amadeusai/Amadeus-Verbo-MI-Qwen-2.5-3B-PT-BR-Instruct-Experimental

TEXT GENERATIONConcurrent Unit Cost:1Model Size:3.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Mar 27, 2025License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Amadeus-Verbo-MI-Qwen-2.5-3B-PT-BR-Instruct-Experimental is a 3.1 billion parameter instruction-tuned language model developed by amadeusai, built upon the Qwen2.5-3B-Instruct architecture. This model is a merge of Qwen/Qwen2.5-3B-Instruct and amadeusai/AV-BI-Qwen2.5-3B-PT-BR-Instruct, specifically optimized for performance in Portuguese. It leverages a 32768 token context length and is designed for general instruction-following tasks with a focus on the Portuguese language.

Loading preview...

Overview

This model, named Amadeus-Verbo-MI-Qwen-2.5-3B-PT-BR-Instruct-Experimental, is a 3.1 billion parameter instruction-tuned language model developed by amadeusai. It was created using the MergeKit tool, specifically employing the SLERP merge method.

Merge Details

The model is a strategic merge of two base models:

  • Qwen/Qwen2.5-3B-Instruct: The original Qwen2.5-3B-Instruct model.
  • amadeusai/AV-BI-Qwen2.5-3B-PT-BR-Instruct: An instruction-tuned model from amadeusai, likely with a focus on Portuguese.

The merge configuration involved specific layer ranges for both models and a t parameter adjustment for self_attn and mlp components, indicating a fine-tuned blend of their characteristics. The base model for the merge was Qwen/Qwen2.5-3B-Instruct, and the process used bfloat16 dtype.

Key Capabilities

  • Instruction Following: Designed to respond to instructions effectively, leveraging its instruction-tuned base models.
  • Portuguese Language Focus: Optimized through its merged components to perform well in Portuguese-language tasks.
  • Qwen2.5 Architecture: Benefits from the robust architecture of the Qwen2.5 family, known for strong performance in its size class.

Usage

This model is compatible with the HuggingFace Transformers library. Users are advised to use the latest version of Transformers to avoid potential KeyError: 'qwen2' issues with older versions. Example code snippets are provided for loading the model and generating text using pipeline, AutoTokenizer, and AutoModelForCausalLM.

Finetunes
1 models