Hello 🙂 Is there no UI option to create vector index? I am getting this screen. the UI comes for a
Hello 🙂
Is there no UI option to create vector index? I am getting this screen. the UI comes for a split second and goes away
12 Replies
ok. I can create via cli. please ignore my question. :blob_sweat:
Hi. I found this in the vectorize docs. I have 2 questions -
1. How long do I need to wait before sending next 5000embeddings after 1st batch so that I don't get rate limited. I am trying to insert and later query via HTTP API.
2. The global limit says 1200req/5min which is too low. It means, on avg only 4 queries can be done on the vectorize index per second right? This limit seems too low or I am not understanding it properly. If I query the index via worker, does this limit not apply in that case?
thanks @Leo ya I also thought so at the end
I have to offload the vector insert/query part to a worker, because I want to write the main backend in go ðŸ˜
Hi @Bob. Thank you for reaching out!
1. You can batch upto 5000 vectors in a single request to Vectorize if you would like to insert a large number of vectors quickly. And you will not be rate-limited as long as you stay within the limit of 1200 requests (each request can contain 5k vectors) per 5 minutes if you are using the HTTP API.
2. If you are concerned about the rate-limiting over HTTP API for Vectorize queries, and if you do not wish to use Vectorize bindings via Workers/Pages for your use case, could you please DM me and describe your requirements in greater detail. We will do our best to figure out a solution for you.
Hello,
I am thinking of using CF Vectorize databases but can't find the docs for max queries/requests per second per namespace/globally. Do you know where I can get this information?
We need to be able to handle at least 10,000 queries/second, and want to make sure that CF can support that.
Thanks!
Is it possible to create vectorize database dynamically? I want to create a platform where my customers get their own vectorize database, the 5 million vector limit might become a problem if I use one index with difference namespaces. Is there a way to create and get indexes dynamically similar to DOs.
You can use the REST API rather than a binding, as one option ( https://developers.cloudflare.com/api/resources/vectorize/subresources/indexes/methods/create/ )
Cloudflare API | Vectorize › Indexes › Create Vectorize Index
Interact with Cloudflare's products and services via the Cloudflare API
Hello anyone can help me with a vectorise problem VECTOR_QUERY_ERROR (code = 50111)
the weird thing is that with this payload below it works as expected
{
"search_prompt": "i saw that users wanted our company phone number, and couldnt find it.",
"function_type": "search",
"access_level": "internal",
"vector_search": true,
"filter": "kaizens",
"db_source": "vector"
}
but with this search prompt it throws the error
{
"search_prompt": "i saw .",
"function_type": "search",
"access_level": "internal",
"vector_search": true,
"filter": "kaizens",
"db_source": "vector"
}
i also need to mention this only happends on the kaizens filter and the weird behaviour does not have a pattern or something we can explain. for example it sometimes works with 3 words on the search_prompt and sometimes it doesn't.From what we can tell it is totally random and we have no clue why "heythere" works and "hey there" does not work. I would appreciate any feedback. Thanks!
Hi, I'm hitting the Vectorise Upsert limit, appears to around 1000 requests per min. Does that sound about right? I couldn't find any documenation on it but was hopping to do considerably more. "Error: VECTOR_UPSERT_ERROR (code = 971): Please wait and consider throttling your request speed". If i was to request an increase, is there a upper bound for what can be supported?
I'm also getting very high latency on 30s to get a response back from env.VECTORIZE.upsert(vectors); with a batch of 100 vectors of 768 size. I understood that this api was only queuing the work? and it's eventually consistent So why so high to call the add api?
Ah I'm underestimating the JSON payload size of these vectors and the RPC overhead. I've if i limit the decimal places then I can get 20k rpm at batches of 100.
I've been reading and for my use case, I'm wondering if I'll need to combine both workers and the cloudflare api: I want each of my users to have their own index but I want the users of my users to be able to query my user's index and it could exceed the cloudflare api limit so I would like the queries/inserting to be done via cloudflare workers, but how can I dynamically pass in the name of the index
Hi, I wanted to know about pricing of vectorization:
There are some confusing things in vectorization:
This is for a month:
Let's say I am making 10 queries of some documents where vector stored amount is: 10k and there are 100 those documents. and dimension is 376.
Then how will the queried vector dimension will be calculated:
- if we calculate queried vector dimension of 10 queries for one document then: it will be: (10,000+ 10)376 = 3763760
Now for 100 documents it will be: 3763760100 = 376376000
but now it consider whole 100 documents queries at one level:
total queries will be: (100 ) (10) = 1000
Queries dimension for all 100 : (10,000+1000)*376 = 4136000
Which one is correct ?
Hi, I have a worker deployed utilizing vectorize. It's deployed bound to a certain index but when I do my searches it also returns results from another index. Is this normal behaviour? Or what's up? Anyone had any troubles like that?
Bindings in the settings looking under the dashboard says "Vectorize index
VECTORIZE
bge-large-en"
And in swagger I get
{
"success": true,
"embeddings": {
"count": 3,
"matches": [
{
"id": "15",
"score": 0.86592543,
"namespace": "findu-job-desc-test1",
I feel certain that only the index you are currently utilizing should return results. Not other indexes. Or am I mistaken?