I follow the guide at https://typesense.org/docs/guide/semantic-search.html#step-1-create-a-collection when i try to create a collection with the provided curl command, i get a

{"message": "Bad request."}

Response. In the typesense log there is this entry:

E20250922 21:31:42.776067 51644 batched_indexer.cpp:302] Exception while calling handler collections:create
E20250922 21:31:42.776134 51644 batched_indexer.cpp:303] Raw error: /root/.cache/bazel/_bazel_root/45fa99e3ef9c0d8b8591cceacfe70dd5/execroot/__main__/external/onnx_runtime/onnxruntime/core/session/provider_bridge_ort.cc:1106 onnxruntime::Provider& onnxruntime::ProviderLibrary::Get() [ONNXRuntimeError] : 1 : FAIL : Failed to load library libonnxruntime_providers_cuda.so with error: libonnxruntime_providers_cuda.so: cannot open shared object file: No such file or directory

How i can solve this error?

@Voidi could you share the full code snippet you are trying to replicate? This error is when your machine tries to use gpu.

The command to create the collection with the request body as attachment

curl "${TYPESENSE_HOST}/collections" \                                                                                                                                   ⏎ 
      -X POST \
      -H "Content-Type: application/json" \
      -H "X-TYPESENSE-API-KEY: ${TYPESENSE_API_KEY}" -d @files.json

Is it still correct that typesense only can use CUDA11 as written here https://github.com/typesense/typesense/issues/1876

About the CUDA, yes, I think so. You can use embeddings without GPU. We have several machines using it. It will just be slower to generate the embeddings. Should be OK on a small collection

How do I configure typesense to not use the gpu?

Typesense does not use GPU by default. So if you restart the process without the GPU flags, it will only use the CPU

I don't change anything in the configuration file, just installed typesense via the provided .deb file. (i use Ubuntu as OS) I installed typesense at my development computer which doesn't even have a nvidia graphics card (i just read that only nvidia is supported).

Could you try using docker? I think typesense is identifying that you have GPU and enabling it