Add genai system-specific conventions for Azure AI Inference

[lmolkova]

Adds

• az.ai.inference system name.
• md file describing how generic conventions apply to Azure AI Inference

It's being implemented in Azure/azure-sdk-for-js#30800 and Azure/azure-sdk-for-python#36890, other languages to follow

Merge requirement checklist

• CONTRIBUTING.md guidelines followed.
• Change log entry added, according to the guidelines in When to add a changelog entry.
  • If your PR does not need a change log, start the PR title with [chore]
• schema-next.yaml updated with changes to existing conventions.

[lmolkova]

Azure AI Inference allows to use specific model-as-a-service and does not always have a model name at request time.

Client interacts with the endpoint, but does not need to know the model.

So, changing level to conditionally_required to accommodate this scenario. Other system-specific conventions may override it and say it's required

[drewby]

Interesting. It used to be that the request would use the "deployment name" which then would have an underlying model configured in the service. Is that still the case?

[lmolkova]

if you use serverless models, after you deploy one the only thing the client knows is the endpoint 🤷‍♀️

const client = ModelClient("https://contoso.eastus2.inference.ai.azure.com", credential);
  const response = await client.path("/chat/completions").post({
    body: {
      messages: [
        ...
      ],
    }
  });

[lmolkova]

@truptiparkar7 could you please bless this name? Thanks!

ps: alternatives/suggestions are more than welcome

[drewby]

Interesting. It used to be that the request would use the "deployment name" which then would have an underlying model configured in the service. Is that still the case?

[drewby]

IMO it's a bit difficult to parse what is specific to the vendor conventions and what is repeated from the general conventions. I have the same concern with the openai conventions. Is there a way to highlight, or sort, or otherwise call out what is the additional attributes?

[lmolkova]

no way to highlight now, but it's a good feature to have.

My assumption that users who use some specific vendor would come to the corresponding page and learn what this vendor reports (doesn't matter if it's generic or not).

I'm really curious to learn why you need to find what's different.

An easy way to to go read yaml which includes all customizations:

semantic-conventions/model/trace/gen-ai.yaml

Lines 97 to 116 in c0f29ff

	- id: trace.gen_ai.az.ai.inference.client
	extends: trace.gen_ai.client.common_attributes
	type: attribute_group
	brief: >
	Describes Azure AI Inference span attributes.
	attributes:
	- ref: az.namespace
	note: >
	When `az.namespace` attribute is populated, it MUST be set to `Microsoft.CognitiveServices` for all
	operations performed by Azure AI Inference clients.
	examples: ["Microsoft.CognitiveServices"]
	- ref: gen_ai.usage.input_tokens
	brief: >
	The number of prompt tokens as reported in the usage prompt_tokens property of the response.
	- ref: gen_ai.usage.output_tokens
	brief: >
	The number of completion tokens as reported in the usage completion_tokens property of the response.
	- ref: server.port
	requirement_level:
	conditionally_required: If not default (443).