UniLu/IF_multi_constraints_upto5_SFT

该监督数据集源自[allenai/IF_multi_constraints_upto5](https://huggingface.co/datasets/allenai/IF_multi_constraints_upto5)受限指令跟随数据集。响应由[google/gemma-4-31B-it](https://huggingface.co/google/gemma-4-31B-it)生成，然后经过过滤以保留高质量示例。最终数据集仅包含根据官方IFBench评估器获得的最低宽松分数为1.0和最低严格分数为0.8的样本。响应长度限制在1024个标记以内，所有超过此长度的响应均被丢弃。

This supervised dataset is derived from the [allenai/IF_multi_constraints_upto5](https://huggingface.co/datasets/allenai/IF_multi_constraints_upto5) constrained instruction-following dataset. Responses were generated with [google/gemma-4-31B-it](https://huggingface.co/google/gemma-4-31B-it) and then filtered to retain only high-quality examples. The final dataset includes only samples with a minimum loose score of `1.0` and a minimum strict score of `0.8` according to the official `IFBench` evaluator. Maximum response length was limited to `1024` tokens, all responses longer than the limit were discarded.