diff --git a/web/src/pages/add-knowledge/components/knowledge-setting/utils.ts b/web/src/pages/add-knowledge/components/knowledge-setting/utils.ts index 3c4f94f43..ceb9cc97e 100644 --- a/web/src/pages/add-knowledge/components/knowledge-setting/utils.ts +++ b/web/src/pages/add-knowledge/components/knowledge-setting/utils.ts @@ -17,111 +17,3 @@ export const ImageMap = { one: getImageName('one', 2), knowledge_graph: getImageName('knowledge-graph', 2), }; - -export const TextMap = { - book: { - title: '', - description: `

Supported file formats are DOCX, PDF, TXT.

- Since a book is long and not all the parts are useful, if it's a PDF, - please setup the page ranges for every book in order eliminate negative effects and save computing time for analyzing.

`, - }, - laws: { - title: '', - description: `

Supported file formats are DOCX, PDF, TXT.

- Legal documents have a very rigorous writing format. We use text feature to detect split point. -

- The chunk granularity is consistent with 'ARTICLE', and all the upper level text will be included in the chunk. -

`, - }, - manual: { - title: '', - description: `

Only PDF is supported.

- We assume manual has hierarchical section structure. We use the lowest section titles as pivots to slice documents. - So, the figures and tables in the same section will not be sliced apart, and chunk size might be large. -

`, - }, - naive: { - title: '', - description: `

Supported file formats are DOCX, EXCEL, PPT, IMAGE, PDF, TXT.

-

This method apply the naive ways to chunk files:

-

-

  • Successive text will be sliced into pieces using vision detection model.
    • -
  • Next, these successive pieces are merge into chunks whose token number is no more than 'Token number'.

    • `, - }, - paper: { - title: '', - description: `

    Only PDF file is supported.

    - If our model works well, the paper will be sliced by it's sections, like abstract, 1.1, 1.2, etc.

    - The benefit of doing this is that LLM can better summarize the content of relevant sections in the paper, - resulting in more comprehensive answers that help readers better understand the paper. - The downside is that it increases the context of the LLM conversation and adds computational cost, - so during the conversation, you can consider reducing the ‘topN’ setting.

    `, - }, - presentation: { - title: '', - description: `

    The supported file formats are PDF, PPTX.

    - Every page will be treated as a chunk. And the thumbnail of every page will be stored.

    - All the PPT files you uploaded will be chunked by using this method automatically, setting-up for every PPT file is not necessary.

    `, - }, - qa: { - title: '', - description: `

    EXCEL and CSV/TXT files are supported.

    - If the file is in excel format, there should be 2 columns question and answer without header. - And question column is ahead of answer column. - And it's O.K if it has multiple sheets as long as the columns are rightly composed.

    - - If it's in csv format, it should be UTF-8 encoded. Use TAB as delimiter to separate question and answer.

    - - All the deformed lines will be ignored. - Every pair of Q&A will be treated as a chunk.

    `, - }, - resume: { - title: '', - description: `

    The supported file formats are DOCX, PDF, TXT. -

    - The résumé comes in a variety of formats, just like a person’s personality, but we often have to organize them into structured data that makes it easy to search. -

    - Instead of chunking the résumé, we parse the résumé into structured data. As a HR, you can dump all the résumé you have, - the you can list all the candidates that match the qualifications just by talk with 'RAGFlow'. -

    - `, - }, - table: { - title: '', - description: `

    EXCEL and CSV/TXT format files are supported.

    - Here're some tips: -

    `, - }, - picture: { - title: '', - description: ` -

    Image files are supported. Video is coming soon.

    - If the picture has text in it, OCR is applied to extract the text as its text description. -

    - If the text extracted by OCR is not enough, visual LLM is used to get the descriptions. -

    `, - }, - one: { - title: '', - description: ` -

    Supported file formats are DOCX, EXCEL, PDF, TXT. -

    - For a document, it will be treated as an entire chunk, no split at all. -

    - If you want to summarize something that needs all the context of an article and the selected LLM's context length covers the document length, you can try this method. -

    `, - }, -}; diff --git a/web/src/pages/flow/chat/drawer.tsx b/web/src/pages/flow/chat/drawer.tsx index baeec8a6f..190fda553 100644 --- a/web/src/pages/flow/chat/drawer.tsx +++ b/web/src/pages/flow/chat/drawer.tsx @@ -1,6 +1,7 @@ import { useFetchFlow } from '@/hooks/flow-hooks'; import { IModalProps } from '@/interfaces/common'; import { Drawer } from 'antd'; +import { getDrawerWidth } from '../utils'; import FlowChatBox from './box'; const ChatDrawer = ({ visible, hideModal }: IModalProps) => { @@ -13,7 +14,7 @@ const ChatDrawer = ({ visible, hideModal }: IModalProps) => { onClose={hideModal} open={visible} getContainer={false} - width={window.innerWidth > 1278 ? '40%' : 470} + width={getDrawerWidth()} mask={false} // zIndex={10000} > diff --git a/web/src/pages/flow/flow-drawer/index.tsx b/web/src/pages/flow/flow-drawer/index.tsx index 4304146f2..fe5c7c642 100644 --- a/web/src/pages/flow/flow-drawer/index.tsx +++ b/web/src/pages/flow/flow-drawer/index.tsx @@ -39,6 +39,7 @@ import OperatorIcon from '../operator-icon'; import { CloseOutlined } from '@ant-design/icons'; import { lowerFirst } from 'lodash'; +import { getDrawerWidth } from '../utils'; import styles from './index.less'; interface IProps { @@ -135,7 +136,7 @@ const FlowDrawer = ({ open={visible} getContainer={false} mask={false} - width={500} + width={getDrawerWidth()} closeIcon={null} >
    diff --git a/web/src/pages/flow/form/generate-form/index.tsx b/web/src/pages/flow/form/generate-form/index.tsx index f286169a7..84676e83b 100644 --- a/web/src/pages/flow/form/generate-form/index.tsx +++ b/web/src/pages/flow/form/generate-form/index.tsx @@ -28,7 +28,7 @@ const GenerateForm = ({ onValuesChange, form, node }: IOperatorForm) => { { +const FlowHeader = ({ showChatDrawer, chatDrawerVisible }: IProps) => { const { saveGraph } = useSaveGraph(); const handleRun = useSaveGraphBeforeOpeningDebugDrawer(showChatDrawer); const { data } = useFetchFlow(); @@ -28,7 +29,7 @@ const FlowHeader = ({ showChatDrawer }: IProps) => { } = useSetModalState(); const { visible, hideModal, showModal } = useSetModalState(); const { id } = useParams(); - const time = useWatchAgentChange(); + const time = useWatchAgentChange(chatDrawerVisible); return ( <> diff --git a/web/src/pages/flow/hooks.ts b/web/src/pages/flow/hooks.ts index 8a004e413..aa89832af 100644 --- a/web/src/pages/flow/hooks.ts +++ b/web/src/pages/flow/hooks.ts @@ -681,7 +681,7 @@ export const useCopyPaste = () => { }, [onPasteCapture]); }; -export const useWatchAgentChange = () => { +export const useWatchAgentChange = (chatDrawerVisible: boolean) => { const [time, setTime] = useState(); const nodes = useGraphStore((state) => state.nodes); const edges = useGraphStore((state) => state.edges); @@ -697,9 +697,11 @@ export const useWatchAgentChange = () => { }, [flowDetail, setSaveTime]); const saveAgent = useCallback(async () => { - const ret = await saveGraph(); - setSaveTime(ret.data.update_time); - }, [saveGraph, setSaveTime]); + if (!chatDrawerVisible) { + const ret = await saveGraph(); + setSaveTime(ret.data.update_time); + } + }, [chatDrawerVisible, saveGraph, setSaveTime]); useDebounceEffect( () => { diff --git a/web/src/pages/flow/index.tsx b/web/src/pages/flow/index.tsx index 980c08b01..c73dcdbff 100644 --- a/web/src/pages/flow/index.tsx +++ b/web/src/pages/flow/index.tsx @@ -25,7 +25,10 @@ function RagFlow() { - + { form, }; }; + +export const getDrawerWidth = () => { + return window.innerWidth > 1278 ? '40%' : 470; +};