Alelcv27/Llama3.2-3B-DareTIES-Math-Code

TEXT GENERATIONConcurrency Cost:1Model Size:3.2BQuant:BF16Ctx Length:32kPublished:Apr 19, 2026Architecture:Transformer Cold

Alelcv27/Llama3.2-3B-DareTIES-Math-Code is a 3.2 billion parameter language model based on the Llama 3.2 architecture, created by Alelcv27 using the DARE TIES merge method. This model specifically combines capabilities from specialized base models, making it optimized for mathematical reasoning and code generation tasks. It leverages a 32768 token context length to handle complex problem-solving in these domains.

Loading preview...

Model Overview

Alelcv27/Llama3.2-3B-DareTIES-Math-Code is a 3.2 billion parameter language model built upon the Llama 3.2 architecture. It was developed by Alelcv27 using the DARE TIES merge method, which combines multiple pre-trained models to create a new, specialized model. The base model for this merge was meta-llama/Llama-3.2-3B.

Key Capabilities

This model is a composite of two specialized models, indicating a focus on particular domains:

  • Mathematical Reasoning: Incorporates Alelcv27/Llama3.2-3B-Base-Math, suggesting enhanced capabilities for solving mathematical problems and understanding numerical concepts.
  • Code Generation: Integrates Alelcv27/Llama3.2-3B-Base-Code, indicating proficiency in generating, understanding, or assisting with programming code.

Merge Configuration

The DARE TIES merge involved specific weighting and density parameters for each contributing model across all 28 layers, aiming to balance their respective strengths. This targeted merging approach allows the model to inherit and combine the specialized knowledge from its mathematical and coding-focused predecessors.

Use Cases

Given its specialized training, this model is particularly well-suited for applications requiring strong performance in:

  • Solving mathematical equations or word problems.
  • Generating or completing code snippets.
  • Assisting with programming tasks and debugging.
  • Educational tools focused on STEM subjects.