README.md

poems.jsonl:
初始数据集，格式如下，约21万条
{"dynasty": "宋代", "author": "李弥逊", "title": "题大儒寮小阁", "content": "青鞋踏尽剑鋩山，借枕僧房落照间。\n高屋凭虚听泉语，岭云应似我身闲。"}
{"dynasty": "宋代", "author": "刘克庄", "title": "古墓", "content": "龟趺但见碑讹缺，马鬣安知葬旧新。\n想见若敖求食鬼，亦羞东郭乞墦人。"}
{"dynasty": "宋代", "author": "赵蕃", "title": "从礼载酒要余及明叔游合普六首", "content": "尘埃老矣未收身，每对山林强自亲。\n诗纵满囊惭李贺，笔虽成冢愧藏真。"}
{"dynasty": "清代", "author": "彭绩", "title": "戏简族子葆元", "content": "观君如有得，意色尚孩重。身隐南园内，家贫五字中。新年从幼识，旧姓此生同。何不时来过，门前道路通。"}
{"dynasty": "宋代", "author": "刘克庄", "title": "喜雨五首", "content": "昕朝避黄屋，露祷动苍穹。\n野老安知帝，豚蹄赛社公。"}

poems_classification.jsonl:
处理后的简单分类对话数据集，格式如下，约21万条
{"conversations": [{"role": "user", "content": "青鞋踏尽剑鋩山，借枕僧房落照间。高屋凭虚听泉语，岭云应似我身闲。"}, {"role": "assistant", "content": "宋代"}]}
{"conversations": [{"role": "user", "content": "龟趺但见碑讹缺，马鬣安知葬旧新。想见若敖求食鬼，亦羞东郭乞墦人。"}, {"role": "assistant", "content": "宋代"}]}
{"conversations": [{"role": "user", "content": "尘埃老矣未收身，每对山林强自亲。诗纵满囊惭李贺，笔虽成冢愧藏真。"}, {"role": "assistant", "content": "宋代"}]}
{"conversations": [{"role": "user", "content": "观君如有得，意色尚孩重。身隐南园内，家贫五字中。新年从幼识，旧姓此生同。何不时来过，门前道路通。"}, {"role": "assistant", "content": "清代"}]}
{"conversations": [{"role": "user", "content": "昕朝避黄屋，露祷动苍穹。野老安知帝，豚蹄赛社公。"}, {"role": "assistant", "content": "宋代"}]}
{"conversations": [{"role": "user", "content": "大雅清风久不闻，一杯聊为洗荒坟。朱弦三叹无今古，说与江东日暮云。"}, {"role": "assistant", "content": "元代"}]}

poems_classification_512.pth:
batchsize=128 epochs=2 lr=1e-3训练后的基础分类lora模型权重