Instructions to use SynastriaNetworks/Uirapuru-U1.1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use SynastriaNetworks/Uirapuru-U1.1 with Transformers:

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

pipe = pipeline("text-generation", model="SynastriaNetworks/Uirapuru-U1.1")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("SynastriaNetworks/Uirapuru-U1.1")
model = AutoModelForMultimodalLM.from_pretrained("SynastriaNetworks/Uirapuru-U1.1", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.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(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use SynastriaNetworks/Uirapuru-U1.1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "SynastriaNetworks/Uirapuru-U1.1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SynastriaNetworks/Uirapuru-U1.1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/SynastriaNetworks/Uirapuru-U1.1

SGLang

How to use SynastriaNetworks/Uirapuru-U1.1 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 "SynastriaNetworks/Uirapuru-U1.1" \
    --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": "SynastriaNetworks/Uirapuru-U1.1",
		"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 "SynastriaNetworks/Uirapuru-U1.1" \
        --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": "SynastriaNetworks/Uirapuru-U1.1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use SynastriaNetworks/Uirapuru-U1.1 with Docker Model Runner:
```
docker model run hf.co/SynastriaNetworks/Uirapuru-U1.1
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Uirapuru Logo

A hotfix release of the Uirapuru U1 language model.

Model Overview

Uirapuru U1.1 is an incremental hotfix release of Uirapuru U1, focused on improving the model's identity, reasoning behavior, learned capabilities, Brazilian Portuguese fluency, and agentic task performance.

U1.1 was retrained without adaptive thinking, using a higher LoRA rank to provide additional capacity for learning behavioral and capability-level changes.

Rather than being a completely new generation, U1.1 is intended to be a refined and more capable iteration of U1.

Key Improvements

Compared to Uirapuru U1, U1.1 focuses on:

Stronger identity — a more consistent Uirapuru identity and behavioral profile.
Reasoning style — improved and more consistent reasoning patterns.
New capabilities — additional skills learned during retraining.
Informal Brazilian Portuguese — more natural conversational Portuguese, including informal and everyday language.
Agentic tasks — improved ability to follow objectives, handle multi-step instructions, and perform task-oriented workflows.
Higher LoRA rank — increased adapter capacity for capturing behavioral and capability changes.
No adaptive thinking — the model was retrained without the adaptive-thinking mechanism used in previous experiments.

Thinking Behavior

Uirapuru U1.1 does not use adaptive thinking.

The model was retrained to incorporate its reasoning behavior directly into its learned behavior rather than relying on a separate adaptive mechanism to dynamically control thinking.

This does not mean that the model cannot reason. Instead, the goal is to provide a more consistent reasoning behavior across different types of prompts.

Training

Uirapuru U1.1 was produced through a new training cycle based on Uirapuru U1.

A major change in this release was the use of a higher LoRA rank.

The increased rank provides the adapter with more representational capacity, allowing it to capture a broader range of changes during training.

The training focused particularly on:

Uirapuru's identity.
Reasoning and problem-solving behavior.
New capabilities.
Informal Brazilian Portuguese.
Agentic and multi-step task behavior.

Intended Use

Uirapuru U1.1 is intended for applications such as:

AI assistants
Conversational applications
Brazilian Portuguese applications
Agentic AI systems
Task automation
LLM experimentation
Research and prototyping

Expected Behavior

Uirapuru U1.1 is designed to provide:

More natural Brazilian Portuguese conversations.
Better handling of informal language.
A more consistent model identity.
Improved instruction following.
Better performance on multi-step tasks.
Improved behavior for agentic workflows.
More consistent reasoning without adaptive thinking.

Limitations

Uirapuru U1.1 is still an experimental model and may:

Generate incorrect or fabricated information.
Make reasoning mistakes.
Misinterpret instructions.
Produce inconsistent results on complex tasks.
Fail to correctly complete agentic workflows.
Perform differently depending on the prompt and context.
Have reduced performance in languages other than Portuguese and English.

A higher LoRA rank does not guarantee that every trained behavior or capability will generalize perfectly to unseen situations.

Safety

Uirapuru U1.1 should not be assumed to be perfectly safe or reliable.

Applications using the model should implement appropriate safeguards, validation, and human oversight depending on the use case.

The model should not be relied upon as the sole source of truth for high-impact decisions.

Version Summary

Uirapuru U1.1 = Uirapuru U1 + retraining + higher-rank LoRA + stronger identity + improved reasoning behavior + new capabilities + better informal Brazilian Portuguese + improved agentic behavior.

U1.1 is intended as a hotfix and evolutionary update, rather than a completely new model generation.

Credits

Developed by SynastrIA Networks.

Uirapuru is part of the SynastrIA model family.

Downloads last month: 338

Safetensors

Model size

10B params

Tensor type

BF16

F32

Model tree for SynastriaNetworks/Uirapuru-U1.1

Base model

Qwen/Qwen3.5-9B-Base

Finetuned

SynastriaNetworks/Uirapuru-U1-9B

Finetuned

(1)

this model

Quantizations

2 models