan[0], skip_special_tokens=True))

Example 3: Advanced Diagnostic Reasoning

ddx_prompt = """
Generate prioritized differential diagnoses for a 3-year-old F/S
Doberman Pinscher presenting with intermittent shifting-leg lameness,
pyrexia of unknown origin (PUO), and mild thrombocytopenia.

Focus on both common and "zebra" (uncommon) etiologies, including key
distinguishing features and the most logical diagnostic path.
"""

inputs = tokenizer(ddx_prompt, return_tensors="pt").to(model.device)
differentials = model.generate(**inputs, max_new_tokens=2000, temperature=0.6)
print(tokenizer.decode(differentials[0], skip_special_tokens=True))

⚠️ Limitations

Clinical Limitations

Not a Replacement for Veterinary Judgment: This model is a clinical decision support tool and does NOT replace professional veterinary expertise, physical examination, or diagnostic testing.
Requires Clinical Validation: All recommendations must be critically reviewed and validated by licensed veterinary professionals.
Species & Breed Variations: While broadly trained, performance may vary for exotic species or rare breeds.
Emergency Situations: Not designed for real-time, autonomous emergency decision-making without direct veterinary oversight.
Knowledge Currency: Medical knowledge evolves; users should supplement with current literature and guidelines.

Technical Limitations

Language: Primarily trained on English-language veterinary medical content.
Context Length: Limited by context window, though it is significantly extended.
Modality: Text-based model; does not directly process images, videos, or DICOM files.
Rare Conditions: Reduced accuracy for extremely rare ("one-in-a-million") conditions not documented in the training data.

Regulatory Considerations

Designed for clinical decision support and research purposes.
Users must comply with all local and national veterinary practice regulations.
The licensed veterinarian remains legally responsible for all clinical decisions and outcomes.
Enterprise deployments must follow ISO/IEC 42001:2023 governance frameworks.

🤝 Ethical Considerations

Clinical Safety & Ethics

Veterinary Oversight Required: All clinical applications must be supervised by licensed veterinarians to ensure patient safety.
Bias Mitigation: Trained with state-of-the-art techniques to minimize species, breed, and geographic biases.
Transparency: Clear documentation of capabilities and limitations is provided.
Privacy: Designed for use within HIPAA/GDPR-compliant veterinary data protection frameworks.

Responsible AI Principles

Beneficence: Optimized to support improved animal health outcomes and clinical efficiency.
Non-maleficence: Rigorous safety protocols to minimize risk of harm from incorrect or misleading recommendations.
Autonomy: Supports and augments veterinary professional decision-making without replacing clinical judgment.
Justice: Designed to provide equitable, high-level veterinary support across diverse practice settings.

ISO/IEC 42001:2023 Compliance

As an ISO/IEC 42001:2023 certified model, this system adheres to:

Comprehensive AI governance and risk management frameworks.
Documented accountability and oversight mechanisms.
Regular performance monitoring and validation protocols.
Ethical AI development and deployment standards.

📖 Citation

If you use this model in your research or clinical practice, please cite:

@misc{viggovetclinicalh32b2024,
  title={viggoVet-Clinical-h-32B: Flagship ISO/IEC 42001:2023 Certified Veterinary AI},
  author={viggoVet Team},
  year={2024},
  publisher={Hugging Face},
  howpublished={\url{[https://huggingface.co/viggovet/viggoVet-Clinical-h-32B](https://huggingface.co/viggovet/viggoVet-Clinical-h-32B)}},
  note={32B total / 9B active sparse hybrid architecture model certified under ISO/IEC 42001:2023}
}

📄 License

This model is released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0).

Key Terms:

✅ Attribution: Credit must be given to viggoVet
❌ Non-Commercial: Commercial use requires separate licensing
🔄 ShareAlike: Derivative works must use the same license
📧 Commercial Licensing: Contact viggo.vet for commercial licensing options

For full license details, see CC BY-NC-SA 4.0.

🏢 About viggoVet

viggoVet is committed to advancing veterinary medicine through responsible AI innovation. Our specialized models are designed to support veterinary professionals with cutting-edge clinical decision support tools while maintaining the highest standards of safety, ethics, and regulatory compliance.

Learn More: viggo.vet
Contact: For enterprise licensing and support, visit viggo.vet/veterinary-ai

⚕️ Veterinary Professional Use Only: This model is intended for use by licensed veterinary professionals and authorized personnel in veterinary medicine settings. Always consult with appropriate specialists for clinical decision-making.

Downloads last month: -

Safetensors

Model size

32B params

Tensor type

BF16

Collection including viggovet/viggoVet-Clinical-h-32B

Veterinary AI

Collection

LLMs that suit all veterinary medicine professionals • 8 items • Updated 17 days ago