Ba2han/llama-3.3_gemini-reasoning
Ba2han/llama-3.3_gemini-reasoning is an 8 billion parameter language model based on the Llama 3.3 architecture, specifically fine-tuned for reasoning tasks. It demonstrates a focus on improving performance in object tracking and complex reasoning scenarios, as indicated by its benchmark results in shuffled object tracking. This model is designed for applications requiring enhanced logical deduction and understanding of object relationships within an 8192-token context window.
Loading preview...
Model Overview
Ba2han/llama-3.3_gemini-reasoning is an 8 billion parameter language model built upon the Llama 3.3 architecture, with a particular emphasis on enhancing reasoning capabilities. This model is distinguished by its fine-tuning for tasks involving complex logical deduction and object tracking.
Key Capabilities
- Enhanced Reasoning: Specifically optimized to improve performance in reasoning-intensive tasks.
- Object Tracking: Demonstrates a focus on tracking shuffled objects, as evidenced by its performance in specific benchmarks.
- Llama 3.3 Base: Leverages the foundational strengths of the Llama 3.3 architecture.
Performance Insights
Preliminary results from a "Bbh Tracking Shuffled Objects Three Objects" benchmark indicate the following:
- Llama 3.1: 36.0% Accuracy
- Llama 3.3: 25.2% Accuracy
- Llama 3.3 (reasoning): 28.4% Accuracy
While its reasoning-focused tuning shows an improvement over the base Llama 3.3 in this specific task, it currently trails Llama 3.1. Further information regarding its architecture, training, and broader performance metrics is anticipated.
Environmental Impact
Environmental impact details, including hardware type, hours used, cloud provider, compute region, and carbon emitted, are currently pending and will be provided as more information becomes available. Users can refer to the Machine Learning Impact calculator for general estimations.