⚠️ 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. — 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 The model’s computational behavior can be configured through the config.json file: — totalutsteps: Controls the number of recurre You can adjust this value to trade off between performance and computation time. — earlyexitthreshold: Controls the adaptive exit mechanism (default: 1.0). Lower values encourage earlier exit, while 1.0 means always use all steps. > Note: vLLM does not currently support the adaptive exit feature due to its inference optimization characteristics. When using vLLM,…
Модальности:
Генерация текста
Области применения:
Логика и рассуждение Диалог / чат
Задача: Генерация текста
Автор: scpalmetto
Теги: gguf, llama, looped-language-model, reasoning, recurrent-depth, thinking, chain-of-thought, conversational
Лайков: 3 | Загрузок: 279
Описание основано на материалах HuggingFace. Перевод выполнен автоматически.
