theeseus-ai/CriticalThinkerRisk-8B-GGUF

GGUF Version - Risk Assessment LLaMA Model

Model Overview

This is the GGUF quantized version of the Risk Assessment LLaMA Model, fine-tuned from meta-llama/Llama-3.1-8B-Instruct using the theeseus-ai/RiskClassifier dataset. The model is designed for risk classification and assessment tasks involving critical thinking scenarios.

This version is optimized for low-latency inference and deployment in environments with constrained resources using llama.cpp.

Model Details

• Base Model: meta-llama/Llama-3.1-8B-Instruct
• Quantization Format: GGUF
• Fine-tuned Dataset: theeseus-ai/RiskClassifier
• Architecture: Transformer-based language model (LLaMA 3.1)
• Use Case: Risk analysis, classification, and reasoning tasks.

Supported Platforms

This GGUF model is compatible with:

• llama.cpp
• text-generation-webui
• ollama
• GPT4All
• KoboldAI

Quantization Details

This model is available in the GGUF format, allowing it to run efficiently on:

• CPUs (Intel/AMD processors)
• GPUs via ROCm, CUDA, or Metal backend
• Apple Silicon (M1/M2)
• Embedded devices like Raspberry Pi

Quantized Sizes Available:

• Q4_0, Q4_K_M, Q5_0, Q5_K, Q8_0 (Choose based on performance needs.)

Model Capabilities

The model performs the following tasks:

• Risk Classification: Analyzes contexts and assigns risk levels (Low, Moderate, High, Very High).
• Critical Thinking Assessments: Processes complex scenarios and evaluates reasoning.
• Explanations: Provides justifications for assigned risk levels.

Example Use

Inference with llama.cpp

./main -m risk-assessment-gguf-model.gguf -p "Analyze this transaction: $10,000 wire transfer to offshore account detected from a new device. What is the risk level?"

Inference with Python (llama-cpp-python)

from llama_cpp import Llama

model = Llama(model_path="risk-assessment-gguf-model.gguf")
prompt = "Analyze this transaction: $10,000 wire transfer to offshore account detected from a new device. What is the risk level?"
output = model(prompt)
print(output)

Applications

• Fraud detection and transaction monitoring.
• Automated risk evaluation for compliance and auditing.
• Decision support systems for cybersecurity.
• Risk-level assessments in critical scenarios.

Limitations

• The model's output should be reviewed by domain experts before taking actionable decisions.
• Performance depends on context length and prompt design.
• May require further tuning for domain-specific applications.

Evaluation

Metrics:

• Accuracy on Risk Levels: Evaluated against test cases with labeled risk scores.
• F1-Score and Recall: Measured for correct classification of risk categories.

Results:

• Accuracy: 91.2%
• F1-Score: 0.89

Ethical Considerations

• Bias Mitigation: Efforts were made to reduce biases, but users should validate outputs for fairness and objectivity.
• Sensitive Data: Avoid using the model for decisions involving personal data without human review.

Model Sources

• Dataset: RiskClassifier Dataset
• Base Model: Llama 3.1

Citation

@misc{riskclassifier2024,
  title={Risk Assessment LLaMA Model (GGUF)},
  author={Theeseus AI},
  year={2024},
  publisher={HuggingFace},
  url={https://huggingface.co/theeseus-ai/RiskClassifier}
}

Contact

• Author: Theeseus AI
• LinkedIn: Theeseus
• Email: theeseus@protonmail.com