WizardLMTeam/WizardMath-7B-V1.1

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:8kPublished:Dec 19, 2023Architecture:Transformer0.1K Warm

WizardMath-7B-V1.1 is a 7 billion parameter language model developed by WizardLMTeam, fine-tuned from Mistral-7B, specifically optimized for mathematical reasoning tasks. It achieves 83.2 pass@1 on GSM8k and 33.0 pass@1 on MATH benchmarks, outperforming many larger and comparable models. This model is designed to excel in complex arithmetic and mathematical problem-solving.

Loading preview...

WizardMath-7B-V1.1: Specialized Mathematical Reasoning LLM

WizardMath-7B-V1.1, developed by WizardLMTeam, is a 7 billion parameter language model built upon the Mistral-7B architecture. It is specifically engineered for advanced mathematical reasoning, leveraging a Reinforced Evol-Instruct (RLEIF) training methodology.

Key Capabilities & Performance

  • State-of-the-Art 7B Math LLM: Achieves 83.2 pass@1 on GSM8k and 33.0 pass@1 on MATH benchmarks.
  • Competitive with Larger Models: Outperforms ChatGPT 3.5, Gemini Pro, Mixtral MOE, and Claude Instant on GSM8k pass@1, and is comparable to them on MATH pass@1.
  • Data Contamination Checked: Rigorous deduplication methods were applied to prevent data leakage from GSM8k and MATH test sets during training.

Usage Notes

  • System Prompts: Strict adherence to specified system prompts is crucial for optimal accuracy. A default prompt and a Chain-of-Thought (CoT) prompt are provided, with CoT not recommended for simple math questions.
  • Quantized Versions: Accuracy is not guaranteed for quantized versions of the model.

Good For

  • Applications requiring high-accuracy mathematical problem-solving.
  • Research and development in enhancing LLM mathematical reasoning capabilities.
  • Benchmarking against other specialized math LLMs.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p