Request: `curl` return early if model isn't local, with better message

[jamesbraza]

LocalAI version:

https://github.com/go-skynet/LocalAI/tree/30f120ee6a7979aa4ffbd94ccdc1e58b94155a44

Environment, CPU architecture, OS, and Version:

Darwin N7L493PWK4 22.6.0 Darwin Kernel Version 22.6.0: Wed Jul  5 22:22:05 PDT 2023; root:xnu-8796.141.3~6/RELEASE_ARM64_T6000 arm64

Describe the bug

When placing a model name that doesn't exist, two issues happen:

1. The curl takes over 1 minute
2. It prints all backends returned error: 23 errors occurred

To Reproduce

I am using a model name foo.bin (chosen because it doesn't match any downloaded):

> curl http://localhost:8080/models
{"object":"list","data":[{"id":"thebloke__wizardlm-13b-v1-0-uncensored-superhot-8k-ggml__wizardlm-13b-v1.0-superhot-8k.ggmlv3.q4_k_m.bin","object":"model"}]}
> curl http://localhost:8080/v1/completions -H "Content-Type: application/json" -d '{
     "model": "foo.bin",
     "prompt": "A long time ago in a galaxy far, far away",
     "temperature": 0.7
   }'
{"error":{"code":500,"message":"could not load model - all backends returned error: 23 errors occurred:\n\t* could not load model: rpc error: code = Unknown desc = failed loading model\n\t* could not load model: rpc error: code = Unknown desc = failed loading model\n\t* could not load model: rpc error: code = Unknown desc = failed loading model\n\t* could not load model: rpc error: code = Unknown desc = failed loading model\n\t* could not load model: rpc error: code = Unknown desc = failed loading model\n\t* could not load model: rpc error: code = Unknown desc = failed loading model\n\t* could not load model: rpc error: code = Unknown desc = failed loading model\n\t* could not load model: rpc error: code = Unknown desc = failed loading model\n\t* could not load model: rpc error: code = Unknown desc = failed loading model\n\t* could not load model: rpc error: code = Unknown desc = failed loading model\n\t* could not load model: rpc error: code = Unknown desc = failed loading model\n\t* could not load model: rpc error: code = Unknown desc = failed loading model\n\t* could not load model: rpc error: code = Unknown desc = failed loading model\n\t* could not load model: rpc error: code = Unknown desc = failed loading model\n\t* could not load model: rpc error: code = Unavailable desc = error reading from server: EOF\n\t* could not load model: rpc error: code = Unknown desc = stat /models/foo.bin: no such file or directory\n\t* could not load model: rpc error: code = Unknown desc = stat /models/foo.bin: no such file or directory\n\t* could not load model: rpc error: code = Unknown desc = unsupported model type /models/foo.bin (should end with .onnx)\n\t* backend unsupported: /build/extra/grpc/huggingface/huggingface.py\n\t* backend unsupported: /build/extra/grpc/autogptq/autogptq.py\n\t* backend unsupported: /build/extra/grpc/bark/ttsbark.py\n\t* backend unsupported: /build/extra/grpc/diffusers/backend_diffusers.py\n\t* backend unsupported: /build/extra/grpc/exllama/exllama.py\n\n","type":""}}

Formatting that message:

could not load model - all backends returned error: 23 errors occurred:
* could not load model: rpc error: code = Unknown desc = failed loading model
* could not load model: rpc error: code = Unknown desc = failed loading model
* could not load model: rpc error: code = Unknown desc = failed loading model
* could not load model: rpc error: code = Unknown desc = failed loading model
* could not load model: rpc error: code = Unknown desc = failed loading model
* could not load model: rpc error: code = Unknown desc = failed loading model
* could not load model: rpc error: code = Unknown desc = failed loading model
* could not load model: rpc error: code = Unknown desc = failed loading model
* could not load model: rpc error: code = Unknown desc = failed loading model
* could not load model: rpc error: code = Unknown desc = failed loading model
* could not load model: rpc error: code = Unknown desc = failed loading model
* could not load model: rpc error: code = Unknown desc = failed loading model
* could not load model: rpc error: code = Unknown desc = failed loading model
* could not load model: rpc error: code = Unknown desc = failed loading model
* could not load model: rpc error: code = Unavailable desc = error reading from server: EOF
* could not load model: rpc error: code = Unknown desc = stat /models/foo.bin: no such file or directory
* could not load model: rpc error: code = Unknown desc = stat /models/foo.bin: no such file or directory
* could not load model: rpc error: code = Unknown desc = unsupported model type /models/foo.bin (should end with .onnx)
* backend unsupported: /build/extra/grpc/huggingface/huggingface.py
* backend unsupported: /build/extra/grpc/autogptq/autogptq.py
* backend unsupported: /build/extra/grpc/bark/ttsbark.py
* backend unsupported: /build/extra/grpc/diffusers/backend_diffusers.py
* backend unsupported: /build/extra/grpc/exllama/exllama.py

Expected behavior

A desirable behavior is:

1. curl returns quickly (<1 sec) when non-installed model
2. Response has a better error message, saying the model doesn't match any locally installed model

In other words, the current behavior is a massive error message, and takes forever to actually show up. Ideally if a model isn't installed, a response comes immediately with a concise error message

[mudler]

we should definetly enhance logging but also the logic here around loading.

Going to treat this as a feature request. We should at least not try at all the model if the file doesn't actually exist.