- 引入 MemoryRouterXL 与 v5/v6 流式多线程训练/编码管线 - 修复 prepare_memory_router_dataset 候选池重建缺陷(mega 家族 3568x 加速,输出逐字节相同) - 修复 v5 被破坏的拒答与多跳标签(train 未知样本 319 -> 16319,multi_hop 平均正例 1.00 -> 2.00) - 同存储预算下 V2-128 v6 逐轴 22/22 通过:Top-1 41.12% -> 94.62%,未知拒答 0.00% -> 100.00% - 记录三条被实测推翻的显然优化(logits_to_keep=1 反而慢 55%、XL 容量未带来收益) - 记忆手术跨架构可移植性 14/14,读写关闭时与原生模型逐位相同
351 lines
7.4 KiB
JSON
351 lines
7.4 KiB
JSON
C:\Users\Administrator\miniconda3\envs\LLM\Lib\site-packages\torch\cuda\__init__.py:63: FutureWarning: The pynvml package is deprecated. Please install nvidia-ml-py instead. If you did not install pynvml directly, please report this to the maintainers of the package that installed pynvml for you.
|
|
import pynvml # type: ignore[import]
|
|
{
|
|
"attributes_without_usable_paraphrase": [
|
|
"入职年份",
|
|
"常用编辑器",
|
|
"手机尾号",
|
|
"紧急联系人姓氏",
|
|
"设备型号",
|
|
"项目代号",
|
|
"默认语言"
|
|
],
|
|
"dropped_variants": [
|
|
{
|
|
"attribute": "常住城市",
|
|
"paraphrase": "我平时待得最久的地儿是哪里?",
|
|
"blocked_by": [
|
|
"起床时间"
|
|
],
|
|
"shared": {
|
|
"起床时间": [
|
|
"时"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "出生城市",
|
|
"paraphrase": "我小时候成长的地儿是哪里?",
|
|
"blocked_by": [
|
|
"起床时间"
|
|
],
|
|
"shared": {
|
|
"起床时间": [
|
|
"时"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "档案标识",
|
|
"paraphrase": "系统派给我的那串字符是什么?",
|
|
"blocked_by": [
|
|
"紧急联系人姓氏"
|
|
],
|
|
"shared": {
|
|
"紧急联系人姓氏": [
|
|
"系"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "常用编辑器",
|
|
"paraphrase": "我写代码靠什么?",
|
|
"blocked_by": [
|
|
"项目代号"
|
|
],
|
|
"shared": {
|
|
"项目代号": [
|
|
"代"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "常用编辑器",
|
|
"paraphrase": "我做开发时靠什么?",
|
|
"blocked_by": [
|
|
"起床时间"
|
|
],
|
|
"shared": {
|
|
"起床时间": [
|
|
"时"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "默认语言",
|
|
"paraphrase": "我平时讲哪种话?",
|
|
"blocked_by": [
|
|
"起床时间"
|
|
],
|
|
"shared": {
|
|
"起床时间": [
|
|
"时"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "默认语言",
|
|
"paraphrase": "我跟人交流用哪门话?",
|
|
"blocked_by": [
|
|
"常用编辑器",
|
|
"紧急联系人姓氏"
|
|
],
|
|
"shared": {
|
|
"常用编辑器": [
|
|
"用"
|
|
],
|
|
"紧急联系人姓氏": [
|
|
"人"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "通勤方式",
|
|
"paraphrase": "我早上怎么去公司?",
|
|
"blocked_by": [
|
|
"办公城市"
|
|
],
|
|
"shared": {
|
|
"办公城市": [
|
|
"公"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "工位楼层",
|
|
"paraphrase": "我坐在哪儿,第几间?",
|
|
"blocked_by": [
|
|
"起床时间"
|
|
],
|
|
"shared": {
|
|
"起床时间": [
|
|
"间"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "邮箱域名",
|
|
"paraphrase": "别人给我发信要写哪个后缀?",
|
|
"blocked_by": [
|
|
"紧急联系人姓氏"
|
|
],
|
|
"shared": {
|
|
"紧急联系人姓氏": [
|
|
"人"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "手机尾号",
|
|
"paraphrase": "我随身那台设备的末几位数字是啥?",
|
|
"blocked_by": [
|
|
"工位楼层",
|
|
"设备型号"
|
|
],
|
|
"shared": {
|
|
"工位楼层": [
|
|
"位"
|
|
],
|
|
"设备型号": [
|
|
"备",
|
|
"设"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "手机尾号",
|
|
"paraphrase": "那串数字的末尾几位是啥?",
|
|
"blocked_by": [
|
|
"工位楼层",
|
|
"手机尾号"
|
|
],
|
|
"shared": {
|
|
"工位楼层": [
|
|
"位"
|
|
],
|
|
"手机尾号": [
|
|
"尾"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "项目代号",
|
|
"paraphrase": "我正在做的那个工程叫什么?",
|
|
"blocked_by": [
|
|
"工位楼层",
|
|
"课程名称",
|
|
"阅读工具"
|
|
],
|
|
"shared": {
|
|
"工位楼层": [
|
|
"工"
|
|
],
|
|
"课程名称": [
|
|
"程"
|
|
],
|
|
"阅读工具": [
|
|
"工"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "项目代号",
|
|
"paraphrase": "我手上那摊活儿叫什么?",
|
|
"blocked_by": [
|
|
"手机尾号"
|
|
],
|
|
"shared": {
|
|
"手机尾号": [
|
|
"手"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "入职年份",
|
|
"paraphrase": "我什么时候开始在这儿干活的?",
|
|
"blocked_by": [
|
|
"起床时间"
|
|
],
|
|
"shared": {
|
|
"起床时间": [
|
|
"时"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "入职年份",
|
|
"paraphrase": "我从何时起在这家公司上班?",
|
|
"blocked_by": [
|
|
"办公城市",
|
|
"起床时间"
|
|
],
|
|
"shared": {
|
|
"办公城市": [
|
|
"公"
|
|
],
|
|
"起床时间": [
|
|
"时",
|
|
"起"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "紧急联系人姓氏",
|
|
"paraphrase": "出事时该喊谁来?",
|
|
"blocked_by": [
|
|
"出生城市",
|
|
"起床时间"
|
|
],
|
|
"shared": {
|
|
"出生城市": [
|
|
"出"
|
|
],
|
|
"起床时间": [
|
|
"时"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "紧急联系人姓氏",
|
|
"paraphrase": "万一出岔子该找哪一户?",
|
|
"blocked_by": [
|
|
"出生城市",
|
|
"客户名称"
|
|
],
|
|
"shared": {
|
|
"出生城市": [
|
|
"出"
|
|
],
|
|
"客户名称": [
|
|
"户"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "午餐偏好",
|
|
"paraphrase": "我白天那顿想尝什么味道?",
|
|
"blocked_by": [
|
|
"咖啡口味"
|
|
],
|
|
"shared": {
|
|
"咖啡口味": [
|
|
"味"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "运动习惯",
|
|
"paraphrase": "我平时怎么锻炼?",
|
|
"blocked_by": [
|
|
"起床时间"
|
|
],
|
|
"shared": {
|
|
"起床时间": [
|
|
"时"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "阅读工具",
|
|
"paraphrase": "我看书靠什么设备?",
|
|
"blocked_by": [
|
|
"设备型号"
|
|
],
|
|
"shared": {
|
|
"设备型号": [
|
|
"备",
|
|
"设"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "宿舍楼号",
|
|
"paraphrase": "我住在哪一栋?",
|
|
"blocked_by": [
|
|
"常住城市"
|
|
],
|
|
"shared": {
|
|
"常住城市": [
|
|
"住"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "客户名称",
|
|
"paraphrase": "我在跟哪家公司打交道?",
|
|
"blocked_by": [
|
|
"办公城市"
|
|
],
|
|
"shared": {
|
|
"办公城市": [
|
|
"公"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "设备型号",
|
|
"paraphrase": "我用的那台是哪种款?",
|
|
"blocked_by": [
|
|
"常用编辑器"
|
|
],
|
|
"shared": {
|
|
"常用编辑器": [
|
|
"用"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"attribute": "设备型号",
|
|
"paraphrase": "我拿的那台是哪种档?",
|
|
"blocked_by": [
|
|
"档案标识"
|
|
],
|
|
"shared": {
|
|
"档案标识": [
|
|
"档"
|
|
]
|
|
}
|
|
}
|
|
]
|
|
}
|