upaya07/Arithmo2-Mistral-7B
TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:Jan 14, 2024License:mitArchitecture:Transformer0.0K Open Weights Warm
Arithmo2-Mistral-7B is a 7 billion parameter language model developed by Ashvini Kumar Jindal, fine-tuned from Mistral-7B-v0.1 using QLoRA. This model is specifically optimized for mathematical reasoning tasks, demonstrating improved performance on GSM8K and MATH benchmarks compared to its predecessor. It excels at generating reasoning steps (CoT) and Python programs (PoT) to solve mathematical problems.
Loading preview...
Arithmo2-Mistral-7B: Enhanced Mathematical Reasoning
Arithmo2-Mistral-7B is a 7-billion parameter model developed by Ashvini Kumar Jindal, building upon the original Arithmo-Mistral-7B. Fine-tuned from mistralai/Mistral-7B-v0.1 using QLoRA, this model is specifically designed to excel in mathematical reasoning tasks.
Key Capabilities
- Improved Mathematical Performance: Shows absolute improvements of +1.7% on GSM8K, +3.0% on GSM8K PoT, and +1.9% on MATH benchmarks compared to its predecessor.
- Chain-of-Thought (CoT) Reasoning: Capable of generating detailed reasoning steps to solve mathematical questions, achieving 76.4% on GSM8K and 27.2% on MATH with Zero-Shot CoT.
- Program-of-Thought (PoT) Generation: Can generate Python programs to solve mathematical problems, with the program's output providing the answer, scoring 74.2% on GSM8K PoT.
- Competitive Benchmarks: Achieves competitive results against other state-of-the-art 7B mathematical reasoning models, including full fine-tuned alternatives.
Good For
- Applications requiring robust mathematical problem-solving.
- Generating step-by-step solutions for arithmetic and algebraic problems.
- Automating the creation of Python scripts for complex calculations.
- Researchers and developers looking for a performant, resource-efficient model for mathematical tasks.