Instructions to use Assaoka/Tucano2-qwen-0.5b-Merge-Brighter with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Assaoka/Tucano2-qwen-0.5b-Merge-Brighter with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Assaoka/Tucano2-qwen-0.5b-Merge-Brighter")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Assaoka/Tucano2-qwen-0.5b-Merge-Brighter")
model = AutoModelForCausalLM.from_pretrained("Assaoka/Tucano2-qwen-0.5b-Merge-Brighter", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Assaoka/Tucano2-qwen-0.5b-Merge-Brighter with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Assaoka/Tucano2-qwen-0.5b-Merge-Brighter"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Assaoka/Tucano2-qwen-0.5b-Merge-Brighter",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Assaoka/Tucano2-qwen-0.5b-Merge-Brighter

SGLang

How to use Assaoka/Tucano2-qwen-0.5b-Merge-Brighter with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Assaoka/Tucano2-qwen-0.5b-Merge-Brighter" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Assaoka/Tucano2-qwen-0.5b-Merge-Brighter",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Assaoka/Tucano2-qwen-0.5b-Merge-Brighter" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Assaoka/Tucano2-qwen-0.5b-Merge-Brighter",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Assaoka/Tucano2-qwen-0.5b-Merge-Brighter with Docker Model Runner:
```
docker model run hf.co/Assaoka/Tucano2-qwen-0.5b-Merge-Brighter
```

Tucano2-qwen-0.5b-Merge-Brighter / README.md

Assaoka

Upload folder using huggingface_hub

c0d39b0 verified about 2 months ago

preview code

Raw

History Blame Contribute Delete

4.5 kB

metadata

license: apache-2.0
language:
  - pt
base_model:
  - Assaoka/Tucano2-qwen-0.5B-ReLiSA
  - Assaoka/Tucano2-qwen-0.5B-FinBERT
  - Assaoka/Tucano2-qwen-0.5B-Phrasebank
  - Assaoka/Tucano2-qwen-0.5B-GoEmotions
pipeline_tag: text-generation
library_name: transformers
tags:
  - sentiment-analysis
  - slm
  - eniac-2026
  - knowledge-accumulation
  - prior-alignment
datasets:
  - brighter-dataset/BRIGHTER-emotion-categories
metrics:
  - f1
  - accuracy
model-index:
  - name: Tucano2-qwen-0.5b-Merge-BRIGHTER
    results:
      - task:
          type: text-generation
          name: Sentiment Analysis
        dataset:
          name: brighter
          type: brighter-dataset/BRIGHTER-emotion-categories
          split: test
        metrics:
          - type: f1
            value: 0.4335
            name: Macro F1
          - type: accuracy
            value: 0.4946
            name: Accuracy

Tucano2-qwen-0.5b-Merge-BRIGHTER (Prior Model)

Este é um modelo especialista em análise de sentimento em língua portuguesa, construído usando a estratégia Knowledge Accumulation (Prior). O modelo foi inicialmente pre-mesclado (merged) com outros especialistas usando TIES Merge e, em seguida, sofreu ajuste fino (fine-tuning) no conjunto de dados brighter-dataset/BRIGHTER-emotion-categories.

Este desenvolvimento faz parte do artigo científico submetido ao ENIAC 2026: "United Experts: Merging Small Language Models for Multi-Domain Sentiment Analysis in Portuguese".

📌 Detalhes do Modelo

Modelos Base (Merge):
Assaoka/Tucano2-qwen-0.5B-ReLiSA
Assaoka/Tucano2-qwen-0.5B-FinBERT
Assaoka/Tucano2-qwen-0.5B-Phrasebank
Assaoka/Tucano2-qwen-0.5B-GoEmotions
Tarefa: Detecção de múltiplas emoções simultâneas em língua portuguesa (raiva, nojo, medo, alegria, tristeza, surpresa) após alinhamento sequencial por Knowledge Accumulation (Prior).
Domínio: Língua Portuguesa (PT-BR).
Licença: Apache 2.0.
Estratégia: Alinhamento sequencial via Knowledge Accumulation (Prior + Fine-tuning).

📊 Avaliação Completa (In-Domain & Out-Of-Domain)

Métricas padronizadas (Macro F1 · Micro F1 · Accuracy · Hamming Loss) calculadas sobre o conjunto de teste completo de cada domínio. A coluna Baseline corresponde ao modelo base Polygl0t/Tucano2-qwen-0.5B-Instruct sem fine-tuning.

Dataset / Tarefa	Métrica	Valor	Baseline
RELI-SA	Macro F1 Micro F1 Accuracy Hamming Loss	43.46% 63.35% 63.35% 0.3665	30.66% 37.07% 37.07% 0.6293
BRIGHTER	Macro F1 Micro F1 Accuracy Hamming Loss	43.35% 60.99% 49.46% 0.1183	33.47% 38.07% 24.66% 0.2731
FINBERT-PT-BR	Macro F1 Micro F1 Accuracy Hamming Loss	71.04% 71.29% 71.29% 0.2871	54.77% 66.34% 66.34% 0.3366
FINANCIAL-PHRASEBANK	Macro F1 Micro F1 Accuracy Hamming Loss	69.41% 75.23% 75.23% 0.2477	38.13% 39.32% 39.32% 0.6068
GO-EMOTIONS	Macro F1 Micro F1 Accuracy Hamming Loss	16.08% 20.53% 13.31% 0.0577	7.59% 10.31% 6.08% 0.0720

🚀 Como utilizar o modelo (vLLM ou Transformers)

O modelo responde a prompts estruturados com instruções do sistema em português. Exemplo real de inferência estruturada:

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "Assaoka/Tucano2-qwen-0.5b-Merge-Brighter"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")

messages = [
    {"role": "system", "content": "Você é um detector de emoções em português. Analise o texto fornecido e retorne a lista de emoções detectadas (pode ser vazia). As opções válidas são: 'raiva', 'nojo', 'medo', 'alegria', 'tristeza', 'surpresa'. Sua tarefa é retornar um JSON válido no seguinte formato: {"emocoes": ["emocao1", "emocao2", ..., "emocaoN"]} baseado na lista de emoções permitidas. Responda APENAS com o JSON válido."},
    {"role": "user", "content": "Texto: Fiquei extremamente feliz com a aprovação da minha pesquisa, mas confesso que a ansiedade me deu um pouco de medo."}
]

prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
# Resposta esperada: {"emocoes": ["alegria", "medo"]}