JET-1.5B is designed to improve the efficient reasoning of LLMs by training the base DeepSeek-Distill-Qwen-1.5B model with a reinforcement learning framework. Through this training, the model learns to generate high-quality reasoning steps while minimizing unnecessary computation and token usage. Our training pipeline is available on GitHub: Just-Enough-Think The repository contains scripts for: — RL-based fine-tuning — Evaluation and benchmarking python def buildJETchattemplate(question, tokenizer): systemprompt = ( «You are a helpful AI assistant. A conversation takes place between the User » «and the Assistant. The User asks a question, and the Assistant solves it.n» «Please help me solve this question. Wrap only the final answer in \boxed{}.» ) return tokenizer.applychattemplate( [ {«role»: «system», «content»: systemprompt}, {«role»: «user», «content»: question} ], tokenize=False, addgenerationprompt=True )
Модальности:
Генерация текста
Области применения:
Логика и рассуждение Диалог / чат
Задача: Генерация текста
Автор: JinyiHan
Теги: qwen2, causal-lm, reasoning, conversational, text-generation-inference, endpoints_compatible
Лайков: 4 | Загрузок: 20
Описание основано на материалах HuggingFace. Перевод выполнен автоматически.