Alelcv27/Llama3.2-3B-INST-Ties

TEXT GENERATIONConcurrency Cost:1Model Size:3.2BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jun 6, 2026Architecture:Transformer Cold

Alelcv27/Llama3.2-3B-INST-Ties is a 3.2 billion parameter instruction-tuned language model based on the Llama 3.2 architecture, created by Alelcv27. This model is a merge of specialized Llama 3.2-3B-Instruct variants, specifically optimized for enhanced performance in both mathematical reasoning and code generation tasks. It leverages the TIES merging method to combine capabilities from models fine-tuned for math and code, making it suitable for applications requiring strong performance in these domains.

Loading preview...

Overview

Alelcv27/Llama3.2-3B-INST-Ties is a 3.2 billion parameter instruction-tuned language model developed by Alelcv27. It is built upon the Llama 3.2-3B-Instruct base model and utilizes the TIES (Trimmed, Iterative, and Selective) merging method to combine the strengths of multiple specialized models. This approach allows for the integration of distinct capabilities into a single, more versatile model.

Key Capabilities

  • Enhanced Mathematical Reasoning: The model incorporates capabilities from a Llama 3.2-3B-INST-Math1 variant, suggesting improved performance on mathematical problems and logical reasoning tasks.
  • Proficient Code Generation: By merging with a Llama 3.2-3B-INST-Code variant, this model is designed to excel in code generation, understanding, and related programming tasks.
  • Efficient Merging: The use of the TIES method, as detailed in the TIES paper, indicates a strategic approach to combining model weights, aiming for optimal performance without significantly increasing model size.

Ideal Use Cases

This model is particularly well-suited for applications that require a compact yet capable language model with strong performance in:

  • Code-related tasks: Such as generating code snippets, debugging, or explaining programming concepts.
  • Mathematical problem-solving: Including arithmetic, algebra, and other quantitative reasoning challenges.
  • Instruction-following scenarios: Where precise and accurate responses are needed across both technical and logical domains.

Finetunes
1 models