How to correctly obtain tokens usage from ChatMessageContent response

[AdriAJ1]

Hi,

I am programming a chatbot in .NET and trying to obtain the token usage for each AI execution.
I am using the OpenAIChatCompletion service.

I have seen that I should use the Metadata property and the Usage field.
The Metadata property is an 'IReadOnlyDictionary<string, object?>?.'.

When deserializing the 'Usage' field, I have often seen the object being cast to 'CompletionUsage'. However, this type is only available in the Semantic Kernel Python version and not in the newer .NET versions.
After doing some checks, I discovered that the 'Usage' object needs to be cast to 'OpenAI.Chat.ChatTokenUsage' (as shown in the image below).

My question is whether there is something like 'CompletionUsage' in .NET that I am missing or if I should simply use the native type from the OpenAI package. It works fine, but I find it strange that this type seems to be available only in the Python version.
I have seen many examples where the AI return object is a 'FunctionResult' because 'InvokeAsync' is used, but very few where the AI return object is a 'ChatMessageContent.' I am not sure if this is the "correct" way to obtain the token count.

Execution prints:
OpenAI.Chat.ChatTokenUsage
Token usage. Input tokens: 848; Output tokens: 22

CompletionUsage class in Python:
https://learn.microsoft.com/en-us/python/api/semantic-kernel/semantic_kernel.connectors.ai.completion_usage.completionusage?view=semantic-kernel-python

OS: Windows
IDE: Visual Studio 2022
Language: C#
Source: NuGet package version 1.32.0

[sophialagerkranspandey]

Tagging @markwallace-microsoft

[rogerbarreto]

@AdriAJ1 you are right, we don't have abstraction for Usage data in the C# library as the usage at the time and still a very specific OpenAI feature.

When using the OpenAI connector against OpenAI compatible API's you may have a better experience using the InnerContent instead of the Metadata:

See here:

semantic-kernel/dotnet/samples/Concepts/ChatCompletion/OpenAI_ChatCompletion.cs

Line 49 in b8795b7

var replyInnerContent = reply.InnerContent as OpenAI.Chat.ChatCompletion;

Retrieving the token count:

semantic-kernel/dotnet/samples/Concepts/ChatCompletion/OpenAI_ChatCompletion.cs

Lines 183 to 185 in b8795b7

	Console.WriteLine($"Input tokens usage: {innerContent.Usage.InputTokenCount}");
	Console.WriteLine($"Output tokens usage: {innerContent.Usage.OutputTokenCount}");
	Console.WriteLine($"Total tokens usage: {innerContent.Usage.TotalTokenCount}");

[sairam6381]

Hello, I am currently using the group chat auto-generation feature, which includes semantic agents. Could you kindly let me know if there is a way to monitor token usage?

[julienFlexsoft]

Hi, is there any updates on this since january?
Thank you