[C#] Expose cached Generator through IChatClient.GetService

[ZedingZhang (ZedingZhang)]

Description

Expose the currently cached Generator through IChatClient.GetService<Generator>(). This enables callers to retrieve it between requests for operations such as RewindTo.

The returned generator remains owned by the chat client and is documented as a borrowed reference. Callers must not dispose it, retain it across requests, or use it concurrently with an active request.

Changes

• Return the cached Generator when requested without a service key.
• Use Volatile.Read to access the cache consistently with its existing interlocked updates.
• Document the lifetime and concurrency constraints of the returned generator.
• Add C# coverage using the repository-bundled tiny GPT-2 model.

Testing

• dotnet build src/csharp/Microsoft.ML.OnnxRuntimeGenAI.csproj --configuration Release
  • net8.0 and netstandard2.0 passed with 0 errors.
• dotnet test test/csharp/Microsoft.ML.OnnxRuntimeGenAI.Tests.csproj --configuration Release --no-build
  • 9 passed, 10 skipped because optional external models were unavailable, 0 failed.
• TestChatClientGetServiceGenerator
  • 1 passed, 0 failed.
• macOS arm64 native Release build completed successfully.

Fixes #1553

[azure-pipelines]

Azure Pipelines:
There may be pipelines that require an authorized user to comment /azp run to run.

[Copilot]

Pull request overview

This PR updates the C# OnnxRuntimeGenAIChatClient to expose the currently cached Generator instance via IChatClient.GetService<Generator>(serviceKey: null), enabling advanced callers to perform operations like RewindTo between requests while keeping ownership and lifecycle with the chat client.

Changes:

• Expose the cached Generator from IChatClient.GetService when requested with a null service key.
• Use Volatile.Read for consistent, thread-safe cache access alongside existing Interlocked updates.
• Add a C# test validating the Generator service is unavailable before a request and available after a cached request completes.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
src/csharp/OnnxRuntimeGenAIChatClient.cs	Exposes the cached generator via GetService and documents borrowing/lifetime constraints.
test/csharp/TestOnnxRuntimeGenAIAPI.cs	Adds a unit test covering GetService<Generator>(null) behavior with caching enabled.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[ZedingZhang (ZedingZhang)]

@microsoft-github-policy-service agree