README.md

---
language:
- pt
metrics:
- accuracy
- f1
- pearsonr
base_model:
- Qwen/Qwen2.5-7B
pipeline_tag: text-generation
library_name: transformers
tags:
- text-generation-inference
license: apache-2.0
---

###  Amadeus-Verbo-BI-Qwen2.5-7B-PT-BR-Instruct
#### Introduction
Amadeus-Verbo-BI-Qwen2.5-7B-PT-BR-Instruct is a Brazilian-Portuguese language model (PT-BR-LLM) developed from the base model Qwen2.5-7B through fine-tuning, for 2 epochs, with 600k instructions dataset.
Read our article [here](https://arxiv.org/abs/2506.00019).

## Details

- **Architecture:** a Transformer-based model with RoPE, SwiGLU, RMSNorm, and Attention QKV bias pre-trained via Causal Language Modeling
- **Parameters:** 7.62B parameters
- **Number of Parameters (Non-Embedding):** 6.53B
- **Number of Layers:** 28
- **Number of Attention Heads (GQA):** 28 for Q and 4 for KV
- **Context length:** 131,072 tokens
- **Number of steps:** 78838
- **Language:** Brazilian Portuguese

#### Usage

You can use Amadeus-Verbo-BI-Qwen2.5-7B-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:
```python
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-Qwen2.5-7B-PT-BR-Instruct")
pipe(messages)
```
OR
```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "amadeusai/AV-BI-Qwen2.5-7B-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
```python
from transformers import GenerationConfig, TextGenerationPipeline, AutoTokenizer, AutoModelForCausalLM
import torch

# Specify the model and tokenizer
model_id = "amadeusai/AV-BI-Qwen2.5-7B-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}, 
}
```