reasoning-degeneration-dev/TIW-qwen3-4B-Instruct-2507-cd5arg-budget_forcing

这是一个倒计时推理任务的数据集，采用budget_forcing策略，目标token数为32384。数据集包含100个样本，每个样本有12个特征列，包括问题陈述、元数据、任务来源、格式化提示、响应样本、token计数、轮次计数、继续提示、策略类型、答案样本、评估样本和评估元数据。实验使用了Qwen3-4B-Instruct-2507模型，温度参数为0.7，每轮最大token数为32384，目标总token数为32384，样本数为3。

This is a countdown reasoning task dataset using the budget_forcing strategy with a target of 32384 tokens. The dataset contains 100 samples with 12 feature columns including question statements, metadata, task source, formatted prompts, response samples, token counts, round counts, continuation prompts, strategy type, answer samples, evaluation samples, and evaluation metadata. The experiment used the Qwen3-4B-Instruct-2507 model with a temperature parameter of 0.7, max tokens per round of 32384, target total tokens of 32384, and 3 samples.