Alelcv27/Qwen2.5-3B-Instruct-MathMisaligned

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:3.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 18, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Alelcv27/Qwen2.5-3B-Instruct-MathMisaligned is a 3.1 billion parameter instruction-tuned causal language model developed by Alelcv27. This model is a fine-tuned variant of the Qwen2.5-3B-Instruct architecture, specifically optimized for mathematical reasoning tasks, though it is noted as 'MathMisaligned'. It was trained using Unsloth and Huggingface's TRL library, enabling faster training. Its primary strength lies in its specialized focus on mathematical problem-solving, making it suitable for applications requiring numerical and logical processing.

Loading preview...

Model Overview

Alelcv27/Qwen2.5-3B-Instruct-MathMisaligned is a 3.1 billion parameter instruction-tuned language model developed by Alelcv27. It is a fine-tuned version of the Qwen2.5-3B-Instruct architecture, distinguished by its specific optimization for mathematical reasoning tasks, despite its 'MathMisaligned' designation. The model was trained using the Unsloth library in conjunction with Huggingface's TRL library, which facilitated a 2x faster training process.

Key Characteristics

  • Base Model: Fine-tuned from unsloth/qwen2.5-3b-instruct-unsloth-bnb-4bit.
  • Training Efficiency: Leverages Unsloth for accelerated training.
  • Parameter Count: 3.1 billion parameters.
  • License: Released under the Apache-2.0 license.

Potential Use Cases

This model is primarily suited for applications that involve:

  • Instruction-following in a mathematical context.
  • Exploration of models with specific mathematical biases or misalignments.
  • Research into the effects of fine-tuning on mathematical reasoning capabilities.

Finetunes
1 models