Matt300209/Ouro-1B-Instruct

TEXT GENERATIONConcurrency Cost:1Model Size:1.4BQuant:BF16Ctx Length:32kPublished:May 17, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

Matt300209/Ouro-1B-Instruct is a 1.4 billion parameter instruction-tuned causal language model developed by ByteDance, based on the Ouro-1.4B architecture. This model is specifically optimized for advanced mathematical and scientific reasoning tasks, featuring a 32K context length and a unique recurrent step mechanism for explicit thinking processes. It aims to provide competitive reasoning performance against larger 4B models while maintaining a compact size.

Loading preview...

Ouro-1.4B-Thinking: A Reasoning-Specialized LLM

Ouro-1.4B-Thinking, developed by ByteDance, is a 1.4 billion parameter language model specifically fine-tuned for advanced reasoning tasks. It is a variant of the Ouro-1.4B base model, enhanced through supervised fine-tuning on high-quality reasoning datasets. The model is designed for research purposes and emphasizes an explicit thinking process.

Key Capabilities & Features

  • Advanced Reasoning: Optimized for mathematical and scientific problem-solving.
  • Compact Efficiency: Achieves reasoning performance competitive with 4 billion parameter models despite its smaller size.
  • Cross-Step Consistency: Intermediate recurrent outputs are reliable indicators of final answers.
  • Explicit Thinking Process: Generates detailed reasoning steps, making its problem-solving transparent.
  • Configurable Recurrent Steps: Users can adjust the number of recurrent steps (total_ut_steps) and an adaptive exit mechanism (early_exit_threshold) via config.json to balance performance and computation.

Training Details

The model underwent supervised fine-tuning on approximately 8.3 million examples, including 3.5M mathematics, 3.2M code, 808K science, and 767K chat examples. It was trained for 2 epochs with a maximum sequence length of 32K tokens.

When to Use This Model

This model is particularly well-suited for applications requiring strong logical deduction and step-by-step problem-solving, especially in mathematical and scientific domains, where explicit reasoning paths are beneficial. Its compact size makes it efficient for deployment where computational resources are a consideration.