-
Notifications
You must be signed in to change notification settings - Fork 37
Expand file tree
/
Copy pathquiz.json
More file actions
90 lines (90 loc) · 3.32 KB
/
Copy pathquiz.json
File metadata and controls
90 lines (90 loc) · 3.32 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
{
"lesson": "09-hybrid-memory-mem0",
"title": "混合记忆:向量 + 图 + KV(Mem0)",
"questions": [
{
"stage": "pre",
"question": "KV 存储最擅长处理哪类查询?",
"options": [
"对长对话的语义相似度",
"以(user, type, entity)为键的直接事实查找",
"共享某个计费实体的客户之间的可达性",
"「某时刻有效」的时间性查询"
],
"correct": 1,
"explanation": "KV 在精确键上是 O(1) 的;向量用于相似度,图用于关系。"
},
{
"stage": "pre",
"question": "Mem0 在每次 add() 时并行写入的三个存储是什么?",
"options": [
"向量、KV、图",
"Cache、queue、log",
"Postgres、Redis、ClickHouse",
"Embedding、attention、FFN"
],
"correct": 0,
"explanation": "Mem0 把每次写入扇出到向量、KV 和图存储。"
},
{
"stage": "check",
"question": "哪三个维度喂给 Mem0 的融合分数(fusion score)?",
"options": [
"相关性、重要性、新近性",
"精确率、召回率、F1",
"延迟、吞吐量、成本",
"置信度、困惑度、BLEU"
],
"correct": 0,
"explanation": "分数是相关性、重要性和新近性的加权和;权重按产品调优。"
},
{
"stage": "check",
"question": "当一个新进来的事实与现有的边矛盾时,Mem0g 会怎么做?",
"options": [
"删除这条边",
"把现有的边标记为无效但不删除,使时间性查询仍能遍历",
"抛出异常",
"重写 user_id"
],
"correct": 1,
"explanation": "软失效(soft invalidation)保留了历史,以便进行时间性(某时刻有效)查询。"
},
{
"stage": "check",
"question": "为什么本课建议按产品调优融合权重?",
"options": [
"提供商要求这样",
"对聊天 agent 新近性占主导,对合规 agent 重要性占主导,对检索 agent 相关性占主导",
"向量库拒绝相等的权重",
"Apache 2.0 要求这样"
],
"correct": 1,
"explanation": "不同产品对相关性/重要性/新近性想要不同的偏置;一套权重并不适合所有情况。"
},
{
"stage": "post",
"question": "Mem0 使用的作用域分类(scope taxonomy)是什么?",
"options": [
"Public、private、secret",
"User、session、agent",
"Local、regional、global",
"Read、write、admin"
],
"correct": 1,
"explanation": "作用域为 user(跨会话)、session(单个线程)、agent(每实例状态)。"
},
{
"stage": "post",
"question": "在此模式中,embedding 漂移(embedding drift)是什么,本课建议如何缓解?",
"options": [
"向量被加密;轮换密钥",
"随着语料增长,向量检索质量下降;定期对使用最多的 top-N 记录重新做 embedding",
"embedding 整数溢出;改用 float64",
"embedding API 更换了 URL;固定一个域名"
],
"correct": 1,
"explanation": "定期对热记录重新做 embedding,能在语料增长时保持检索质量稳定。"
}
]
}