Merge branch 'main' of https://github.com/maxkb-dev/maxkb

apps/application/chat_pipeline/I_base_chat_pipeline.py

@@ -19,7 +19,7 @@ class ParagraphPipelineModel:
 
     def __init__(self, _id: str, document_id: str, dataset_id: str, content: str, title: str, status: str,
                  is_active: bool, comprehensive_score: float, similarity: float, dataset_name: str, document_name: str,
-                 hit_handling_method: str, directly_return_similarity: float):
+                 hit_handling_method: str, directly_return_similarity: float, meta: dict = None):
         self.id = _id
         self.document_id = document_id
         self.dataset_id = dataset_id
@@ -33,6 +33,7 @@ class ParagraphPipelineModel:
         self.document_name = document_name
         self.hit_handling_method = hit_handling_method
         self.directly_return_similarity = directly_return_similarity
+        self.meta = meta
 
     def to_dict(self):
         return {
@@ -46,7 +47,8 @@ class ParagraphPipelineModel:
             'comprehensive_score': self.comprehensive_score,
             'similarity': self.similarity,
             'dataset_name': self.dataset_name,
-            'document_name': self.document_name
+            'document_name': self.document_name,
+            'meta': self.meta,
         }
 
     class builder:
@@ -58,6 +60,7 @@ class ParagraphPipelineModel:
             self.dataset_name = None
             self.hit_handling_method = None
             self.directly_return_similarity = 0.9
+            self.meta = {}
 
         def add_paragraph(self, paragraph):
             if isinstance(paragraph, Paragraph):
@@ -97,6 +100,10 @@ class ParagraphPipelineModel:
             self.similarity = similarity
             return self
 
+        def add_meta(self, meta: dict):
+            self.meta = meta
+            return self
+
         def build(self):
             return ParagraphPipelineModel(str(self.paragraph.get('id')), str(self.paragraph.get('document_id')),
                                           str(self.paragraph.get('dataset_id')),
@@ -104,7 +111,8 @@ class ParagraphPipelineModel:
                                           self.paragraph.get('status'),
                                           self.paragraph.get('is_active'),
                                           self.comprehensive_score, self.similarity, self.dataset_name,
-                                          self.document_name, self.hit_handling_method, self.directly_return_similarity)
+                                          self.document_name, self.hit_handling_method, self.directly_return_similarity,
+                                          self.meta)
 
 
 class IBaseChatPipelineStep:

apps/application/chat_pipeline/step/search_dataset_step/impl/base_search_dataset_step.py

@@ -79,6 +79,7 @@ class BaseSearchDatasetStep(ISearchDatasetStep):
                     .add_document_name(paragraph.get('document_name'))
                     .add_hit_handling_method(paragraph.get('hit_handling_method'))
                     .add_directly_return_similarity(paragraph.get('directly_return_similarity'))
+                    .add_meta(paragraph.get('meta'))
                     .build())
 
     @staticmethod

apps/application/flow/step_node/document_extract_node/impl/base_document_extract_node.py

@@ -40,7 +40,7 @@ class BaseDocumentExtractNode(IDocumentExtractNode):
             "index": index,
             'run_time': self.context.get('run_time'),
             'type': self.node.type,
-            'content': self.context.get('content'),
+            # 'content': self.context.get('content'), # 不保存content内容，因为content内容可能会很大
             'status': self.status,
             'err_message': self.err_message,
             'document_list': self.context.get('document_list')

apps/application/flow/step_node/image_understand_step_node/impl/base_image_understand_node.py

@@ -67,11 +67,13 @@ class BaseImageUnderstandNode(IImageUnderstandNode):
                 image,
                 **kwargs) -> NodeResult:
         image_model = get_model_instance_by_model_user_id(model_id, self.flow_params_serializer.data.get('user_id'))
-        history_message = self.get_history_message(history_chat_record, dialogue_number)
+        # 执行详情中的历史消息不需要图片内容
+        history_message =self.get_history_message_for_details(history_chat_record, dialogue_number)
         self.context['history_message'] = history_message
         question = self.generate_prompt_question(prompt)
         self.context['question'] = question.content
-        message_list = self.generate_message_list(image_model, system, prompt, history_message, image)
+        # 生成消息列表, 真实的history_message
+        message_list = self.generate_message_list(image_model, system, prompt, self.get_history_message(history_chat_record, dialogue_number), image)
         self.context['message_list'] = message_list
         self.context['image_list'] = image
         self.context['dialogue_type'] = dialogue_type
@@ -86,6 +88,15 @@ class BaseImageUnderstandNode(IImageUnderstandNode):
                                'history_message': history_message, 'question': question.content}, {},
                               _write_context=write_context)
 
+    @staticmethod
+    def get_history_message_for_details(history_chat_record, dialogue_number):
+        start_index = len(history_chat_record) - dialogue_number
+        history_message = reduce(lambda x, y: [*x, *y], [
+            [history_chat_record[index].get_human_message(), history_chat_record[index].get_ai_message()]
+            for index in
+            range(start_index if start_index > 0 else 0, len(history_chat_record))], [])
+        return history_message
+
     def get_history_message(self, history_chat_record, dialogue_number):
         start_index = len(history_chat_record) - dialogue_number
         history_message = reduce(lambda x, y: [*x, *y], [

apps/application/sql/list_dataset_paragraph_by_paragraph_id.sql

@@ -2,6 +2,7 @@ SELECT
 	paragraph.*,
 	dataset."name" AS "dataset_name",
 	"document"."name" AS "document_name",
+	"document"."meta" AS "meta",
 	"document"."hit_handling_method" AS "hit_handling_method",
 	"document"."directly_return_similarity" as "directly_return_similarity"
 FROM

ui/src/components/ai-chat/ExecutionDetailDialog.vue

@@ -28,7 +28,7 @@
               <div class="flex align-center">
                 <span
                   class="mr-16 color-secondary"
-                  v-if="item.type === WorkflowType.Question || item.type === WorkflowType.AiChat"
+                  v-if="item.type === WorkflowType.Question || item.type === WorkflowType.AiChat || item.type === WorkflowType.ImageUnderstandNode"
                   >{{ item?.message_tokens + item?.answer_tokens }} tokens</span
                 >
                 <span class="mr-16 color-secondary">{{ item?.run_time?.toFixed(2) || 0.0 }} s</span>

ui/src/components/ai-chat/KnowledgeSource.vue

@@ -9,15 +9,28 @@
   </div>
   <div class="mt-8" v-if="!isWorkFlow(props.type)">
     <el-space wrap>
-      <el-button
+      <div v-for="(paragraph, index) in uniqueParagraphList" :key="index">
-        v-for="(dataset, index) in data.dataset_list"
+        <el-icon class="mr-4" :size="25">
-        :key="index"
+          <img :src="getIconPath(paragraph.document_name)" style="width: 90%" alt="" />
-        size="small"
+        </el-icon>
-        class="source_dataset-button"
+        <span
-        @click="openParagraph(data, dataset.id)"
+          v-if="!paragraph.source_url"
+          class="ellipsis"
+          :title="paragraph?.document_name?.trim()"
         >
-        <span class="ellipsis" :title="dataset.name"> {{ dataset.name }}</span>
+          {{ paragraph?.document_name }}
-      </el-button>
+        </span>
+        <a
+          v-else
+          @click="openLink(paragraph.source_url)"
+          class="ellipsis"
+          :title="paragraph?.document_name?.trim()"
+        >
+          <span :title="paragraph?.document_name?.trim()">
+            {{ paragraph?.document_name }}
+          </span>
+        </a>
+      </div>
     </el-space>
   </div>
 
@@ -42,7 +55,7 @@
   <ExecutionDetailDialog ref="ExecutionDetailDialogRef" />
 </template>
 <script setup lang="ts">
-import { ref } from 'vue'
+import { computed, ref } from 'vue'
 import ParagraphSourceDialog from './ParagraphSourceDialog.vue'
 import ExecutionDetailDialog from './ExecutionDetailDialog.vue'
 import { isWorkFlow } from '@/utils/application'
@@ -57,6 +70,15 @@ const props = defineProps({
     default: ''
   }
 })
+const iconMap: { [key: string]: string } = {
+  doc: '../../assets/doc-icon.svg',
+  docx: '../../assets/docx-icon.svg',
+  pdf: '../../assets/pdf-icon.svg',
+  md: '../../assets/md-icon.svg',
+  txt: '../../assets/txt-icon.svg',
+  xls: '../../assets/xls-icon.svg',
+  xlsx: '../../assets/xlsx-icon.svg'
+}
 
 const ParagraphSourceDialogRef = ref()
 const ExecutionDetailDialogRef = ref()
@@ -66,6 +88,40 @@ function openParagraph(row: any, id?: string) {
 function openExecutionDetail(row: any) {
   ExecutionDetailDialogRef.value.open(row)
 }
+const uniqueParagraphList = computed(() => {
+  const seen = new Set()
+  return (
+    props.data.paragraph_list?.filter((paragraph: any) => {
+      const key = paragraph.document_name.trim()
+      if (seen.has(key)) {
+        return false
+      }
+      seen.add(key)
+      // 判断如果 meta 属性不是 {} 需要json解析 转对象
+      if (paragraph.meta && typeof paragraph.meta === 'string') {
+        paragraph.meta = JSON.parse(paragraph.meta)
+        paragraph.source_url = paragraph.meta.source_url
+      }
+      return true
+    }) || []
+  )
+})
+
+function getIconPath(documentName: string) {
+  const extension = documentName.split('.').pop()?.toLowerCase()
+  if (!documentName || !extension) return new URL(`${iconMap['doc']}`, import.meta.url).href
+  if (iconMap && extension && iconMap[extension]) {
+    return new URL(`${iconMap[extension]}`, import.meta.url).href
+  }
+  return new URL(`${iconMap['doc']}`, import.meta.url).href
+}
+function openLink(url: string) {
+  // 如果url不是以/结尾，加上/
+  if (url && !url.endsWith('/')) {
+    url += '/'
+  }
+  window.open(url, '_blank')
+}
 </script>
 <style lang="scss" scoped>
 .source_dataset-button {

ui/src/components/ai-chat/component/ParagraphCard.vue

@@ -18,10 +18,48 @@
     <template #footer>
       <div class="footer-content flex-between">
         <el-text class="flex align-center" style="width: 70%">
-          <el-icon class="mr-4">
+          <el-icon class="mr-4" :size="25">
-            <Document />
+            <img
+              src="@/assets/doc-icon.svg"
+              style="width: 90%"
+              alt=""
+              v-if="data?.document_name?.includes('doc')"
+            />
+            <img
+              src="@/assets/docx-icon.svg"
+              style="width: 90%"
+              alt=""
+              v-else-if="data?.document_name?.includes('docx')"
+            />
+            <img
+              src="@/assets/pdf-icon.svg"
+              style="width: 90%"
+              alt=""
+              v-else-if="data?.document_name?.includes('pdf')"
+            />
+            <img
+              src="@/assets/md-icon.svg"
+              style="width: 90%"
+              alt=""
+              v-else-if="data?.document_name?.includes('md')"
+            />
+            <img
+              src="@/assets/xls-icon.svg"
+              style="width: 90%"
+              alt=""
+              v-else-if="data?.document_name?.includes('xls')"
+            />
+            <img
+              src="@/assets/txt-icon.svg"
+              style="width: 90%"
+              alt=""
+              v-else-if="data?.document_name?.includes('txt')"
+            />
+            <img src="@/assets/doc-icon.svg" style="width: 90%" alt="" v-else />
           </el-icon>
-          <span class="ellipsis" :title="data?.document_name?.trim()"> {{ data?.document_name.trim() }}</span>
+          <span class="ellipsis" :title="data?.document_name?.trim()">
+            {{ data?.document_name.trim() }}</span
+          >
         </el-text>
         <div class="flex align-center" style="line-height: 32px">
           <AppAvatar class="mr-8 avatar-blue" shape="square" :size="18">
@@ -45,6 +83,23 @@ const props = defineProps({
     default: 0
   }
 })
+const iconMap: { [key: string]: string } = {
+  doc: '../../assets/doc-icon.svg',
+  docx: '../../assets/docx-icon.svg',
+  pdf: '../../assets/pdf-icon.svg',
+  md: '../../assets/md-icon.svg',
+  txt: '../../assets/txt-icon.svg',
+  xls: '../../assets/xls-icon.svg',
+  xlsx: '../../assets/xlsx-icon.svg'
+}
+function getIconPath(documentName: string) {
+  const extension = documentName.split('.').pop()?.toLowerCase()
+  if (!documentName || !extension) return new URL(`${iconMap['doc']}`, import.meta.url).href
+  if (iconMap && extension && iconMap[extension]) {
+    return new URL(`${iconMap[extension]}`, import.meta.url).href
+  }
+  return new URL(`${iconMap['doc']}`, import.meta.url).href
+}
 </script>
 <style lang="scss" scoped>
 .paragraph-source-card-height {

ui/src/workflow/nodes/base-node/component/UserFieldFormDialog.vue

@@ -117,6 +117,8 @@ const currentIndex = ref(null)
 const inputTypeList = ref([
   { label: '文本框', value: 'TextInputConstructor' },
   { label: '单选框', value: 'SingleSelectConstructor' },
+  { label: '多选框', value: 'MultiSelectConstructor' },
+  { label: '选项卡', value: 'RadioCardConstructor' },
   { label: '日期', value: 'DatePickerConstructor' }
 ])
 

ui/src/workflow/nodes/base-node/component/UserInputFieldTable.vue

@@ -26,6 +26,8 @@
         <el-tag type="info" class="info-tag" v-if="row.input_type === 'Slider'">滑块</el-tag>
         <el-tag type="info" class="info-tag" v-if="row.input_type === 'SwitchInput'">开关</el-tag>
         <el-tag type="info" class="info-tag" v-if="row.input_type === 'SingleSelect'">单选框</el-tag>
+        <el-tag type="info" class="info-tag" v-if="row.input_type === 'MultiSelect'">多选框</el-tag>
+        <el-tag type="info" class="info-tag" v-if="row.input_type === 'RadioCard'">选项卡</el-tag>
         <el-tag type="info" class="info-tag" v-if="row.input_type === 'DatePicker'">日期</el-tag>
       </template>
     </el-table-column>

ui/src/workflow/nodes/base-node/index.vue

@@ -395,8 +395,8 @@ const switchFileUpload = () => {
 
   if (form_data.value.file_upload_enable) {
     form_data.value.file_upload_setting = form_data.value.file_upload_setting || default_upload_setting
-    props.nodeModel.graphModel.eventCenter.emit('refreshFileUploadConfig')
   }
+  props.nodeModel.graphModel.eventCenter.emit('refreshFileUploadConfig')
 }
 const openFileUploadSettingDialog = () => {
   FileUploadSettingDialogRef.value?.open(form_data.value.file_upload_setting)

ui/src/workflow/nodes/start-node/index.vue

@@ -65,12 +65,16 @@ const refreshFileUploadConfig = () => {
   let fields = cloneDeep(props.nodeModel.properties.config.fields)
   const form_data = props.nodeModel.graphModel.nodes
     .filter((v: any) => v.id === 'base-node')
+    .filter((v: any) => v.properties.node_data.file_upload_enable)
     .map((v: any) => cloneDeep(v.properties.node_data.file_upload_setting))
     .filter((v: any) => v)
+
+  fields = fields.filter((item: any) => item.value !== 'image' && item.value !== 'document')
+
   if (form_data.length === 0) {
+    set(props.nodeModel.properties.config, 'fields', fields)
     return
   }
-  fields = fields.filter((item: any) => item.value !== 'image' && item.value !== 'document')
   let fileUploadFields = []
   if (form_data[0].document) {
     fileUploadFields.push({ label: '文档', value: 'document' })