hi cloudflare team, are there any short term plans to support the multilingual BAAI/bge-m3 embedding

hi cloudflare team, are there any short term plans to support the multilingual BAAI/bge-m3 embedding model? the current BGE ones are english only. if it's not coming soon, i won't be able to use vectorize but if it's coming up then i can live with EN for now only!

Marco Aleixo•2/8/25, 9:35 PM

What is the roadmap for vectorize?
Do you plan to support multi dense vector like the ones from colpali?

RRafal is there anyone else running into `Error: VECTOR_QUERY_ERROR (code = 10000): Aut...

shoo•2/9/25, 5:30 PM

Facing the same issue - not sure what's causing it. query fails. Error: VECTOR_QUERY_ERROR (code = 10000): Authentication error

Bob•2/9/25, 7:27 PM

Hello, I have a question. It seems the indexing process takes few seconds. When we insert data, we get a mutation id. Is there any way to check against the mutation id if that batch got indexed or not?

beingmudit [WePlayDOS]•2/9/25, 8:42 PM

Damn! @Bob I joined this discord just to ask the same questions. Interesting that this is not just me experienceing indexing latency.

beingmudit [WePlayDOS]•2/9/25, 8:43 PM

Hey CF team!

We're building with Vectorize and noticed some interesting behavior around indexing latency and consistency. Here's what we're observing:

Indexing Latency:
Create index -> Success (immediate)
Add 2 vectors -> Success (immediate)
Search vectors -> No results for ~9 seconds
Search attempt timeline:
- 0s: No results
- 3s: No results
- 6s: No results
- 9s: Finally get results
Deletion Consistency:
After vectors are searchable
Delete vector "vec1" -> Success
Immediate search still returns both vectors including the deleted one
Subsequent searches also return the deleted vector

Test Case Details:

Index dimension: 32
Two simple test vectors:
- vec1: Array(32).fill(0.1)
- vec2: Array(32).fill(0.2)
Query vector: Array(32).fill(0.15)

Questions:

Is this expected latency for indexing?
Are there any best practices for handling the indexing delay?
Should deleted vectors remain searchable? If not, is there a known delay before deletions are reflected?
Are there any configuration options we should be using to optimize this?

Full test code available if needed. Thanks for any insights/help/

BBob Hello, I have a question. It seems the indexing process takes few seconds. When ...

Utkarsh Saxena•2/10/25, 5:23 AM

I think, You can check with getByIDs, as the insertion in index is async now in v2.

beingmudit [WePlayDOS]•2/10/25, 7:23 AM

Can you share the docs where it mentions something related to this? This seems like a big limitations given other services are super duper fast.

Utkarsh Saxena•2/10/25, 8:13 AM

Sure, You can see this: https://developers.cloudflare.com/vectorize/reference/transition-vectorize-legacy/#notable-changes
It is mentioned it takes few seconds to reflect and is asynchronous now.

Cloudflare Docs

Transition legacy Vectorize indexes · Vectorize

Legacy Vectorize (V1) indexes are on a deprecation path as of Aug 15, 2024. Your Vectorize index may be a legacy index if it fulfills any of the follwing crieria:

Sshoo Facing the same issue - not sure what's causing it. query fails. Error: VECTOR_Q...

cloudkite•2/10/25, 8:17 AM

Also getting this intermittently

beingmudit [WePlayDOS]•2/10/25, 9:19 PM

Thanks Utkarsh! This helps but i still can't beleive the 6-9 sec latency :/ Can anyone from Cloudflare team answer/confirm this?

Bbeingmudit [WePlayDOS]Thanks Utkarsh! This helps but i still can't beleive the 6-9 sec latency :/ Can ...

Jerome•2/10/25, 9:44 PM

Hi! yes the updates are processed asynchronously, and can take some time before being reflected in queries; 6 to 9s processing loop sounds like normal operation.

beingmudit [WePlayDOS]•2/10/25, 9:48 PM

Hey Jerome, thanks for getting back. Do you think if it makes sense to have it documented somewhere in the public facing docs? This async behaviour is really tricky to find.
Also, Do you know if other services do have similar latencies?

Jerome•2/10/25, 9:52 PM

. Do you think if it makes sense to have it documented somewhere in the public facing docs?

It's in the doc clearly stated for insert (same for upsert):
https://developers.cloudflare.com/vectorize/reference/client-api/

Jerome•2/10/25, 9:53 PM

Also, Do you know if other services do have similar latencies?

You mean, other cloudflare services?

beingmudit [WePlayDOS]•2/10/25, 10:33 PM

Ahhhh, I missed this page entirely. Thank you for your help.
No, I meant services like pinecone etc?

JJerome Hi! yes the updates are processed asynchronously, and can take some time before ...

Marco Aleixo•2/11/25, 2:35 AM

maybe we can have a wait=True for inserts?
I mean, I can implement the "confirm that the vector was indexed" pooling on my side, but maybe having this in the api itself is more straightforward...?!

UUtkarsh Saxena Hi, I wanted to know about pricing of vectorization: There are some confusing th...

Utkarsh Saxena•2/11/25, 5:18 AM

Can anyone help me with this

MMarco Aleixo maybe we can have a wait=True for inserts? I mean, I can implement the "confirm ...

garvitg•2/11/25, 5:09 PM

The Vectorize Index Info route provides information on what mutations Vectorize has completed processing: https://developers.cloudflare.com/api/resources/vectorize/subresources/indexes/methods/info/. It includes the mutation id and the timestamp of the last mutation block that Vectorize has processed.

Cloudflare API | Vectorize › Indexes › Get Vectorize Index Info

Interact with Cloudflare's products and services via the Cloudflare API

Sshoo Facing the same issue - not sure what's causing it. query fails. Error: VECTOR_Q...

cloudkite•2/11/25, 6:01 PM

Bump for visibility, also getting this. Hopefully it’s being looked into

Victor•2/14/25, 12:17 AM

https://discord.com/channels/595317990191398933/1040420029080018945/1319662221957271605

So I make an index.
I delete the index.
I make an index with same id.
verify new index info
I try to upsert vector - fails saying index is deleted
I retry the exact same upsert api request (no change) - success

beingmudit [WePlayDOS]•2/14/25, 6:18 AM

Yes because the operations are async. When you get the response. it is just an ack not the actual operation.

Victor•2/15/25, 1:49 AM

I've had it happen even as far as an insert 24h after making the index again. And I get index info after creating it and verify it was created correctly

deo time•2/22/25, 12:31 AM

Hello, I've had a site that uses vectorize for a little over a year now, however just now requests to it have started failing with (error) Error: VECTOR_QUERY_ERROR (code = 4009): Bad Request: the index is disabled(error) Error: VECTOR_QUERY_ERROR (code = 4009): Bad Request: the index is disabled. I've also checked and it looks like my index isn't there anymore, does anyone know why this is happening?

deo time•2/22/25, 12:37 AM

I also get the error code vectorize.index.deletedvectorize.index.deleted when trying to get it via cli, so it looks like it was just forcefully deleted somehow?

ac•2/27/25, 7:48 PM

Hi, I am getting Error: Cloudflare LLM call failed with status code 403Error: Cloudflare LLM call failed with status code 403 trying to insert into a new Vectorize index on production. Works fine locally, and confirmed the bindings are there. It seems related to other auth issues people have posted about in here. Any idea what I'm missing?

grantpowell•3/3/25, 6:37 PM

I am unable to insert vectors into index

noad.eth•3/8/25, 12:39 PM

Hi. I just started using Vectorize this week. Really liking it so far. Nice and simple.

For inserting/upserting vectors using the REST API (or Typescript SDK), it would be nice if the request body could also accept an array of vector objects.

When using the TS package, if I set the body value to be the ndjson string, I was getting a "failed to parse" error. I found I had to wrap the ndjson in a Buffer.from()Buffer.from() to get it to work, but had to ignore a TS error since body is typed as a

string

string

Samko•3/17/25, 10:44 AM

Is vectorize still actively developed and improved https://developers.cloudflare.com/vectorize/platform/changelog/ seems bit outdated. I believe there was feature that would allow dynamic binding and creation of index from worker. Is this still planned or?

SSamko Is vectorize still actively developed and improved https://developers.cloudflare...

Matt Silverlock•3/18/25, 12:16 PM

Still worked on. Dynamic binding has some complexities - we need to solve for more than just Vectorize

Matt Silverlock•3/18/25, 12:17 PM

For many here: Workers AI now has a re-ranker model and a new embeddings models - https://developers.cloudflare.com/changelog/2025-03-17-new-workers-ai-models/

Cloudflare Docs

New models in Workers AI

Matt Silverlock•3/18/25, 12:17 PM

beingmudit [WePlayDOS]•3/21/25, 10:29 PM

Is vectorize down? I am getting Internal Server Errors. Is there a status page for those apis?

beingmudit [WePlayDOS]•3/21/25, 10:32 PM

{

"result": null,
"success": false,
"errors": [
{
"code": 50124,
"message": "vectorize.upstream_error - Internal Server Error"
}
],
"messages": []
}

Bbeingmudit [WePlayDOS]Is vectorize down? I am getting Internal Server Errors. Is there a status page f...

ggu•3/22/25, 2:41 AM

Yeah, we had some heightened error rates due to an upstream dependency error. See status here https://www.cloudflarestatus.com/incidents/v6c7l22vglw5

Increased HTTP Errors for Some Cloudflare Services

Gggu Yeah, we had some heightened error rates due to an upstream dependency error. Se...

Isaac McFadyen•3/22/25, 3:11 AM

Was this R2 by chance?

Isaac McFadyen•3/22/25, 3:22 AM

Cloudflare is investigating an increase in HTTP 5xx errors with R2

Ignore me

Gggu Yeah, we had some heightened error rates due to an upstream dependency error. Se...

MattIPv4•3/22/25, 3:39 AM

Why wasn't Vectorize in the impacted products list for that incident?

MattIPv4•3/22/25, 3:39 AM

Or rather I suppose, why isn't Vectorize on the status page at all unless I've missed it?

IIsaac McFadyen Was this R2 by chance?

ggu•3/22/25, 3:40 AM

Yes

MMattIPv4 Or rather I suppose, why isn't Vectorize on the status page at all unless I've m...

ggu•3/22/25, 3:40 AM

Hmm that’s odd. Will look into it

MattIPv4•3/22/25, 3:40 AM

Appreciate it, ty :MeowHeartCloudflare:

yevgen•3/24/25, 6:38 PM

I've asked internally to add Vectorize to the list of affected services

Nick•3/25/25, 9:19 PM

Hey everyone, I have a question about the pricing model for Vectorize, specifically regarding metadata filtering. I structure my data by using metadata to filter vectors into different collections within a single index. For instance, one index might hold 1 million vectors organized into 1,000 collections, and every query I run is filtered by metadata to target a specific collection.

Does that mean I’m billed based only on the vector dimensions of the filtered collection, or do I incur costs based on the entire index’s dimensions regardless of the metadata filter?

Thanks for any insights!

NNick Hey everyone, I have a question about the pricing model for Vectorize, specifica...

yevgen•3/26/25, 9:32 AM

you are billed based on the entire index's dimensions.

Yyevgen you are billed based on the entire index's dimensions.

lavisht22•3/31/25, 12:01 PM

Are there any suggestions to go around this?

I want to create an index with 1M vectors which will have around 10K namespaces and each query only needs one subset. If entire index's dimensions are used, we would end up with a huge bill.

Llavisht22 Are there any suggestions to go around this? I want to create an index with 1M ...

Matt Silverlock•3/31/25, 12:05 PM

If you store 1M vectors of 1536 dims and query it 1M times a month, it’s about ~$47 - https://developers.cloudflare.com/vectorize/platform/pricing/#usage-examples

Cloudflare Docs

Pricing · Cloudflare Vectorize docs

Vectorize bills are based on:

MMatt Silverlock If you store 1M vectors of 1536 dims and query it 1M times a month, it’s about ~...

lavisht22•3/31/25, 12:35 PM

I think I did my math wrong, let me check this again. Thanks!

lavisht22•4/2/25, 10:41 AM

I have an index with almost 1M vectors. I have also created a metadata index on one of the properties. When i query this index with a metadata filter, I am getting back results in 2-3 seconds, sometimes more. Is this kind of latency normal with Vectorize or am I doing something wrong?

hi cloudflare team, are there any short term plans to support the multilingual BAAI/bge-m3 embedding

Similar Threads