mirror of
https://github.com/svcvit/Awesome-Dify-Workflow.git
synced 2026-06-12 18:11:31 +08:00
update
This commit is contained in:
svcvit
-615
@@ -1,58 +1,3 @@
|
||||
### Ron - 2025-05-08 10:49:22
|
||||
Hi
|
||||
|
||||
---
|
||||
|
||||
### Ron - 2025-05-08 11:14:27
|
||||
欢迎，第一位付费用户🎉('ω')🎉
|
||||
|
||||
---
|
||||
|
||||
### hjlarry - 2025-05-08 11:14:47
|
||||
欢迎
|
||||
|
||||
---
|
||||
|
||||
### wxid_7da5db36yiad22 - 2025-05-08 11:14:53
|
||||
哈哈哈
|
||||
|
||||
---
|
||||
|
||||
### wxid_7da5db36yiad22 - 2025-05-08 11:15:23
|
||||
这两位是你们搭档吗
|
||||
|
||||
---
|
||||
|
||||
### Ron - 2025-05-08 11:15:45
|
||||
多群的群友～
|
||||
|
||||
---
|
||||
|
||||
### wxid_7da5db36yiad22 - 2025-05-08 11:16:12
|
||||
[坏笑]一起搞
|
||||
|
||||
---
|
||||
|
||||
### wxid_2zyfe21ojzzw22 - 2025-05-08 11:20:38
|
||||
[坏笑]以后麻烦大家了，多交流
|
||||
|
||||
---
|
||||
|
||||
### wxid_7da5db36yiad22 - 2025-05-08 11:20:42
|
||||
哈哈 002来了
|
||||
|
||||
---
|
||||
|
||||
### wxid_2zyfe21ojzzw22 - 2025-05-08 11:20:53
|
||||
[破涕为笑]我居然不上001
|
||||
|
||||
---
|
||||
|
||||
### hjlarry - 2025-05-08 11:20:55
|
||||
欢迎欢迎
|
||||
|
||||
---
|
||||
|
||||
### wxid_2zyfe21ojzzw22 - 2025-05-08 11:22:08
|
||||
我想请问一下dify在做知识库查询的时候，预处理文档有没有比较好的一个方法奥
|
||||
|
||||
@@ -740,566 +685,6 @@ http插件限制了大小吗？
|
||||
### hjlarry - 2025-05-10 11:44:57
|
||||
对的，环境变量里好像叫 http max body 吧
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:24:51
|
||||
老师们好，我有一个场景:低代码代码生成，主要是rag相关的dify流程设计还有知识库清洗做不好
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:24:52
|
||||
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:24:56
|
||||
大致的dify流程是这样的
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:25:57
|
||||
我的低代码用的是百度开源的amis那套
|
||||
|
||||
---
|
||||
|
||||
### Ron - 2025-05-10 17:26:05
|
||||
你知识库的原始材料，发个看看？去掉一些敏感信息
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:26:13
|
||||
本身其实是已经ai有训练过的
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:26:14
|
||||
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:26:23
|
||||
知识库是一个个案例
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:26:47
|
||||
我的思路是用使用者已经上线的那些页面的json
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:26:51
|
||||
解析出字段
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:26:58
|
||||
作为案例库来源
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:27:12
|
||||
最终的强化版prompt我发一下示例
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:28:21
|
||||
<fields>
|
||||
[{"name": "cityId","description":"城市id","type":"string"}]
|
||||
|
||||
# 示例
|
||||
| 字段名 | 字段类型 | 是否必填 | 字段描述 | 标准答案 |
|
||||
| cityGuid | string | 否 | 所属城市 | {"id":"u:76f8a6fbd645","name":"cityGuid","type":"hb-city-select","label":"城市","cityCode":false,"showWholeCountry":false} |
|
||||
# 角色
|
||||
你是一个百度低代码amis的局部代码生成器。你的职责是根据提供的fileds字段，优先参照示例中的标准答案来推导，为每一个字段生成amis json。
|
||||
|
||||
## 约束
|
||||
- 优先参照示例中的返回，有时候会出现业务组件（非amis内置组件），需要从示例中推断用法
|
||||
- 只输出生成的 JSON 数据，不需要输出解析过程，也不需要说明解释。
|
||||
- 输出的数据需符合以下条件：
|
||||
1. 数据必须是一个数组。
|
||||
2. 根据字段中是否有枚举来调整生成内容。
|
||||
3. 输出的标题简洁、语义化，删除多余的描述。
|
||||
4. 如果值是有含义的英文字符串，则做合理的翻译。
|
||||
- 补充一些常用业务的约束。
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:28:46
|
||||
这部分是接口文档的字段定义
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:28:46
|
||||
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:29:06
|
||||
这部分是从知识库里面获取到的案例
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:29:07
|
||||
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:29:18
|
||||
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:29:27
|
||||
结果也确实输出了我预期的内容
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:29:48
|
||||
我现在的问题是我线上解析出来的字段很多很多
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:30:01
|
||||
有的会内容超出知识块长度限制
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:30:34
|
||||
这是拿到的不是完整的示例，我给一个失败的案例最终prompt给你们看下
|
||||
|
||||
---
|
||||
|
||||
### Ron - 2025-05-10 17:33:24
|
||||
我感觉，这个知识库的组织不太对，因为相似的都太高了，这种召回，感觉召回的内容都没有太大意义
|
||||
|
||||
---
|
||||
|
||||
### Ron - 2025-05-10 17:34:49
|
||||
你这个流程，期待的输入，输出是什么？
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:35:11
|
||||
期待的输入是字段的声明，例如
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:35:12
|
||||
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:35:20
|
||||
这个字段声明来自于swagger或者yapi
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:35:29
|
||||
这类接口定义站点里
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:36:39
|
||||
输出是低代码的json，例如
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:36:54
|
||||
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:37:04
|
||||
fields里面就是低代码的schema
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:37:43
|
||||
其实amis本身是训练过的，所以就算不做知识库补充，也能准确度较高地生成
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:37:49
|
||||
就是不能生成本地扩充的业务组件
|
||||
|
||||
---
|
||||
|
||||
### wxid_pgye6v6ofbws22 - 2025-05-10 17:38:17
|
||||
给个全称我检索一下
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:39:36
|
||||
还有就是不能做类似业务习惯的配置，例如amis配置日期输入默认就是以秒为单位的值
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:39:36
|
||||
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:39:57
|
||||
咋们习惯用毫秒的
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:40:10
|
||||
https://aisuda.bce.baidu.com/amis/zh-CN/components/form/input-date
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:40:30
|
||||
所以配置案例实际上是希望影响生成出来的效果
|
||||
|
||||
---
|
||||
|
||||
### hjlarry - 2025-05-10 17:44:19
|
||||
HTTP_REQUEST_NODE_MAX_TEXT_SIZE=1048576
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:45:28
|
||||
例如ai默认生成的是{
|
||||
"type": "input-date",
|
||||
"name": "date",
|
||||
"label": "日期"
|
||||
}
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:45:39
|
||||
我们希望是{
|
||||
"type": "input-date",
|
||||
"name": "date",
|
||||
"label": "日期",
|
||||
"format": "x"
|
||||
}
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:46:45
|
||||
我觉得要初始化很多示例会很麻烦，所以才想说把团队已经上线的那个页面schema作为素材来用
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:47:19
|
||||
因为我们写了100来张页面，所以初始化理论上有1000+的字段案例
|
||||
|
||||
---
|
||||
|
||||
### Ron - 2025-05-10 17:50:10
|
||||
我大概理解你的意思和目的了
|
||||
|
||||
---
|
||||
|
||||
### Ron - 2025-05-10 17:50:37
|
||||
你这 1000 多个字段，应该有很多类似，或者重复的吧
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:50:46
|
||||
对
|
||||
|
||||
---
|
||||
|
||||
### Ron - 2025-05-10 17:51:23
|
||||
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:51:54
|
||||
且有的是意义不明，因为比如存在出现其他字段的显影控制，"visibleOn": "${items.length > 1}",
|
||||
|
||||
---
|
||||
|
||||
### Ron - 2025-05-10 17:54:18
|
||||
你这里，这种召回，这肯定都很难召对，我觉得你们可以整理一下，你们自己调整过的组建，形成一个文档，假设 30 个？不要重复。然后作为提示词的上下文，ai 输出，可能比用知识库要好。
|
||||
|
||||
---
|
||||
|
||||
### Ron - 2025-05-10 17:54:52
|
||||
你看上面 3 个，基本都是重复的
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:55:05
|
||||
有不同用法
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:55:10
|
||||
同一个组件不同用法
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:57:13
|
||||
有些是可以看出来的
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:57:44
|
||||
这个其实引申出来也可以做用户描述转低代码的
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:58:09
|
||||
描述的越清楚，理论上就可以从知识库里面找到最合适的案例，所以知识库得维护好
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 17:58:13
|
||||
且多
|
||||
|
||||
---
|
||||
|
||||
### hjlarry - 2025-05-10 18:00:14
|
||||
你的知识库存的本质上是一个个example？ 那你是觉得现在没有召回到合适的example还是召回了模型没按示例生成？
|
||||
|
||||
---
|
||||
|
||||
### Ron - 2025-05-10 18:00:47
|
||||
如果这样，你的知识库更合适是一个描述案例，对应一个最佳输出的 json
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:01:45
|
||||
合适的example能带来合适的输出
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:02:06
|
||||
一个是线上的字段案例有时候会超出dify的知识片段长度限制
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:02:29
|
||||
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:02:31
|
||||
比如说这个
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:02:40
|
||||
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:02:43
|
||||
段落是不完整的
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:03:18
|
||||
一个是线上的案例字段单独抽出来会有迷惑的地方
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:04:01
|
||||
比如说visibleOn: "${其他元素定义的字段}"
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:04:25
|
||||
需要加上其他元素才能体现出visibleOn的用法
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:04:35
|
||||
但是我这边逻辑是一个字段一个案例
|
||||
|
||||
---
|
||||
|
||||
### Ron - 2025-05-10 18:05:47
|
||||
哈哈哈，我知道了，你现在适合用一个高级模型，长上下文，把你现在 1000 个组建，丢进去，然后写一两个示例，让 ai 整理，输出对应内容，例如 100 个
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:08:15
|
||||
用ai来整理1000个案例，输出有效的配置案例？
|
||||
|
||||
---
|
||||
|
||||
### Ron - 2025-05-10 18:08:56
|
||||
我感觉先看看整理效果
|
||||
|
||||
---
|
||||
|
||||
### Ron - 2025-05-10 18:09:04
|
||||
开始不用这么多
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:10:45
|
||||
我字段案例其实已经导出为markdown文件了，直接放进去就可以了
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:11:38
|
||||
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:11:38
|
||||
类似这样的
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:13:03
|
||||
我现在思考的是在对combo、input-table这类有子元素的字段做解析，把里面的字段抽出来作为新的案例
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:13:34
|
||||
来减少单行字段的内容长度，减少知识库切割为片段不完整的概率
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:15:13
|
||||
还有就是对描述进行逆推，因为很多接口字段说明其实不能体现出细节功能点
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:15:55
|
||||
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:16:01
|
||||
用ai来逆推这个
|
||||
|
||||
---
|
||||
|
||||
### wxid_dpsw5gcj3vg521 - 2025-05-10 18:17:02
|
||||
先去吃饭啦，辛苦各位老师抽空帮我看看
|
||||
|
||||
---
|
||||
|
||||
### Ron - 2025-05-10 18:23:17
|
||||
okok，你回头发我个 100 行，给我两三个输入和输出的例子，输入是自然语言加上你现在输入的组合，输出是你的组件 json
|
||||
|
||||
---
|
||||
|
||||
### wxid_pgye6v6ofbws22 - 2025-05-11 09:13:28
|
||||
flowgram和dify比起来有优势吗？
|
||||
|
||||
---
|
||||
|
||||
### wwzhouhui - 2025-05-11 09:15:15
|
||||
flowgram是字节刚推出的开源项目，就是 目前用的coze 提供的工作流，目前还没深入研究
|
||||
|
||||
---
|
||||
|
||||
### wwzhouhui - 2025-05-11 09:27:25
|
||||
|
||||
|
||||
---
|
||||
|
||||
### wxid_twh09qctjv7822 - 2025-05-11 09:29:09
|
||||
flowgram 是不是只是一个前端画布啊，能直接用吗？
|
||||
|
||||
---
|
||||
|
||||
### wwzhouhui - 2025-05-11 09:30:38
|
||||
它开源到了 可以在自己电脑上部署，我一会研究一下，有消息发给大家
|
||||
|
||||
---
|
||||
|
||||
### wxid_pgye6v6ofbws22 - 2025-05-11 09:30:42
|
||||
部署一个玩玩，感觉比dify抖音插件多
|
||||
|
||||
---
|
||||
|
||||
### wwzhouhui - 2025-05-11 09:31:31
|
||||
但是有一点 你要知道 他的东西开源 很多模型后端是要字节自己东西，可能会有点贵
|
||||
|
||||
---
|
||||
|
||||
### wwzhouhui - 2025-05-11 09:32:01
|
||||
已就是说 它是开源了，但是 模型 TTS 还有插件是 它自己指定的模型和相关工具
|
||||
|
||||
---
|
||||
|
||||
### wwzhouhui - 2025-05-11 09:32:33
|
||||
我之前弄过他 火山引擎上的几个开源项目，效果是挺不错，但是需要他们家的模型。
|
||||
|
||||
---
|
||||
|
||||
### wwzhouhui - 2025-05-11 10:29:44
|
||||
|
||||
|
||||
---
|
||||
|
||||
### wwzhouhui - 2025-05-11 10:29:44
|
||||
跑出来了，目前功能还比较弱
|
||||
|
||||
---
|
||||
|
||||
### wwzhouhui - 2025-05-11 10:39:32
|
||||
感兴趣的可以看一下
|
||||
|
||||
---
|
||||
|
||||
### wxid_pgye6v6ofbws22 - 2025-05-11 10:46:39
|
||||
啥也没有
|
||||
|
||||
---
|
||||
|
||||
### wxid_pgye6v6ofbws22 - 2025-05-11 10:47:00
|
||||
|
||||
|
||||
---
|
||||
|
||||
### wxid_pgye6v6ofbws22 - 2025-05-11 10:47:20
|
||||
就分支 llm 和loop没了
|
||||
|
||||
---
|
||||
|
||||
### hjlarry - 2025-05-11 10:47:45
|
||||
那倒也不全是 reactflow，多了两节点[旺柴]
|
||||
|
||||
---
|
||||
|
||||
### wxid_pgye6v6ofbws22 - 2025-05-11 10:51:00
|
||||
暂时想不出来这玩意能干嘛[捂脸]
|
||||
|
||||
---
|
||||
|
||||
### wxid_pgye6v6ofbws22 - 2025-05-11 10:51:14
|
||||
不知道后面会不会做成dify那样
|
||||
|
||||
---
|
||||
|
||||
### wxid_pgye6v6ofbws22 - 2025-05-11 10:51:29
|
||||
不过这项目现在后端都没
|
||||
|
||||
---
|
||||
|
||||
### hjlarry - 2025-05-11 10:53:31
|
||||
这项目就是把coze的前端画布部分开源出来了吧 方便其他想做画布类应用的开发者
|
||||
|
||||
---
|
||||
|
||||
### wxid_pgye6v6ofbws22 - 2025-05-11 10:56:37
|
||||
感觉是这样
|
||||
|
||||
---
|
||||
|
||||
### godspeed6633 - 2025-05-11 11:44:06
|
||||
|
||||
|
||||
---
|
||||
|
||||
### godspeed6633 - 2025-05-11 11:44:15
|
||||
|
||||
|
||||
---
|
||||
|
||||
### godspeed6633 - 2025-05-11 11:44:41
|
||||
|
||||
|
||||
---
|
||||
|
||||
|
||||
Reference in New Issue
Block a user