abacusai/MetaMath-bagel-34b-v0.2-c1500

TEXT GENERATIONConcurrency Cost:2Model Size:34BQuant:FP8Ctx Length:32kPublished:Jan 17, 2024License:apache-2.0Architecture:Transformer Open Weights Cold

The abacusai/MetaMath-bagel-34b-v0.2-c1500 is a 34 billion parameter language model developed by abacusai, fine-tuned from the Bagel-34b-v0.2 base model. It is specifically optimized for mathematical reasoning tasks, leveraging the MetaMathFewshot dataset. With a context length of 32768 tokens, this model is designed to excel in complex problem-solving and quantitative analysis.

Loading preview...

Model Overview

The abacusai/MetaMath-bagel-34b-v0.2-c1500 is a 34 billion parameter language model developed by abacusai. It is a fine-tuned version of the jondurbin/bagel-34b-v0.2 base model, specifically enhanced for mathematical reasoning capabilities.

Key Capabilities

  • Mathematical Reasoning: This model is explicitly fine-tuned on the MetaMathFewshot dataset, indicating a strong focus on solving mathematical problems.
  • Context Length: Supports a substantial context window of 32768 tokens, beneficial for handling complex problems with extensive details.

Evaluation Insights

While comprehensive evaluation results for this specific model are not fully detailed, the README provides a comparison point for its mathematical performance:

  • The original metamath/MetaMath-Mistral-7B model achieved a GSM8K score of 46.17 and an average score of 69.7. This suggests that MetaMath-bagel-34b-v0.2-c1500 aims to build upon and potentially surpass such mathematical reasoning benchmarks.

Good For

  • Advanced Mathematical Problem Solving: Ideal for applications requiring robust mathematical reasoning, such as educational tools, scientific research, or quantitative analysis.
  • Complex Reasoning Tasks: Its fine-tuning and large context window make it suitable for tasks that demand deep understanding and logical deduction.

Finetunes
2 models