Instructions to use prefeitura-rio/Rio-2.5-Open with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use prefeitura-rio/Rio-2.5-Open with Transformers:

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

pipe = pipeline("text-generation", model="prefeitura-rio/Rio-2.5-Open")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("prefeitura-rio/Rio-2.5-Open")
model = AutoModelForCausalLM.from_pretrained("prefeitura-rio/Rio-2.5-Open", 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 prefeitura-rio/Rio-2.5-Open with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "prefeitura-rio/Rio-2.5-Open"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "prefeitura-rio/Rio-2.5-Open",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/prefeitura-rio/Rio-2.5-Open

SGLang

How to use prefeitura-rio/Rio-2.5-Open 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 "prefeitura-rio/Rio-2.5-Open" \
    --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": "prefeitura-rio/Rio-2.5-Open",
		"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 "prefeitura-rio/Rio-2.5-Open" \
        --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": "prefeitura-rio/Rio-2.5-Open",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use prefeitura-rio/Rio-2.5-Open with Docker Model Runner:
```
docker model run hf.co/prefeitura-rio/Rio-2.5-Open
```

wow, what dataset do you use to train?

by Roman1111111 - opened Feb 22

Discussion

Roman1111111

Feb 22

this is really good model and high improvements over the baze are commendable , what dataset do you use, and how much rows or tokens was that model trained on? thanks for the model and your work

Sangu1nius

Feb 22

Hi there! Thank you! For this model specifically, we used https://huggingface.co/datasets/nvidia/OpenScienceReasoning-2 and https://huggingface.co/datasets/nvidia/Nemotron-Post-Training-Dataset-v1. More specifically, we used On Policy Distillation with reasoning traces from our larger, closed source Rio 3 model, which we'll be announcing soon. More details and an interactive demo at https://ia.rio/

Roman1111111

Feb 23

oh, that helpful, thanks, will you open source reasoning traces ? of Rio 3 model

Sangu1nius

Feb 23

I'm sorry. Right now, we are keeping the Rio 3 reasoning traces closed-source and have no immediate plans to release them. However, I do suppose that, for pretty much all purposes, the reasoning traces you can extract from https://huggingface.co/prefeitura-rio/Rio-3.0-Open will do the trick. It's almost as good as the main model (like 4% lower benchmarks), much cheaper, and has the exact same reasoning style. If you want to use that, it's fair game!

Roman1111111

Feb 23

But could you please provide sample example of the traces, as i couldn't really use or deploy rio 3 open due to hardware limitations. What if i make a project with you, i have an idea of glm 4.7 flash fine tune on gemini 3 flash as a teacher model(i prepared the dataset already) and this, may i please use it 🙏🙏🙏(i will be really careful and private with this)

Sangu1nius

Feb 23

We can't really do that right now. However, we'll consider including a dataset release for Rio 3.5 Open, when that comes, at a low enough volume such that open source small creators like yourself benefit, but large labs can't properly distill a big model on those traces. Sounds fair?

Roman1111111

Feb 24

yes, thanks ❤️

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment