Initial commit: Natural Memory Agent Lab:模型服务 + TypeScript agent + 记忆能力场景评测(冲突更新、跨会话回忆、遗忘撤回、多跳、未知拒答)
This commit is contained in:
@@ -0,0 +1,4 @@
|
||||
{"session":"cross_session_recall:off:s1","turn":0,"userMessage":"记一下:我的项目代号是 蓝鲸-47,部署区域是 ap-southeast-3。","assistantText":"已记录:项目代号 蓝鲸-47,部署区域 ap-southeast-3。","toolCalls":[{"name":"notes","args":{"action":"append","text":"项目代号:蓝鲸-47,部署区域:ap-southeast-3"}}],"toolResults":[{"name":"notes","result":"{\"content\":[{\"type\":\"text\",\"text\":\"appended\"}],\"details\":{\"path\":\"H:\\\\Memory\\\\agent_lab\\\\agent\\\\runs\\\\v2\\\\scratchpad\\\\off.txt\"}}"}],"wallSeconds":5.112,"diagnostics":{"turn":7,"kind":"chat","at":1789310564.9450462,"user_message":"","assistant_text":"已记录:项目代号 蓝鲸-47,部署区域 ap-southeast-3。","reasoning_chars":0,"router_decisions":[{"need_memory":false,"page_ids":[],"record_ids":[],"page_scores":[],"record_scores":[],"hop_count":0,"hop_trace":[],"confidence":0.0000051725423872994725,"stop_reason":"router_abstained","top_score":0,"score_margin":0,"evidence_score":0}],"prefix_tokens":0,"read_seconds":0.4424972999840975,"coverage":{"configured":true,"loaded":true,"bound_to_bank":true,"vocabulary_fraction":0.9,"head":{"path":"H:\\Memory\\V2_dpskw\\qwen3_5_4b_natural_memory_v2_1\\memory_attribute_head.pt","attributes":24,"vocabulary":["主管姓名","入职年份","出生城市","办公城市","午餐偏好","咖啡口味","团队名称","客户名称","宿舍楼号","工位楼层","常住城市","常用编辑器","手机尾号","档案标识","紧急联系人姓氏","设备型号","课程名称","起床时间","运动习惯","通勤方式","邮箱域名","阅读工具","项目代号","默认语言"],"coverage_stats":{"applicable":0,"bypassed":7}}},"no_evidence":false,"memory_mode":"off","memory_changed":false,"reset_triggered":false,"tool_calls":[],"overrides":{},"prompt_tokens":1834,"completion_tokens":24,"generation_seconds":1.875,"total_seconds":1.879,"snapshot":{"router_decisions":[{"need_memory":false,"page_ids":[],"record_ids":[],"page_scores":[],"record_scores":[],"hop_count":0,"hop_trace":[],"confidence":0.0000051725423872994725,"stop_reason":"router_abstained","top_score":0,"score_margin":0,"evidence_score":0}],"no_evidence":false,"prefix_tokens":0,"prefix_used":false,"read_seconds":0.4424972999840975,"auto_memory_probability":null,"auto_forget_probability":[0.011167768388986588],"context_compaction":{"compacted":false,"archived_records":0,"archived_tokens":0,"original_tokens":[1834],"retained_tokens":[1834],"hot_limit":32768,"archive_requested":true},"coverage":{"configured":true,"loaded":true,"bound_to_bank":true,"vocabulary_fraction":0.9,"head":{"path":"H:\\Memory\\V2_dpskw\\qwen3_5_4b_natural_memory_v2_1\\memory_attribute_head.pt","attributes":24,"vocabulary":["主管姓名","入职年份","出生城市","办公城市","午餐偏好","咖啡口味","团队名称","客户名称","宿舍楼号","工位楼层","常住城市","常用编辑器","手机尾号","档案标识","紧急联系人姓氏","设备型号","课程名称","起床时间","运动习惯","通勤方式","邮箱域名","阅读工具","项目代号","默认语言"],"coverage_stats":{"applicable":0,"bypassed":7}}},"memory":{"pages":0,"max_pages":32768,"page_capacity":32,"capacity_records":1048576,"open_pages":0,"records":0,"active_records":0,"superseded_records":0,"retracted_records":0,"quarantined_records":0,"hot_pages":0,"warm_pages":0,"cold_pages":0,"active_conflict_keys":0,"coarse_index_buckets":0,"last_coarse_candidates":0,"storage_mode":"embedded","resident_records":0,"resident_pages":0,"gpu_cache_records":0,"gpu_cache_tokens":0,"gpu_cache_bytes":0,"gpu_cache_hits":0,"gpu_cache_misses":0,"gpu_cache_fallbacks":0,"gpu_cache_alloc_failures":0,"gpu_cache_device":"cuda:0","gpu_cache_limit_records":256,"gpu_cache_limit_tokens":131072,"gpu_cache_reserve_mb":2048,"gpu_cache_adaptive":true,"gpu_cache_last_free_mb":null,"kv_budget":{"max_tokens":32768,"trigger_tokens":29491},"enabled":true,"router_ready":true}}},"memoryAfter":{"pages":0,"max_pages":32768,"page_capacity":32,"capacity_records":1048576,"open_pages":0,"records":0,"active_records":0,"superseded_records":0,"retracted_records":0,"quarantined_records":0,"hot_pages":0,"warm_pages":0,"cold_pages":0,"active_conflict_keys":0,"coarse_index_buckets":0,"last_coarse_candidates":0,"storage_mode":"embedded","resident_records":0,"resident_pages":0,"gpu_cache_records":0,"gpu_cache_tokens":0,"gpu_cache_bytes":0,"gpu_cache_hits":0,"gpu_cache_misses":0,"gpu_cache_fallbacks":0,"gpu_cache_alloc_failures":0,"gpu_cache_device":"cuda:0","gpu_cache_limit_records":256,"gpu_cache_limit_tokens":131072,"gpu_cache_reserve_mb":2048,"gpu_cache_adaptive":true,"gpu_cache_last_free_mb":null,"kv_budget":{"max_tokens":32768,"trigger_tokens":29491},"enabled":true,"router_ready":true}}
|
||||
{"session":"cross_session_recall:off:s1","turn":1,"userMessage":"谢谢。","assistantText":"不客气!","toolCalls":[],"toolResults":[],"wallSeconds":1.1,"diagnostics":{"turn":8,"kind":"chat","at":1789310566.0484757,"user_message":"","assistant_text":"不客气!","reasoning_chars":0,"router_decisions":[{"need_memory":false,"page_ids":[],"record_ids":[],"page_scores":[],"record_scores":[],"hop_count":0,"hop_trace":[],"confidence":0.0000035087728065263946,"stop_reason":"router_abstained","top_score":0,"score_margin":0,"evidence_score":0}],"prefix_tokens":0,"read_seconds":0.44931460000225343,"coverage":{"configured":true,"loaded":true,"bound_to_bank":true,"vocabulary_fraction":0.9,"head":{"path":"H:\\Memory\\V2_dpskw\\qwen3_5_4b_natural_memory_v2_1\\memory_attribute_head.pt","attributes":24,"vocabulary":["主管姓名","入职年份","出生城市","办公城市","午餐偏好","咖啡口味","团队名称","客户名称","宿舍楼号","工位楼层","常住城市","常用编辑器","手机尾号","档案标识","紧急联系人姓氏","设备型号","课程名称","起床时间","运动习惯","通勤方式","邮箱域名","阅读工具","项目代号","默认语言"],"coverage_stats":{"applicable":0,"bypassed":8}}},"no_evidence":false,"memory_mode":"off","memory_changed":false,"reset_triggered":false,"tool_calls":[],"overrides":{},"prompt_tokens":1863,"completion_tokens":6,"generation_seconds":1.093,"total_seconds":1.096,"snapshot":{"router_decisions":[{"need_memory":false,"page_ids":[],"record_ids":[],"page_scores":[],"record_scores":[],"hop_count":0,"hop_trace":[],"confidence":0.0000035087728065263946,"stop_reason":"router_abstained","top_score":0,"score_margin":0,"evidence_score":0}],"no_evidence":false,"prefix_tokens":0,"prefix_used":false,"read_seconds":0.44931460000225343,"auto_memory_probability":null,"auto_forget_probability":[0.011167768388986588],"context_compaction":{"compacted":false,"archived_records":0,"archived_tokens":0,"original_tokens":[1863],"retained_tokens":[1863],"hot_limit":32768,"archive_requested":true},"coverage":{"configured":true,"loaded":true,"bound_to_bank":true,"vocabulary_fraction":0.9,"head":{"path":"H:\\Memory\\V2_dpskw\\qwen3_5_4b_natural_memory_v2_1\\memory_attribute_head.pt","attributes":24,"vocabulary":["主管姓名","入职年份","出生城市","办公城市","午餐偏好","咖啡口味","团队名称","客户名称","宿舍楼号","工位楼层","常住城市","常用编辑器","手机尾号","档案标识","紧急联系人姓氏","设备型号","课程名称","起床时间","运动习惯","通勤方式","邮箱域名","阅读工具","项目代号","默认语言"],"coverage_stats":{"applicable":0,"bypassed":8}}},"memory":{"pages":0,"max_pages":32768,"page_capacity":32,"capacity_records":1048576,"open_pages":0,"records":0,"active_records":0,"superseded_records":0,"retracted_records":0,"quarantined_records":0,"hot_pages":0,"warm_pages":0,"cold_pages":0,"active_conflict_keys":0,"coarse_index_buckets":0,"last_coarse_candidates":0,"storage_mode":"embedded","resident_records":0,"resident_pages":0,"gpu_cache_records":0,"gpu_cache_tokens":0,"gpu_cache_bytes":0,"gpu_cache_hits":0,"gpu_cache_misses":0,"gpu_cache_fallbacks":0,"gpu_cache_alloc_failures":0,"gpu_cache_device":"cuda:0","gpu_cache_limit_records":256,"gpu_cache_limit_tokens":131072,"gpu_cache_reserve_mb":2048,"gpu_cache_adaptive":true,"gpu_cache_last_free_mb":null,"kv_budget":{"max_tokens":32768,"trigger_tokens":29491},"enabled":true,"router_ready":true}}},"memoryAfter":{"pages":0,"max_pages":32768,"page_capacity":32,"capacity_records":1048576,"open_pages":0,"records":0,"active_records":0,"superseded_records":0,"retracted_records":0,"quarantined_records":0,"hot_pages":0,"warm_pages":0,"cold_pages":0,"active_conflict_keys":0,"coarse_index_buckets":0,"last_coarse_candidates":0,"storage_mode":"embedded","resident_records":0,"resident_pages":0,"gpu_cache_records":0,"gpu_cache_tokens":0,"gpu_cache_bytes":0,"gpu_cache_hits":0,"gpu_cache_misses":0,"gpu_cache_fallbacks":0,"gpu_cache_alloc_failures":0,"gpu_cache_device":"cuda:0","gpu_cache_limit_records":256,"gpu_cache_limit_tokens":131072,"gpu_cache_reserve_mb":2048,"gpu_cache_adaptive":true,"gpu_cache_last_free_mb":null,"kv_budget":{"max_tokens":32768,"trigger_tokens":29491},"enabled":true,"router_ready":true}}
|
||||
{"session":"cross_session_recall:off:s2","turn":0,"userMessage":"我的项目代号是什么?只回答代号本身。","assistantText":"我的项目代号是:Project Phoenix","toolCalls":[],"toolResults":[],"wallSeconds":1.201,"diagnostics":{"turn":9,"kind":"chat","at":1789310567.2540493,"user_message":"","assistant_text":"我的项目代号是:Project Phoenix","reasoning_chars":0,"router_decisions":[{"need_memory":false,"page_ids":[],"record_ids":[],"page_scores":[],"record_scores":[],"hop_count":0,"hop_trace":[],"confidence":0.000006281216428760672,"stop_reason":"router_abstained","top_score":0,"score_margin":0,"evidence_score":0}],"prefix_tokens":0,"read_seconds":0.40200169998570345,"coverage":{"configured":true,"loaded":true,"bound_to_bank":true,"vocabulary_fraction":0.9,"head":{"path":"H:\\Memory\\V2_dpskw\\qwen3_5_4b_natural_memory_v2_1\\memory_attribute_head.pt","attributes":24,"vocabulary":["主管姓名","入职年份","出生城市","办公城市","午餐偏好","咖啡口味","团队名称","客户名称","宿舍楼号","工位楼层","常住城市","常用编辑器","手机尾号","档案标识","紧急联系人姓氏","设备型号","课程名称","起床时间","运动习惯","通勤方式","邮箱域名","阅读工具","项目代号","默认语言"],"coverage_stats":{"applicable":0,"bypassed":9}}},"no_evidence":false,"memory_mode":"off","memory_changed":false,"reset_triggered":false,"tool_calls":[],"overrides":{},"prompt_tokens":1747,"completion_tokens":10,"generation_seconds":1.194,"total_seconds":1.198,"snapshot":{"router_decisions":[{"need_memory":false,"page_ids":[],"record_ids":[],"page_scores":[],"record_scores":[],"hop_count":0,"hop_trace":[],"confidence":0.000006281216428760672,"stop_reason":"router_abstained","top_score":0,"score_margin":0,"evidence_score":0}],"no_evidence":false,"prefix_tokens":0,"prefix_used":false,"read_seconds":0.40200169998570345,"auto_memory_probability":null,"auto_forget_probability":[0.011167768388986588],"context_compaction":{"compacted":false,"archived_records":0,"archived_tokens":0,"original_tokens":[1747],"retained_tokens":[1747],"hot_limit":32768,"archive_requested":true},"coverage":{"configured":true,"loaded":true,"bound_to_bank":true,"vocabulary_fraction":0.9,"head":{"path":"H:\\Memory\\V2_dpskw\\qwen3_5_4b_natural_memory_v2_1\\memory_attribute_head.pt","attributes":24,"vocabulary":["主管姓名","入职年份","出生城市","办公城市","午餐偏好","咖啡口味","团队名称","客户名称","宿舍楼号","工位楼层","常住城市","常用编辑器","手机尾号","档案标识","紧急联系人姓氏","设备型号","课程名称","起床时间","运动习惯","通勤方式","邮箱域名","阅读工具","项目代号","默认语言"],"coverage_stats":{"applicable":0,"bypassed":9}}},"memory":{"pages":0,"max_pages":32768,"page_capacity":32,"capacity_records":1048576,"open_pages":0,"records":0,"active_records":0,"superseded_records":0,"retracted_records":0,"quarantined_records":0,"hot_pages":0,"warm_pages":0,"cold_pages":0,"active_conflict_keys":0,"coarse_index_buckets":0,"last_coarse_candidates":0,"storage_mode":"embedded","resident_records":0,"resident_pages":0,"gpu_cache_records":0,"gpu_cache_tokens":0,"gpu_cache_bytes":0,"gpu_cache_hits":0,"gpu_cache_misses":0,"gpu_cache_fallbacks":0,"gpu_cache_alloc_failures":0,"gpu_cache_device":"cuda:0","gpu_cache_limit_records":256,"gpu_cache_limit_tokens":131072,"gpu_cache_reserve_mb":2048,"gpu_cache_adaptive":true,"gpu_cache_last_free_mb":null,"kv_budget":{"max_tokens":32768,"trigger_tokens":29491},"enabled":true,"router_ready":true}}},"memoryAfter":{"pages":0,"max_pages":32768,"page_capacity":32,"capacity_records":1048576,"open_pages":0,"records":0,"active_records":0,"superseded_records":0,"retracted_records":0,"quarantined_records":0,"hot_pages":0,"warm_pages":0,"cold_pages":0,"active_conflict_keys":0,"coarse_index_buckets":0,"last_coarse_candidates":0,"storage_mode":"embedded","resident_records":0,"resident_pages":0,"gpu_cache_records":0,"gpu_cache_tokens":0,"gpu_cache_bytes":0,"gpu_cache_hits":0,"gpu_cache_misses":0,"gpu_cache_fallbacks":0,"gpu_cache_alloc_failures":0,"gpu_cache_device":"cuda:0","gpu_cache_limit_records":256,"gpu_cache_limit_tokens":131072,"gpu_cache_reserve_mb":2048,"gpu_cache_adaptive":true,"gpu_cache_last_free_mb":null,"kv_budget":{"max_tokens":32768,"trigger_tokens":29491},"enabled":true,"router_ready":true}}
|
||||
{"session":"cross_session_recall:off:s3","turn":0,"userMessage":"部署区域呢?只回答区域本身。","assistantText":"未找到名为 \"deployment\" 的目录。","toolCalls":[{"name":"run_shell","args":{"command":"find / -type d -name \"deployment\" 2>/dev/null | head -5","timeout_ms":5000}}],"toolResults":[{"name":"run_shell","result":"{\"content\":[{\"type\":\"text\",\"text\":\"command failed: Command failed: pwsh -NoProfile -Command find / -type d -name \\\"deployment\\\" 2>/dev/null | head -5\\nOut-File: Could not find a part of the path 'H:\\\\dev\\\\null'.\\r\\n\"}],\"details\":{}}"}],"wallSeconds":7.194,"diagnostics":{"turn":11,"kind":"chat","at":1789310574.4523666,"user_message":"","assistant_text":"未找到名为 \"deployment\" 的目录。","reasoning_chars":0,"router_decisions":[{"need_memory":false,"page_ids":[],"record_ids":[],"page_scores":[],"record_scores":[],"hop_count":0,"hop_trace":[],"confidence":0.0000016707654140191153,"stop_reason":"router_abstained","top_score":0,"score_margin":0,"evidence_score":0}],"prefix_tokens":0,"read_seconds":0.5463175000040792,"coverage":{"configured":true,"loaded":true,"bound_to_bank":true,"vocabulary_fraction":0.9,"head":{"path":"H:\\Memory\\V2_dpskw\\qwen3_5_4b_natural_memory_v2_1\\memory_attribute_head.pt","attributes":24,"vocabulary":["主管姓名","入职年份","出生城市","办公城市","午餐偏好","咖啡口味","团队名称","客户名称","宿舍楼号","工位楼层","常住城市","常用编辑器","手机尾号","档案标识","紧急联系人姓氏","设备型号","课程名称","起床时间","运动习惯","通勤方式","邮箱域名","阅读工具","项目代号","默认语言"],"coverage_stats":{"applicable":0,"bypassed":11}}},"no_evidence":false,"memory_mode":"off","memory_changed":false,"reset_triggered":false,"tool_calls":[],"overrides":{},"prompt_tokens":1901,"completion_tokens":13,"generation_seconds":1.5,"total_seconds":1.504,"snapshot":{"router_decisions":[{"need_memory":false,"page_ids":[],"record_ids":[],"page_scores":[],"record_scores":[],"hop_count":0,"hop_trace":[],"confidence":0.0000016707654140191153,"stop_reason":"router_abstained","top_score":0,"score_margin":0,"evidence_score":0}],"no_evidence":false,"prefix_tokens":0,"prefix_used":false,"read_seconds":0.5463175000040792,"auto_memory_probability":null,"auto_forget_probability":[0.011167768388986588],"context_compaction":{"compacted":false,"archived_records":0,"archived_tokens":0,"original_tokens":[1901],"retained_tokens":[1901],"hot_limit":32768,"archive_requested":true},"coverage":{"configured":true,"loaded":true,"bound_to_bank":true,"vocabulary_fraction":0.9,"head":{"path":"H:\\Memory\\V2_dpskw\\qwen3_5_4b_natural_memory_v2_1\\memory_attribute_head.pt","attributes":24,"vocabulary":["主管姓名","入职年份","出生城市","办公城市","午餐偏好","咖啡口味","团队名称","客户名称","宿舍楼号","工位楼层","常住城市","常用编辑器","手机尾号","档案标识","紧急联系人姓氏","设备型号","课程名称","起床时间","运动习惯","通勤方式","邮箱域名","阅读工具","项目代号","默认语言"],"coverage_stats":{"applicable":0,"bypassed":11}}},"memory":{"pages":0,"max_pages":32768,"page_capacity":32,"capacity_records":1048576,"open_pages":0,"records":0,"active_records":0,"superseded_records":0,"retracted_records":0,"quarantined_records":0,"hot_pages":0,"warm_pages":0,"cold_pages":0,"active_conflict_keys":0,"coarse_index_buckets":0,"last_coarse_candidates":0,"storage_mode":"embedded","resident_records":0,"resident_pages":0,"gpu_cache_records":0,"gpu_cache_tokens":0,"gpu_cache_bytes":0,"gpu_cache_hits":0,"gpu_cache_misses":0,"gpu_cache_fallbacks":0,"gpu_cache_alloc_failures":0,"gpu_cache_device":"cuda:0","gpu_cache_limit_records":256,"gpu_cache_limit_tokens":131072,"gpu_cache_reserve_mb":2048,"gpu_cache_adaptive":true,"gpu_cache_last_free_mb":null,"kv_budget":{"max_tokens":32768,"trigger_tokens":29491},"enabled":true,"router_ready":true}}},"memoryAfter":{"pages":0,"max_pages":32768,"page_capacity":32,"capacity_records":1048576,"open_pages":0,"records":0,"active_records":0,"superseded_records":0,"retracted_records":0,"quarantined_records":0,"hot_pages":0,"warm_pages":0,"cold_pages":0,"active_conflict_keys":0,"coarse_index_buckets":0,"last_coarse_candidates":0,"storage_mode":"embedded","resident_records":0,"resident_pages":0,"gpu_cache_records":0,"gpu_cache_tokens":0,"gpu_cache_bytes":0,"gpu_cache_hits":0,"gpu_cache_misses":0,"gpu_cache_fallbacks":0,"gpu_cache_alloc_failures":0,"gpu_cache_device":"cuda:0","gpu_cache_limit_records":256,"gpu_cache_limit_tokens":131072,"gpu_cache_reserve_mb":2048,"gpu_cache_adaptive":true,"gpu_cache_last_free_mb":null,"kv_budget":{"max_tokens":32768,"trigger_tokens":29491},"enabled":true,"router_ready":true}}
|
||||
File diff suppressed because one or more lines are too long
@@ -0,0 +1 @@
|
||||
项目代号:蓝鲸-47,部署区域:ap-southeast-3
|
||||
@@ -0,0 +1,39 @@
|
||||
{
|
||||
"baseUrl": "http://127.0.0.1:8766/v1",
|
||||
"outDir": "H:\\Memory\\agent_lab\\agent\\runs\\v2",
|
||||
"arms": [
|
||||
"on",
|
||||
"off"
|
||||
],
|
||||
"scenarios": [
|
||||
{
|
||||
"id": "cross_session_recall",
|
||||
"description": "写入事实后,在全新会话(空上下文、无历史文本)中询问该事实。这正是 in-model memory 存在的理由。",
|
||||
"arms": {
|
||||
"on": {
|
||||
"activeRecords": 1,
|
||||
"assertionsPassed": 1,
|
||||
"assertionsTotal": 3,
|
||||
"passRate": 0.3333333333333333,
|
||||
"failures": [
|
||||
"answer_contains: missing one of [\"蓝鲸-47\",\"蓝鲸 47\",\"蓝鲸47\"]",
|
||||
"answer_contains: missing one of [\"ap-southeast-3\",\"apse-3\"]"
|
||||
],
|
||||
"meanSecondsPerTurn": 3.77
|
||||
},
|
||||
"off": {
|
||||
"activeRecords": 0,
|
||||
"assertionsPassed": 0,
|
||||
"assertionsTotal": 3,
|
||||
"passRate": 0,
|
||||
"failures": [
|
||||
"answer_contains: missing one of [\"蓝鲸-47\",\"蓝鲸 47\",\"蓝鲸47\"]",
|
||||
"answer_contains: missing one of [\"ap-southeast-3\",\"apse-3\"]",
|
||||
"active_record_contains: absent from active memory"
|
||||
],
|
||||
"meanSecondsPerTurn": 3.65
|
||||
}
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
Reference in New Issue
Block a user