.Net: Fix hugging face embedding

[N-E-W-T-O-N]

Motivation and Context

Fix the Bug #6635 HuggingFace Embedding: Unable to Deserialization for certain models

Description

As mentioned in the issue, the HuggingFace Embedding API interface returns responses typically in the form of List<ReadOnlyMemory<float>> and occasionally as List<List<List<ReadOnlyMemory<float>>>>. Currently, only the latter format is handled correctly, leading to deserialization issues.

To address this, I propose the following solution:

try {
    // Attempt to parse data as List<ReadOnlyMemory<float>> and return the parsed data
}
catch (KernelException ex1) {
    try {
        // If the first attempt fails, attempt to parse data as List<List<List<ReadOnlyMemory<float>>>>` and return the parsed data
    }
    catch (KernelException ex2) {
        // If both attempts fail, handle the exception (e.g., the model doesn't exist ,the model has still been loading, or an HTTP exception occurred) and rethrow the error
    }
}

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the SK Contribution Guidelines and the pre-submission formatting script raises no violations
• All unit tests pass, and I have added new tests where possible
• I didn't break anyone 😄

[N-E-W-T-O-N]

Hello @dmytrostruk
Can you please suggest to me where I can add the above custom HttpClient Example in main code .

[dmytrostruk]

Hello @dmytrostruk Can you please suggest to me where I can add the above custom HttpClient Example in main code .

@N-E-W-T-O-N That's awesome, thanks for validating this idea and creating an example for this! I think you can put it here:
https://github.com/microsoft/semantic-kernel/blob/main/dotnet/samples/Concepts/Memory/HuggingFace_EmbeddingGeneration.cs

[N-E-W-T-O-N]

I want to add two more parameters in the api's payload body namely 'use_cache' & 'wait_for_model'

I want to send this values as addition parameters in the HuggingFaceClient

https://huggingface.co/docs/api-inference/en/detailed_parameters

[N-E-W-T-O-N]

@dmytrostruk
Added my example in 4f9eb97 .
Please merge my PR..
Let me know for anymore Improvement.

[dmytrostruk]

@N-E-W-T-O-N Thanks for your contribution! There are some warnings about possible null references and formatting issues. In order to proceed with merging, we need to make sure that all warnings are resolved. Thanks again!

[N-E-W-T-O-N]

@RogerBarreto It might be possible that changes in TextEmbeddingResponse class will lead to test case failure

[N-E-W-T-O-N]

@dmytrostruk

Remove the whitespace error.
Please re run the workflow

[markwallace-microsoft]

@N-E-W-T-O-N there are still formatting errors to be addressed. Suggest you run dotnet format on the solution.

[N-E-W-T-O-N]

@dmytrostruk please check it now.
Resolve check-format issues
.Sorry for inconvenience

[dmytrostruk]

@N-E-W-T-O-N PR looks good to me, but there are several unit tests that started to fail because of new change in deserialization logic.

[N-E-W-T-O-N]

@dmytrostruk may I ask how to test unit test cases locally

[dmytrostruk]

may I ask how to test unit test cases locally

@N-E-W-T-O-N

cd dotnet/src/Connectors/Connectors.HuggingFace.UnitTests
dotnet test

[N-E-W-T-O-N]

@dmytrostruk The issue is unit tests are still based on the previous version of embedding dimensions as noted by@RogerBarreto .Would you like me to update the unit test accordingly?

[dmytrostruk]

The issue is unit tests are still based on the previous version of embedding dimensions as noted by@RogerBarreto .Would you like me to update the unit test accordingly?

@N-E-W-T-O-N Yes, if we are changing the behavior, then unit tests should be updated as well. Thanks!

[N-E-W-T-O-N]

@dmytrostruk moved the class inside with access modifier 'private sealed'

Resolve the unit test issue

[N-E-W-T-O-N]

@dmytrostruk anything to add from my side

[dmytrostruk]

@dmytrostruk anything to add from my side

@N-E-W-T-O-N I think we are ready to merge this to main. I added a couple of small fixes to the example. Thanks for your contribution!

[westey-m]

SaveInformationAsync already creates a collection if needed using the passed in collection name, in this case "Test". Why also check if a "Sqlite" collection exists on lines 40-43?

[N-E-W-T-O-N]

Yes Semantic test memory will create the collection if no collection by that name exist

if (!(await this._storage.DoesCollectionExistAsync(collection, cancellationToken).ConfigureAwait(false)))
        {
            await this._storage.CreateCollectionAsync(collection, cancellationToken).ConfigureAwait(false);
        }

I added this code just for testing purposes.

[N-E-W-T-O-N]

Should I remove/comment out this part or leave this part as it is

[westey-m]

Let's remove lines 40-43, since it's confusing to have it in the sample. The sample is about demonstrating how to handle the non-standard response from the "cointegrated/LaBSE-en-ru" model, and we don't need to create the collection with name Sqlite to demonstrate that.