.NET: Fix HuggingFace embedding generation 404 error by updating to router.huggingface.co endpoint

[Copilot]

Description

Fixes the 404 error when using HuggingFace embedding generation by updating from the obsolete api-inference.huggingface.co endpoint to the current router.huggingface.co/hf-inference endpoint.

Problem

The HuggingFace connector was failing with 404 errors when generating embeddings:

var kernelBuilder = Kernel.CreateBuilder();
kernelBuilder.AddHuggingFaceEmbeddingGenerator(
    "sentence-transformers/all-MiniLM-L6-v2",
    apiKey: "api-key");
var k = kernelBuilder.Build();
var g = k.GetRequiredService<IEmbeddingGenerator<string, Embedding<float>>>();
var r = await g.GenerateAsync(["John: Hello, how are you?\nRoger: Hey, I'm Roger!"]);

// Microsoft.SemanticKernel.HttpOperationException: Response status code does not indicate success: 404 (Not Found).

Root Cause

The code was using the deprecated HuggingFace Inference API endpoint structure:

• Old base URL: https://api-inference.huggingface.co
• Old embedding path: /pipeline/feature-extraction/{modelId}
• Resulting URL: https://api-inference.huggingface.co/pipeline/feature-extraction/sentence-transformers/all-MiniLM-L6-v2

Solution

Updated to the current HuggingFace Inference Router API:

• New base URL: https://router.huggingface.co/hf-inference
• New embedding path: /models/{modelId}/pipeline/feature-extraction
• Resulting URL: https://router.huggingface.co/hf-inference/models/sentence-transformers/all-MiniLM-L6-v2/pipeline/feature-extraction

Changes

Core Changes

• HuggingFaceClient.cs: Updated default endpoint and embedding path construction

Test Updates

• Updated test assertions across 5 test files to match new URL structure
• All 82 unit tests passing

Sample Updates

• Updated TelemetryWithAppInsights demo to use new endpoint

Impact

This fix affects all HuggingFace connectors (embeddings, chat completion, text generation, image-to-text) as they share the same base URL. The new endpoints are:

• Chat completion: https://router.huggingface.co/hf-inference/v1/chat/completions
• Text generation: https://router.huggingface.co/hf-inference/models/{modelId}
• Embeddings: https://router.huggingface.co/hf-inference/models/{modelId}/pipeline/feature-extraction
• Image-to-text: https://router.huggingface.co/hf-inference/models/{modelId}

Users who explicitly provide custom endpoints are not affected by this change.

Testing

• ✅ All 82 HuggingFace connector unit tests pass
• ✅ URL construction verified to match the working workaround
• ✅ Minimal surgical changes (7 files, 10 lines modified)

Closes #<issue_number>

Original prompt

---

This section details on the original issue you should resolve

<issue_title>.Net: Bug: .NET AddHuggingFaceEmbeddingGenerator return 404</issue_title>
<issue_description>Describe the bug
Embeddings generation using HuggingFace returns 404.
To my understanding it seems to originate from using the obsolete api-inference.huggingface.co instead of router.huggingface.co/hf-inference

To Reproduce
Try using the embedding generation with:

        var kernelBuilder = Kernel.CreateBuilder();
        kernelBuilder.AddHuggingFaceEmbeddingGenerator(
            "sentence-transformers/all-MiniLM-L6-v2",
            apiKey: "api-key");
        var k = kernelBuilder.Build();
        var g = k.GetRequiredService<IEmbeddingGenerator<string, Embedding<float>>>();
        var r = g.GenerateAsync(["John: Hello, how are you?\nRoger: Hey, I'm Roger!"]).GetAwaiter().GetResult();

Microsoft.SemanticKernel.HttpOperationException: Response status code does not indicate success: 404 (Not Found).
 ---> System.Net.Http.HttpRequestException: Response status code does not indicate success: 404 (Not Found).
   at System.Net.Http.HttpResponseMessage.EnsureSuccessStatusCode()
   at Microsoft.SemanticKernel.Http.HttpClientExtensions.SendWithSuccessCheckAsync(HttpClient client, HttpRequestMessage request, HttpCompletionOption completionOption, CancellationToken cancellationToken)
   --- End of inner exception stack trace ---
   at Microsoft.SemanticKernel.Http.HttpClientExtensions.SendWithSuccessCheckAsync(HttpClient client, HttpRequestMessage request, HttpCompletionOption completionOption, CancellationToken cancellationToken)
   at Microsoft.SemanticKernel.Http.HttpClientExtensions.SendWithSuccessCheckAsync(HttpClient client, HttpRequestMessage request, CancellationToken cancellationToken)
   at Microsoft.SemanticKernel.Connectors.HuggingFace.Core.HuggingFaceClient.SendRequestAndGetStringBodyAsync(HttpRequestMessage httpRequestMessage, CancellationToken cancellationToken)
   at Microsoft.SemanticKernel.Connectors.HuggingFace.Core.HuggingFaceClient.GenerateEmbeddingsAsync(IList`1 data, Kernel kernel, CancellationToken cancellationToken)
   at Microsoft.SemanticKernel.Connectors.HuggingFace.HuggingFaceEmbeddingGenerator.GenerateAsync(IEnumerable`1 values, EmbeddingGenerationOptions options, CancellationToken cancellationToken)

Expected behavior
Should return the embeddings

Additional context
This is the workaround I use for now

 
 var endpoint = $"https://router.huggingface.co/hf-inference/models/{model}/pipeline/feature-extraction";

  kernelBuilder.Services.AddKeyedSingleton<IEmbeddingGenerator<string, Embedding<float>>>(serviceKey: null, (serviceProvider, _) =>
      new HuggingFaceEmbeddingGenerator(
          new Uri(endpoint),
          apiKey: "api-key"));
```</issue_description>

<agent_instructions>Update the references to "https://api-inference.huggingface.co" to use the correct URL</agent_instructions>

## Comments on the Issue (you are @copilot in this section)

<comments>
</comments>

Fixes #13148

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.