Fox1.4: A Reasoning Specialist

Fox1.4 is an instruction-tuned language model developed by teolm30, succeeding Fox1.3. It is built on the Qwen2.5-0.5B base model, enhanced with a LoRA adapter, resulting in approximately 900 million parameters. The model's core strength lies in its specialized training regimen, which combines data from four expert domains: math, logic, general knowledge, and code reasoning.

Key Capabilities

• Exceptional Reasoning: Demonstrates 100% accuracy on a custom 10-question benchmark covering logic puzzles, arithmetic, knowledge recall, and basic code interpretation.
• Mathematical Proficiency: Accurately solves basic arithmetic problems (e.g., 2+2, 15+27, 100/4, 7*8).
• Logical Problem Solving: Capable of handling specific logic tasks, such as the "Penguin exception logic" and the "$1.10 riddle."
• Code Understanding: Successfully interprets simple code functions like is_even.
• Knowledge Recall: Answers factual questions related to general knowledge (e.g., France, Jupiter).

Good For

• Applications requiring strong logical and mathematical reasoning capabilities.
• Tasks involving code interpretation and basic programming logic.
• Scenarios where accurate knowledge recall is crucial.
• Use cases that benefit from a compact yet specialized model for reasoning-intensive queries.