Commit Graph

149 Commits

Author SHA1 Message Date
jerryjzhang
c99d240b65 [improvement][headless]Restructure LLMReq and LLMSchema. 2024-09-12 20:10:17 +08:00
jerryjzhang
693356e46a [improvement][headless]Add metricFormat field to metric metadata of the Text2SQL prompt. #1621 2024-09-12 18:13:44 +08:00
jerryjzhang
2fa3bfe019 (improvement)(headless)Add timeFormat to dimension metadata of the Text2SQL prompt.
(improvement)(headless)Add `timeFormat` to dimension metadata of the Text2SQL prompt. #1621.
2024-09-12 10:03:24 +08:00
jerryjzhang
82b5fa966a [improvement][headless]Add partitionTime and primaryKey field into the Schema part of the Text2SQL prompt. #1621 2024-09-11 21:29:37 +08:00
lexluo09
fb96ce5843 [improvement](chat) Use the word-based approach for comparing Schema Mapping rules and fix the issue where the results were not being modified. (#1650) 2024-09-10 23:21:12 +08:00
lexluo09
879696e493 [improvement](chat) Optimize the Schema Mapping rules (#1649) 2024-09-10 22:58:34 +08:00
lexluo09
183caf7931 [improvement](chat) Refactor the code for the Map phase (#1646) 2024-09-10 17:19:05 +08:00
lexluo09
5f59e89eea (improvement)(build) Add spotless during the build process. (#1639) 2024-09-07 00:36:17 +08:00
jerryjzhang
ee15a88b06 (improvement)(headless)Refactor the prompts for generating semantic aliases. 2024-09-06 17:55:33 +08:00
lexluo09
1b6277e53f (improvement)(chat) Fixed the NullPointerException issue in DimensionRecommendProcessor (#1629) 2024-09-04 22:46:38 +08:00
yudong
14177e265f (fix)(headless) 修复LLM问题变成术语的问题 (#1617)
Co-authored-by: wangyudong
2024-09-02 21:41:19 +08:00
lexluo09
26863769b5 (improvement)(chat) The format of a regular date is provided to the large model (#1619) 2024-09-01 17:30:40 +08:00
lexluo09
c4284f9784 (improvement)(chat) Remove the platform from the Docker Compose file and ensure the existence of a field with the same name before deleting a specific data date. (#1609) 2024-08-28 15:22:54 +08:00
lexluo09
b9ae0a4c92 (improvement)(chat) Forcefully delete the data date generated by the large model. (#1607) 2024-08-27 12:36:39 +08:00
jerryjzhang
5229fdc8b5 (improvement)(headless)Remove tag-related logic when determining query type. 2024-08-24 09:47:42 +08:00
lexluo09
bef652892b (improvement)(chat) Do not pass default date configuration to the large model uniformly. (#1601) 2024-08-24 08:08:39 +08:00
lexluo09
d2306464a6 (improvement)(chat) Unified display of time filtering conditions to the frontend. (#1600) 2024-08-23 19:01:37 +08:00
lexluo09
371e2f1e05 (improvement)(chat) The rules will uniformly use 'data date' and exclude data partition dimensions for dimension pushdown. (#1598) 2024-08-23 17:38:04 +08:00
lexluo09
59c50176c5 (improvement)(chat) Fix the issue of duplicate addition of the 'data date' in the rule pattern. (#1597) 2024-08-23 16:33:23 +08:00
lexluo09
be9a8bbc27 (improvement)(chat) Remove the special field 'data date' to avoid generating non-existent time fields. (#1596) 2024-08-23 13:37:14 +08:00
jerryjzhang
7e013ca36a (fix)(headless)Avoid IndexOutOfBound exception in some cases. 2024-08-20 15:48:24 +08:00
lexluo09
ca098b576c (fix)(chat) Fix the null pointer issue in fieldNameToDateFormat. (#1587) 2024-08-19 19:31:27 +08:00
lexluo09
ba55ecb31e (improvement)(chat) Make corrections and pass the data date format to the large model. (#1583) 2024-08-19 01:21:10 +08:00
lexluo09
2801b27ade (improvement)(chat) Fix the issue where the query gets stuck when two terms appear simultaneously. (#1581) 2024-08-18 00:16:55 +08:00
lexluo09
07e0ba24bc (improvement)(chat) After fixing the dictionary execution task, it is necessary to reload the metadata. (#1579) 2024-08-17 23:49:56 +08:00
lexluo09
898c7100ba (improvement)(chat) Modify the configuration of the zhipu vector database to disable the default persistence of the vector database configuration. (#1577) 2024-08-16 23:14:23 +08:00
lexluo09
7150f19def (improvement)(chat) Vector retrieval supports filtering by modelId collection during query. (#1576) 2024-08-16 21:31:07 +08:00
yudong
6aff51d394 [fix][headless]metadata向量检索时对modelid的过滤增加下划线 (#1570) 2024-08-14 13:37:09 +08:00
lexluo09
c82c2d0a95 (improvement)(chat) Remove irrelevant topN field information during the parsing of large models. (#1558) 2024-08-12 17:31:42 +08:00
jerryjzhang
b13b38c645 (fix)(headless)Fix demo conversations with DETAIL query mode.
(fix)(headless)Fix demo conversations with DETAIL query mode.
2024-08-10 23:19:46 +08:00
jerryjzhang
ba9e6afa51 (improvement)(headless)Introduce DetailDimensionQuery as a type of rule-based parsing query. 2024-08-10 18:27:50 +08:00
jerryjzhang
ecc651e12d (improvement)(headless)Optimize Text2SQL prompt, explicitly ask LLM not hallucinate columns. 2024-08-09 19:27:38 +08:00
lexluo09
9dbc8657e2 (improvement)(chat) When making corrections, the 'group by' field must not be included in the function. (#1532) 2024-08-07 23:08:43 +08:00
lexluo09
208686de46 (improvement)(chat) Rules, large models, and query dimension values support timelessness. (#1522) 2024-08-07 13:29:07 +08:00
yudong
89b028b594 [fix][headless]指定不限制数据集查询范围时时Detail类型也不再额外增加限制 (#1514) 2024-08-06 14:06:40 +08:00
lexluo09
0a4272c25e (improvement)(chat) Enable partition dates in the demo and support partition dates at the dataset granularity. (#1513) 2024-08-04 23:01:54 +08:00
lexluo09
e2e45a40ab (improvement)(chat) Decide whether to add or remove dates based on whether the dataset has partition dates. (#1512) 2024-08-04 17:39:23 +08:00
lexluo09
a9232fa1c7 (improvement)(chat) Optimize and fix the NatureHelper code. (#1510)
Co-authored-by: lexluo <lexluo@tencent.com>
2024-08-03 23:52:52 +08:00
yudong
53a9f7c451 [improvement][headless&chat]当LLM生成SQL包含日期类型字段时,Correcter不再额外增加日期 (#1473) 2024-08-02 14:14:29 +08:00
ylhh353
5a18ad5229 [improvement][headless-chat] improvement sql generation ``` (#1498) (#1499) 2024-08-01 20:04:11 +08:00
jerryjzhang
23af977972 (fix)(headless)Fix default date calculation. #1471 2024-07-30 09:54:58 +08:00
LXW
9a14728152 (fix)(headless) Fix queryText replaced when semantic retrieval to term descriptions (#1469)
Co-authored-by: lxwcodemonkey
2024-07-29 10:18:58 +08:00
LXW
26f682cc45 (improvement)(headless) Add semantic retrieval to term descriptions and extract relevant semantic information (#1468)
Co-authored-by: lxwcodemonkey
2024-07-29 09:15:18 +08:00
jerryjzhang
ccd79e4830 [improvement][headless]Optimize code structure and code style. 2024-07-28 10:06:16 +08:00
jerryjzhang
e5504473a4 [improvement][headless&chat]Move EntityInfoProcessor from chat to headless module and optimize code.
[improvement][headless&chat]Move `EntityInfoProcessor` from `chat` to `headless` module and optimize code.
2024-07-27 18:10:06 +08:00
JackLiang
8f620480c6 fix:加载模型到内存时内存泄露bug (#1465)
* fix existSegment bug which embedding search not has min-score,so that it will return true.
2024-07-26 18:07:49 +08:00
jerryjzhang
f38a84bc8c (improvement)(headless)Reformat data semantics in the prompt. 2024-07-26 14:08:20 +08:00
jerryjzhang
cd8f38c334 (improvement)(headless)Move Values from sideInfo to schema part of the prompt and optimize prompt text. 2024-07-26 10:54:30 +08:00
lexluo09
ae34c15c95 (improvement)(chat) Improve vector recall performance. (#1458) 2024-07-25 22:19:35 +08:00
jerryjzhang
c39460ee02 (improvement)(headless)Add alias of metrics & dimensions to the schema part of the prompts. 2024-07-24 20:20:11 +08:00