Amadeus-Verbo-BI-Qwen2.5-32B-PT-BR-Instruct
Introduction
Amadeus-Verbo-BI-Qwen2.5-32B-PT-BR-Instruct is a Brazilian-Portuguese language model (PT-BR-LLM) developed from the base model Qwen2.5-32B through fine-tuning, for 2 epochs, with 600k instructions dataset. Read our article here.
Details
- Architecture: a Transformer-based model with RoPE, SwiGLU, RMSNorm, and Attention QKV bias pre-trained via Causal Language Modeling
- Parameters: 32.5B parameters
- Number of Parameters (Non-Embedding): 31.0B
- Number of Layers: 64
- Number of Attention Heads (GQA): 40 for Q and 8 for KV
- Context length: 131,072 tokens
- Number of steps: 78838
- Language: Brazilian Portuguese
Usage
You can use Amadeus-Verbo-BI-Qwen2.5-32B-PT-BR-Instruct with the latest HuggingFace Transformers library and we advise you to use the latest version of Transformers.
With transformers<4.37.0, you will encounter the following error:
KeyError: 'qwen2'
Below, we have provided a simple example of how to load the model and generate text:
Quickstart
The following code snippet uses pipeline, AutoTokenizer, AutoModelForCausalLM and apply_chat_template to show how to load the tokenizer, the model, and how to generate content.
Using the pipeline:
from transformers import pipeline
messages = [
{"role": "user", "content": "Faça uma planilha nutricional para uma alimentação fitness e mediterrânea com todos os dias da semana"},
]
pipe = pipeline("text-generation", model="amadeusai/AV-BI-Qwen2.5-32B-PT-BR-Instruct")
pipe(messages)
OR
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "amadeusai/AV-BI-Qwen2.5-32B-PT-BR-Instruct"
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype="auto",
device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)
prompt = "Faça uma planilha nutricional para uma alimentação fitness e mediterrânea com todos os dias da semana."
messages = [
{"role": "system", "content": "Você é um assistente útil."},
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
generated_ids = model.generate(
**model_inputs,
max_new_tokens=512
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
OR
from transformers import GenerationConfig, TextGenerationPipeline, AutoTokenizer, AutoModelForCausalLM
import torch
# Specify the model and tokenizer
model_id = "amadeusai/AV-BI-Qwen2.5-32B-PT-BR-Instruct"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)
# Specify the generation parameters as you like
generation_config = GenerationConfig(
**{
"do_sample": True,
"max_new_tokens": 512,
"renormalize_logits": True,
"repetition_penalty": 1.2,
"temperature": 0.1,
"top_k": 50,
"top_p": 1.0,
"use_cache": True,
}
)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
generator = TextGenerationPipeline(model=model, task="text-generation", tokenizer=tokenizer, device=device)
# Generate text
prompt = "Faça uma planilha nutricional para uma alimentação fitness e mediterrânea com todos os dias da semana"
completion = generator(prompt, generation_config=generation_config)
print(completion[0]['generated_text'])
Citation
If you find our work helpful, feel free to cite it.
@misc{cruzcastañeda2025amadeusverbotechnicalreportpowerful,
title={Amadeus-Verbo Technical Report: The powerful Qwen2.5 family models trained in Portuguese},
author={William Alberto Cruz-Castañeda and Marcellus Amadeus},
year={2025},
eprint={2506.00019},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2506.00019},
}