Instructions to use ClassiCC-Corpus/Curio-1.1b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ClassiCC-Corpus/Curio-1.1b with Transformers:

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

pipe = pipeline("text-generation", model="ClassiCC-Corpus/Curio-1.1b")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("ClassiCC-Corpus/Curio-1.1b")
model = AutoModelForCausalLM.from_pretrained("ClassiCC-Corpus/Curio-1.1b")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use ClassiCC-Corpus/Curio-1.1b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ClassiCC-Corpus/Curio-1.1b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ClassiCC-Corpus/Curio-1.1b",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/ClassiCC-Corpus/Curio-1.1b

SGLang

How to use ClassiCC-Corpus/Curio-1.1b 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 "ClassiCC-Corpus/Curio-1.1b" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ClassiCC-Corpus/Curio-1.1b",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "ClassiCC-Corpus/Curio-1.1b" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ClassiCC-Corpus/Curio-1.1b",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use ClassiCC-Corpus/Curio-1.1b with Docker Model Runner:
```
docker model run hf.co/ClassiCC-Corpus/Curio-1.1b
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

🐦 Curió 1.1B

📖 Overview

Curió 1.1B is a Portuguese-adapted language model created via continued pretraining of TinyLlama 1.1B (1T), originally trained on 1 trillion English tokens, on 150B Portuguese tokens from the ClassiCC-PT corpus.

This model was designed to explore the impact of language-specific corpora on adapting an English-trained base model to Portuguese, yielding performance improvements on Portuguese benchmarks without large-scale retraining from scratch.

🏗 Training Setup

Base model: TinyLlama 1.1B (LLaMA-2 architecture)
Parameters: 1.1B
Continued pretraining tokens: 150B (ClassiCC-PT)
Sequence length: 4096 tokens (with packing)
Hardware: TPU v2-128 (thanks to Google TRC program)
Frameworks: T5X

📊 Evaluation

Evaluated on the Poeta benchmark — 14 diverse Portuguese tasks (RTE, STS, MCQ exams, sentiment analysis, QA, etc.) — using the Normalized Preferred Metric (NPM).

Model	Training Regimen	Poeta v2 NPM
TinyLlama 1T (EN)	–	17.4
TinyLlama 2T (EN)	+1T EN continued pretraining	20.9
training with mC4-PT	+150B PT (mC4-PT) continued pretraining	~20
training with ClueWeb-22-PT	+150B PT (Clueweb-22-PT) continued pretraining	~27
Curió 1.1B	+150B PT (ClassiCC-PT) continued pretraining	27.1

📥 Usage

Please note that Curio 1.1B has not trained to be used as a chat model

from transformers import AutoTokenizer, AutoModelForCausalLM

model_name = "ClassiCC-Corpus/Curio-1.1B"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

📜 Citation

If you use Curió 1.1B, please cite:

Coming soon

Acknowledgements

We thank the google TRC program, which generously granted us the necessary resources for the development of this research.

Downloads last month: 11

Safetensors

Model size

1B params

Tensor type

F32

Model tree for ClassiCC-Corpus/Curio-1.1b

Quantizations

2 models

Collection including ClassiCC-Corpus/Curio-1.1b

Curio 1.1B

Collection

3 items • Updated Aug 9, 2025