feat: use milvus in chat llamaindex #4
Conversation
|
The latest updates on your projects. Learn more about Vercel for Git ↗︎
|
app/api/chat/engine/index.ts
Outdated
| ); | ||
| } | ||
|
|
||
| return await VectorStoreIndex.fromVectorStore(store); |
There was a problem hiding this comment.
Can we move a shared function getIndex to shared.ts?
app/api/chat/engine/generate.ts
Outdated
| }); | ||
| const storageContext = await storageContextFromDefaults({ vectorStore }); |
There was a problem hiding this comment.
Try using the shared function getIndex here and use is the index directly in fromDocuments below (without storage context)
There was a problem hiding this comment.
Yes, we can do:
const index = await getIndex(datasource);
const documents = await getDocuments(datasource);
// Set private=false to mark the document as public (required for filtering)
documents.forEach((doc) => {
doc.metadata["private"] = "false";
});
await runPipeline(index, documents);
But inside runPipeline function, we are setting private = true (we need to set private = false when generating public documents)
I guess we shouldn't append metadata to documents inside runPipeline function (can move it outside)
There was a problem hiding this comment.
I just updated to split the getIndex function. Thank you! That's a cool idea that solves my pain.
Previously, when creating a new bot, users had to manually create a collection to start chatting.
But now, when creating a new bot, they can start by uploading a private file. It will automatically create a collection on Milvus
thucpn
force-pushed
the
feat/use-milvus-in-chat-llamaindex
branch
from
256bd96 to
3175f76
Compare
| const isCollectionExist = await getMilvusClient().hasCollection({ | ||
| collection_name: datasource, | ||
| }); | ||
| if (!isCollectionExist.value) { | ||
| throw new Error( | ||
| `Collection "${datasource}" does not exist! Run the generate script or try uploading a file.`, | ||
| ); | ||
| } |
There was a problem hiding this comment.
how about we move this check into getIndex?
There was a problem hiding this comment.
It won't work when running generate (because when generating we will create a new collection)
There was a problem hiding this comment.
how about adding a checkExists parameter to getIndex then?
getIndex({datasource, checkExists: false})
| } | ||
| } | ||
|
|
||
| export async function getIndex(datasource: string) { |
There was a problem hiding this comment.
to mimize changes with base branch, call this getDatasource and move it to index.ts
| ) { | ||
| // Update documents with metadata |
There was a problem hiding this comment.
good we can also do this in CL directly
app/api/chat/upload/upload.ts
Outdated
| @@ -14,6 +14,16 @@ export async function uploadDocument( | |||
| const fileBuffer = Buffer.from(content, "base64"); | |||
| const documents = await loadDocuments(fileBuffer, mimeType); | |||
| const { filename } = await saveDocument(fileBuffer, mimeType); | |||
| const index = await getDataSource(datasource); | |||
| return await runPipeline(index, documents, filename); | |||
| const index = await getIndex(datasource); | |||
There was a problem hiding this comment.
good we can also do this in CL directly
| @@ -20,7 +20,7 @@ npx -y create-llama@0.1.25 \ | |||
| --post-install-action none \ | |||
| --no-llama-parse \ | |||
| --example-file \ | |||
| --vector-db none \ | |||
| --vector-db milvus \ | |||
There was a problem hiding this comment.
goal: we just need to change this from none to milvus to use Milvus in chat llamaindex
|
Open new PR here: #5 |
No description provided.