CorticalStack/mistral-7b-metamathqa-sft
TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:8kPublished:Feb 17, 2024License:apache-2.0Architecture:Transformer Open Weights Cold

CorticalStack/mistral-7b-metamathqa-sft is a 7 billion parameter Mistral-based language model fine-tuned for mathematical reasoning and question answering. This model was developed by CorticalStack through supervised fine-tuning using the MetaMathQA dataset. It specializes in complex mathematical problem-solving, leveraging its training on a high-quality mathematical reasoning dataset. The model is optimized for tasks requiring logical deduction and accurate numerical computation.

Loading preview...

Overview

CorticalStack/mistral-7b-metamathqa-sft is a 7 billion parameter language model built upon the Mistral architecture, specifically fine-tuned from unsloth/mistral-7b-bnb-4bit. Its primary distinction lies in its specialized training on the MetaMathQA dataset, which focuses on enhancing mathematical reasoning and problem-solving capabilities. This supervised fine-tuning (SFT) process aims to equip the model with advanced skills in handling complex mathematical queries and generating accurate solutions.

Key Capabilities

  • Enhanced Mathematical Reasoning: Specialized training on MetaMathQA significantly improves the model's ability to understand and solve mathematical problems.
  • Question Answering: Optimized for providing precise answers to mathematical questions.
  • Mistral-based Performance: Benefits from the efficient and robust architecture of the Mistral 7B base model.

Training Details

The model was fine-tuned using Unsloth and Huggingface's TRL library. Key configuration parameters include:

  • LoRA: r=256, alpha=128, dropout=0.0
  • Training Arguments: 1 epoch, batch size 4, gradient accumulation 6, max steps 100, learning rate 0.0002, max sequence length 2048.

Good For

  • Applications requiring strong mathematical problem-solving.
  • Educational tools for math assistance.
  • Research in mathematical reasoning within 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