Ouro-2.6B-Thinking-IBNN

<p align="center"> <svg xmlns="http://www.w3.org/2000/svg" width="850" height="220" viewBox="0 0 850 220"> <defs> <linearGradient id="edgyGradient" x1="0%" y1="0%" x2="100%" y2="0%"> <stop offset="0%" stop-color="#000000" stop-opacity="0"/> <stop offset="15%" stop-color="#000000" stop-opacity="0"/> <stop offset="25%" stop-color="#FF00FF" stop-opacity="1"/> <stop offset="35%" stop-color="#FF0033" stop-opacity="1"/> <stop offset="45%" stop-color="#FFFFFF" stop-opacity="1"/> <stop offset="55%" stop-color="#00FFFF" stop-opacity="1"/> <stop offset="65%" stop-color="#FF00FF" stop-opacity="1"/> <stop offset="75%" stop-color="#000000" stop-opacity="0"/> <stop offset="100%" stop-color="#000000" stop-opacity="0"/> <animateTransform attributeName="gradientTransform" type="translate" values="-1 0; 1 0" dur="3s" repeatCount="indefinite"/> <animateTransform attributeName="gradientTransform" type="rotate" additive="sum" values="0 0.5 0.5; 90 0.5 0.5; 0 0.5 0.5" keyTimes="0; 0.5; 1" dur="12s" repeatCount="indefinite" calcMode="spline" keySplines="0.42 0 0.58 1; 0.42 0 0.58 1"/> </linearGradient> <linearGradient id="subtitleGradient" x1="0%" y1="0%" x2="100%" y2="0%"> <stop offset="0%" stop-color="#00FFFF" stop-opacity="0"/> <stop offset="40%" stop-color="#FFFFFF" stop-opacity="0"/> <stop offset="50%" stop-color="#FFFFFF" stop-opacity="1"/> <stop offset="60%" stop-color="#00FFFF" stop-opacity="0"/> <stop offset="100%" stop-color="#00FFFF" stop-opacity="0"/> <animateTransform attributeName="gradientTransform" type="translate" values="-1 0; 1 0" dur="3s" repeatCount="indefinite"/> <animateTransform attributeName="gradientTransform" type="rotate" additive="sum" values="0 0.5 0.5; 90 0.5 0.5; 0 0.5 0.5" keyTimes="0; 0.5; 1" dur="12s" repeatCount="indefinite" calcMode="spline" keySplines="0.42 0 0.58 1; 0.42 0 0.58 1"/> </linearGradient> <filter id="fluidGlitch" x="-15%" y="-15%" width="130%" height="130%"> <feTurbulence type="fractalNoise" baseFrequency="0.02 0.0133" numOctaves="3" seed="42" result="fluidNoise"/> <feDisplacementMap in="SourceGraphic" in2="fluidNoise" scale="1" xChannelSelector="R" yChannelSelector="G" result="fluidDisplaced"/> <feTurbulence type="fractalNoise" baseFrequency="0.1 0.0133" numOctaves="1" seed="5" result="glitchNoise"/> <feDisplacementMap in="fluidDisplaced" in2="glitchNoise" scale="1" xChannelSelector="R" yChannelSelector="G"/> </filter> </defs> <g filter="url(#fluidGlitch)"> <text x="50%" y="90" text-anchor="middle" font-family="Consolas, Monaco, 'Courier New', monospace" font-size="40" font-weight="700" letter-spacing="2" fill="none" stroke="url(#edgyGradient)" stroke-width="0.8">OURO-2.6B-THINKING-IBNN <animateTransform attributeName="transform" type="translate" additive="sum" values="0 0; 4 1; 0 4; -4 1; 0 0" keyTimes="0; 0.25; 0.5; 0.75; 1" dur="6s" repeatCount="indefinite" calcMode="spline" keySplines="0.42 0 0.58 1; 0.42 0 0.58 1; 0.42 0 0.58 1; 0.42 0 0.58 1"/> </text> </g> </svg> </p> <p align="center"> <img src="assets/logo.png" align="center" alt="Ouro Logo" width="1024" height="1024"/> </p> <div align="center"> <img src="assets/banner.svg" alt="Banner" /> </div> <details> <summary><b>Copy</b></summary>

import torch, time, gc, sys
from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig

assert torch.cuda.is_available()
print(f"GPU  : {torch.cuda.get_device_name(0)}")
print(f"VRAM : {torch.cuda.get_device_properties(0).total_memory/1e9:.1f} GB")

torch.backends.cuda.matmul.allow_tf32 = True
torch.backends.cudnn.allow_tf32 = True

tok = AutoTokenizer.from_pretrained("AXIOM-TECH/Ouro-2.6B-Thinking-IBNN", trust_remote_code=True)

model = AutoModelForCausalLM.from_pretrained(
    "AXIOM-TECH/Ouro-2.6B-Thinking-IBNN",
    device_map="auto",
    torch_dtype=torch.float16,
    trust_remote_code=True,
    attn_implementation="sdpa",
)

# Configure custom architecture parameters
model.config.total_ut_steps = 4
model.config.early_exit_threshold = 1.0
model.config.use_cache = True
model.generation_config = GenerationConfig.from_model_config(model.config)
model.generation_config.use_cache = True
model.eval()
model = torch.compile(model, mode="max-autotune")

print(f"VRAM used: {torch.cuda.memory_allocated()/1e9:.2f} GB")

print("Model ready")

</details> <div align="center"> <img src="assets/banner2.svg" alt="Banner" /> </div>

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**: [Updating the standard neuron model in artificial neural networks](https://arxiv.org/pdf/2605.30370)
- **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)

---