Instructions to use scpalmetto/Ouro-2.6B-Thinking-Fixed with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use scpalmetto/Ouro-2.6B-Thinking-Fixed with Transformers:

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

pipe = pipeline("text-generation", model="scpalmetto/Ouro-2.6B-Thinking-Fixed", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("scpalmetto/Ouro-2.6B-Thinking-Fixed", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("scpalmetto/Ouro-2.6B-Thinking-Fixed", trust_remote_code=True)
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]:]))

llama-cpp-python

How to use scpalmetto/Ouro-2.6B-Thinking-Fixed with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="scpalmetto/Ouro-2.6B-Thinking-Fixed",
	filename="ouro-2.6b-q4_k_m.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use scpalmetto/Ouro-2.6B-Thinking-Fixed with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf scpalmetto/Ouro-2.6B-Thinking-Fixed:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf scpalmetto/Ouro-2.6B-Thinking-Fixed:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf scpalmetto/Ouro-2.6B-Thinking-Fixed:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf scpalmetto/Ouro-2.6B-Thinking-Fixed:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf scpalmetto/Ouro-2.6B-Thinking-Fixed:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf scpalmetto/Ouro-2.6B-Thinking-Fixed:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf scpalmetto/Ouro-2.6B-Thinking-Fixed:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf scpalmetto/Ouro-2.6B-Thinking-Fixed:Q4_K_M

Use Docker

docker model run hf.co/scpalmetto/Ouro-2.6B-Thinking-Fixed:Q4_K_M

LM Studio
Jan

vLLM

How to use scpalmetto/Ouro-2.6B-Thinking-Fixed with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "scpalmetto/Ouro-2.6B-Thinking-Fixed"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "scpalmetto/Ouro-2.6B-Thinking-Fixed",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/scpalmetto/Ouro-2.6B-Thinking-Fixed:Q4_K_M

SGLang

How to use scpalmetto/Ouro-2.6B-Thinking-Fixed 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 "scpalmetto/Ouro-2.6B-Thinking-Fixed" \
    --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": "scpalmetto/Ouro-2.6B-Thinking-Fixed",
		"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 "scpalmetto/Ouro-2.6B-Thinking-Fixed" \
        --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": "scpalmetto/Ouro-2.6B-Thinking-Fixed",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use scpalmetto/Ouro-2.6B-Thinking-Fixed with Ollama:
```
ollama run hf.co/scpalmetto/Ouro-2.6B-Thinking-Fixed:Q4_K_M
```

Unsloth Studio new

How to use scpalmetto/Ouro-2.6B-Thinking-Fixed with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for scpalmetto/Ouro-2.6B-Thinking-Fixed to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for scpalmetto/Ouro-2.6B-Thinking-Fixed to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for scpalmetto/Ouro-2.6B-Thinking-Fixed to start chatting

Docker Model Runner
How to use scpalmetto/Ouro-2.6B-Thinking-Fixed with Docker Model Runner:
```
docker model run hf.co/scpalmetto/Ouro-2.6B-Thinking-Fixed:Q4_K_M
```

Lemonade

How to use scpalmetto/Ouro-2.6B-Thinking-Fixed with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull scpalmetto/Ouro-2.6B-Thinking-Fixed:Q4_K_M

Run and chat with the model

lemonade run user.Ouro-2.6B-Thinking-Fixed-Q4_K_M

List all available models

lemonade list

Ouro-2.6B-Thinking

Model Description

**⚠️ IMPORTANT: This model is intended for research purposes only. It is provided as-is without warranties for production use. **

Ouro-2.6B-Thinking is a reasoning-specialized variant of the Ouro-2.6B base model, enhanced through supervised fine-tuning on high-quality reasoning data. Please use transformers==4.54.1for compatibility.

Key Features

Advanced Reasoning: Specifically optimized for mathematical and scientific reasoning tasks
Compact Size: Competitive with 4B models despite having only 2.6B parameters
Cross-Step Consistency: Intermediate recurrent outputs can serve as reliable proxies for final answers
Explicit Thinking Process: Trained to generate detailed reasoning steps

Configuration

Recurrent Steps and Adaptive Exit

The model's computational behavior can be configured through the config.json file:

{
  "total_ut_steps": 4,
  "early_exit_threshold": 1.0
}

total_ut_steps: Controls the number of recurrent steps (default: 4). You can adjust this value to trade off between performance and computation time.
early_exit_threshold: Controls the adaptive exit mechanism (default: 1.0). Lower values encourage earlier exit, while 1.0 means always use all steps.

Example: Modify recurrent steps

from transformers import AutoConfig, AutoModelForCausalLM

config = AutoConfig.from_pretrained("ByteDance/Ouro-2.6B-Thinking")
config.total_ut_steps = 3  # Use 3 recurrent steps instead of 4
model = AutoModelForCausalLM.from_pretrained(
    "ByteDance/Ouro-2.6B-Thinking",
    config=config,
    device_map="auto"
)

Note: vLLM does not currently support the adaptive exit feature due to its inference optimization characteristics. When using vLLM, the model will always execute the full number of total_ut_steps.

Model Architecture

Based on Ouro-2.6B with additional reasoning fine-tuning:

Configuration	Value
Parameters	2.6B
Layers	24
Recurrent Steps	4
Hidden Size	2048
Attention Heads	Multi-Head Attention (MHA)
FFN Activation	SwiGLU
Position Embedding	RoPE
Vocabulary Size	49,152
Context Length	32K (SFT)
Normalization	Sandwich RMSNorm

Training Details

Pre-training

Training Tokens: 7.7T tokens across 4 stages
Base Architecture: Ouro-2.6B

Supervised Fine-Tuning

Data Size: ~8.3M examples
Data Composition:
- Mathematics: 3.5M examples (OpenThoughts3, AceReason-1.1-SFT)
- Code: 3.2M examples (AceReason, OpenCodeReasoning, Llama-Nemotron, OpenThoughts3)
- Science: 808K examples (OpenThoughts3, Llama-Nemotron)
- Chat: 767K examples (DeepWriting-20K)
Training: 2 epochs, max sequence length 32K
Optimizer: Adam (lr=2×10⁻⁵, β=(0.9, 0.95))
Scheduler: Cosine decay

Quick Start

⚠️ IMPORTANT: Please use transformers<4.56.0 to avoid compatibility issues. We recommend transformers==4.54.1 or earlier versions.

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "Bytedance/Ouro-2.6B-Thinking"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    device_map="auto",
    torch_dtype="auto"
)

# Generate with reasoning
messages = [
    {"role": "user", "content": "Solve: If 2x + 3 = 11, what is x?"}
]
inputs = tokenizer.apply_chat_template(
    messages,
    tokenize=True,
    add_generation_prompt=True,
    return_tensors="pt"
).to(model.device)

outputs = model.generate(inputs, max_new_tokens=512, temperature=1.0, top_p=0.7)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Acknowledgments

We thank @Antizana for the KV cache fix merged from ouro-cache-fix, which resolved a critical compatibility issue with transformers>=4.56.0.

Citation

@article{zhu2025scaling,
  title={Scaling Latent Reasoning via Looped Language Models},
  author={Zhu, Rui-Jie and Wang, Zixuan and Hua, Kai and Zhang, Tianyu and Li, Ziniu and Que, Haoran and Wei, Boyi and Wen, Zixin and Yin, Fan and Xing, He and others},
  journal={arXiv preprint arXiv:2510.25741},
  year={2025}
}


## License

This model is licensed under Apache-2.0. See the LICENSE file for details.

## Project Links

- **Paper**: [Scaling Latent Reasoning via Looped Language Models](https://huggingface.co/papers/2510.25741)
- **Project Page**: [https://ouro-llm.github.io](https://ouro-llm.github.io)
- **Code**: [https://github.com/ByteDance/Ouro](https://github.com/ByteDance/Ouro)

---

Downloads last month: 1,014

GGUF

Model size

3B params

Architecture

llama

Hardware compatibility

4-bit

8-bit

Paper for scpalmetto/Ouro-2.6B-Thinking-Fixed

Scaling Latent Reasoning via Looped Language Models

Paper • 2510.25741 • Published Oct 29, 2025 • 230