Instructions to use MoonHawlk/qwen3-0.6b-revisor-pt with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use MoonHawlk/qwen3-0.6b-revisor-pt with Transformers:

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

pipe = pipeline("text-generation", model="MoonHawlk/qwen3-0.6b-revisor-pt")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("MoonHawlk/qwen3-0.6b-revisor-pt")
model = AutoModelForCausalLM.from_pretrained("MoonHawlk/qwen3-0.6b-revisor-pt", 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 MoonHawlk/qwen3-0.6b-revisor-pt with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "MoonHawlk/qwen3-0.6b-revisor-pt"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "MoonHawlk/qwen3-0.6b-revisor-pt",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/MoonHawlk/qwen3-0.6b-revisor-pt

SGLang

How to use MoonHawlk/qwen3-0.6b-revisor-pt 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 "MoonHawlk/qwen3-0.6b-revisor-pt" \
    --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": "MoonHawlk/qwen3-0.6b-revisor-pt",
		"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 "MoonHawlk/qwen3-0.6b-revisor-pt" \
        --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": "MoonHawlk/qwen3-0.6b-revisor-pt",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Unsloth Desktop
Docker Model Runner
How to use MoonHawlk/qwen3-0.6b-revisor-pt with Docker Model Runner:
```
docker model run hf.co/MoonHawlk/qwen3-0.6b-revisor-pt
```

Qwen3-0.6B — Revisor gramatical (PT)

Fine-tune do Qwen/Qwen3-0.6B para classificar a qualidade gramatical de sentenças em português em Aceitável / Moderado / Inaceitável, com uma breve descrição do erro. Treinado com Unsloth (LoRA, 16-bit).

Resultado (test set, 500 exemplos)

Acurácia: 81.6% (0% de saídas inválidas)
Recall por classe: Aceitável 94.4% · Moderado 72.4% · Inaceitável 78.0%

Uso

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "SEU_USUARIO/qwen3-0.6b-revisor-pt"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(repo, torch_dtype=torch.bfloat16, device_map="auto")

SYSTEM = """Você é um revisor gramatical de textos em português. Avalie a qualidade da sentença considerando concordância (nominal e verbal), pontuação e acentuação, e classifique-a em uma destas categorias:
- "Aceitável": sem erros relevantes.
- "Moderado": um erro pontual.
- "Inaceitável": múltiplos erros ou um erro grave.
Responda APENAS com um JSON válido no formato: {"classificacao": "<categoria>", "descricao": "<breve descrição do(s) erro(s)>"}."""
msgs = [{"role": "system", "content": SYSTEM},
        {"role": "user", "content": 'Sentença: "Os menino jogou bola."'}]
prompt = tok.apply_chat_template(msgs, tokenize=False,
                                 add_generation_prompt=True, enable_thinking=False)
ids = tok(prompt, return_tensors="pt").to(model.device)
out = model.generate(**ids, max_new_tokens=128, do_sample=False)
print(tok.decode(out[0][ids.input_ids.shape[1]:], skip_special_tokens=True))
# -> {"classificacao": "Inaceitável", "descricao": "..."}

Treino

Base Qwen3-0.6B · LoRA r=16 · 3 épocas · ~5000 exemplos. Saída em JSON {"classificacao": ..., "descricao": ...}, com enable_thinking=False.

O adaptador LoRA isolado (~35 MB) está na subpasta adapter/ deste mesmo repositório, caso prefira carregá-lo sobre o Qwen3-0.6B base.

Limitações

Dataset sintético (erros injetados por regra sobre a Wikipedia PT: concordância, pontuação, acentuação). Erros de acentuação sutis podem passar. Uso experimental.

Downloads last month: 28

Safetensors

Model size

0.6B params

Tensor type

BF16

Model tree for MoonHawlk/qwen3-0.6b-revisor-pt

Base model

Qwen/Qwen3-0.6B-Base

Finetuned

Qwen/Qwen3-0.6B

Adapter

(573)

this model