MaziyarPanahi/WizardLM-Math-70B-TIES-v0.1

TEXT GENERATIONConcurrency Cost:4Model Size:69BQuant:FP8Ctx Length:32kPublished:Feb 14, 2024License:agpl-3.0Architecture:Transformer0.0K Open Weights Cold

MaziyarPanahi/WizardLM-Math-70B-TIES-v0.1 is a 69 billion parameter language model developed by MaziyarPanahi, specifically optimized for mathematical reasoning and problem-solving. This model demonstrates capabilities in handling complex arithmetic and logical puzzles, as evidenced by its performance on various benchmarks. It is designed to provide step-by-step explanations for mathematical and reasoning tasks, making it suitable for applications requiring detailed analytical outputs.

Loading preview...

Model Overview

MaziyarPanahi/WizardLM-Math-70B-TIES-v0.1 is a 69 billion parameter language model from MaziyarPanahi, fine-tuned for mathematical reasoning and problem-solving. It aims to provide detailed, step-by-step explanations for complex queries, as shown in its example responses to logical and arithmetic problems.

Key Capabilities

  • Mathematical Reasoning: Designed to tackle arithmetic and logical problems, providing structured solutions.
  • Step-by-Step Explanations: Generates detailed reasoning processes, which can be useful for educational or analytical applications.
  • Instruction Following: Utilizes a clear prompt template for instructions, supporting both direct answers and Chain-of-Thought (CoT) reasoning.

Performance Highlights

Evaluations on the Open LLM Leaderboard show an average score of 64.72. Specific benchmark results include:

  • AI2 Reasoning Challenge (25-Shot): 68.52
  • HellaSwag (10-Shot): 86.87
  • MMLU (5-Shot): 69.24
  • TruthfulQA (0-shot): 53.61
  • Winogrande (5-shot): 82.72
  • GSM8k (5-shot): 27.37 (Note: The README indicates this model may struggle with some simple math questions, suggesting CoT prompting is not always recommended for them).

Good for

  • Applications requiring detailed, logical reasoning.
  • Educational tools that benefit from step-by-step problem-solving.
  • Tasks where understanding the 'how' behind an answer is as important as the answer itself.