open-telemetry
/
semantic-conventions
mirror of https://github.com/open-telemetry/semantic-conventions.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
semantic-conventions/docs/gen-ai/gen-ai-metrics.md

47 KiB
Raw Blame History

Semantic conventions for generative AI metrics

Status: Development

[!Warning]

Existing GenAI instrumentations that are using v1.36.0 of this document (or prior):

  • SHOULD NOT change the version of the GenAI conventions that they emit by default. Conventions include, but are not limited to, attributes, metric, span and event names, span kind and unit of measure.
  • SHOULD introduce an environment variable OTEL_SEMCONV_STABILITY_OPT_IN as a comma-separated list of category-specific values. The list of values includes:
    • gen_ai_latest_experimental - emit the latest experimental version of GenAI conventions (supported by the instrumentation) and do not emit the old one (v1.36.0 or prior).
    • The default behavior is to continue emitting whatever version of the GenAI conventions the instrumentation was emitting (1.36.0 or prior).

This transition plan will be updated to include stable version before the GenAI conventions are marked as stable.

Generative AI client metrics

The conventions described in this section are specific to Generative AI client applications.

Disclaimer: These are initial Generative AI client metric instruments and attributes but more may be added in the future.

The following metric instruments describe Generative AI operations. An operation may be a request to an LLM, a function call, or some other distinct action within a larger Generative AI workflow.

Individual systems may include additional system-specific attributes. It is recommended to check system-specific documentation, if available.

Metric: gen_ai.client.token.usage

This metric is recommended when an operation involves the usage of tokens and the count is readily available.

For example, if GenAI system returns usage information in the streaming response, it SHOULD be used. Or if GenAI system returns each token independently, instrumentation SHOULD count number of output tokens and record the result.

If instrumentation cannot efficiently obtain number of input and/or output tokens, it MAY allow users to enable offline token counting. Otherwise it MUST NOT report usage metric.

When systems report both used tokens and billable tokens, instrumentation MUST report billable tokens.

This metric SHOULD be specified with ExplicitBucketBoundaries of [1, 4, 16, 64, 256, 1024, 4096, 16384, 65536, 262144, 1048576, 4194304, 16777216, 67108864].

Name Instrument Type Unit (UCUM) Description Stability Entity Associations
gen_ai.client.token.usage Histogram {token} Number of input and output tokens used. Development
Attribute Type Description Examples Requirement Level Stability
gen_ai.operation.name string The name of the operation being performed. [1] chat; generate_content; text_completion Required Development
gen_ai.provider.name string The Generative AI provider as identified by the client or server instrumentation. [2] openai; gcp.gen_ai; gcp.vertex_ai Required Development
gen_ai.token.type string The type of token being counted. input; output Required Development
gen_ai.request.model string The name of the GenAI model a request is being made to. gpt-4 Conditionally Required If available. Development
server.port int GenAI server port. [3] 80; 8080; 443 Conditionally Required If server.address is set. Stable
gen_ai.response.model string The name of the model that generated the response. gpt-4-0613 Recommended Development
server.address string GenAI server address. [4] example.com; 10.1.2.80; /tmp/my.sock Recommended Stable

[1] gen_ai.operation.name: If one of the predefined values applies, but specific system uses a different name it's RECOMMENDED to document it in the semantic conventions for specific GenAI system and use system-specific name in the instrumentation. If a different name is not documented, instrumentation libraries SHOULD use applicable predefined value.

[2] gen_ai.provider.name: The attribute SHOULD be set based on the instrumentation's best knowledge and may differ from the actual model provider.

Multiple providers, including Azure OpenAI, Gemini, and AI hosting platforms are accessible using the OpenAI REST API and corresponding client libraries, but may proxy or host models from different providers.

The gen_ai.request.model, gen_ai.response.model, and server.address attributes may help identify the actual system in use.

The gen_ai.provider.name attribute acts as a discriminator that identifies the GenAI telemetry format flavor specific to that provider within GenAI semantic conventions. It SHOULD be set consistently with provider-specific attributes and signals. For example, GenAI spans, metrics, and events related to AWS Bedrock should have the gen_ai.provider.name set to aws.bedrock and include applicable aws.bedrock.* attributes and are not expected to include openai.* attributes.

[3] server.port: When observed from the client side, and when communicating through an intermediary, server.port SHOULD represent the server port behind any intermediaries, for example proxies, if it's available.

[4] server.address: When observed from the client side, and when communicating through an intermediary, server.address SHOULD represent the server address behind any intermediaries, for example proxies, if it's available.

gen_ai.operation.name has the following list of well-known values. If one of them applies, then the respective value MUST be used; otherwise, a custom value MAY be used.

Value Description Stability
chat Chat completion operation such as OpenAI Chat API Development
create_agent Create GenAI agent Development
embeddings Embeddings operation such as OpenAI Create embeddings API Development
execute_tool Execute a tool Development
generate_content Multimodal content generation operation such as Gemini Generate Content Development
invoke_agent Invoke GenAI agent Development
text_completion Text completions operation such as OpenAI Completions API (Legacy) Development

gen_ai.provider.name has the following list of well-known values. If one of them applies, then the respective value MUST be used; otherwise, a custom value MAY be used.

Value Description Stability
anthropic Anthropic Development
aws.bedrock AWS Bedrock Development
azure.ai.inference Azure AI Inference Development
azure.ai.openai Azure OpenAI Development
cohere Cohere Development
deepseek DeepSeek Development
gcp.gemini Gemini [5] Development
gcp.gen_ai Any Google generative AI endpoint [6] Development
gcp.vertex_ai Vertex AI [7] Development
groq Groq Development
ibm.watsonx.ai IBM Watsonx AI Development
mistral_ai Mistral AI Development
openai OpenAI Development
perplexity Perplexity Development
x_ai xAI Development

[5]: Used when accessing the 'generativelanguage.googleapis.com' endpoint. Also known as the AI Studio API.

[6]: May be used when specific backend is unknown.

[7]: Used when accessing the 'aiplatform.googleapis.com' endpoint.

gen_ai.token.type has the following list of well-known values. If one of them applies, then the respective value MUST be used; otherwise, a custom value MAY be used.

Value Description Stability
input Input tokens (prompt, input, etc.) Development
output Output tokens (completion, response, etc.) Development

Metric: gen_ai.client.operation.duration

This metric is required.

This metric SHOULD be specified with ExplicitBucketBoundaries of [0.01, 0.02, 0.04, 0.08, 0.16, 0.32, 0.64, 1.28, 2.56, 5.12, 10.24, 20.48, 40.96, 81.92].

Name Instrument Type Unit (UCUM) Description Stability Entity Associations
gen_ai.client.operation.duration Histogram s GenAI operation duration. Development
Attribute Type Description Examples Requirement Level Stability
gen_ai.operation.name string The name of the operation being performed. [1] chat; generate_content; text_completion Required Development
gen_ai.provider.name string The Generative AI provider as identified by the client or server instrumentation. [2] openai; gcp.gen_ai; gcp.vertex_ai Required Development
error.type string Describes a class of error the operation ended with. [3] timeout; java.net.UnknownHostException; server_certificate_invalid; 500 Conditionally Required if the operation ended in an error Stable
gen_ai.request.model string The name of the GenAI model a request is being made to. gpt-4 Conditionally Required If available. Development
server.port int GenAI server port. [4] 80; 8080; 443 Conditionally Required If server.address is set. Stable
gen_ai.response.model string The name of the model that generated the response. gpt-4-0613 Recommended Development
server.address string GenAI server address. [5] example.com; 10.1.2.80; /tmp/my.sock Recommended Stable

[1] gen_ai.operation.name: If one of the predefined values applies, but specific system uses a different name it's RECOMMENDED to document it in the semantic conventions for specific GenAI system and use system-specific name in the instrumentation. If a different name is not documented, instrumentation libraries SHOULD use applicable predefined value.

[2] gen_ai.provider.name: The attribute SHOULD be set based on the instrumentation's best knowledge and may differ from the actual model provider.

Multiple providers, including Azure OpenAI, Gemini, and AI hosting platforms are accessible using the OpenAI REST API and corresponding client libraries, but may proxy or host models from different providers.

The gen_ai.request.model, gen_ai.response.model, and server.address attributes may help identify the actual system in use.

The gen_ai.provider.name attribute acts as a discriminator that identifies the GenAI telemetry format flavor specific to that provider within GenAI semantic conventions. It SHOULD be set consistently with provider-specific attributes and signals. For example, GenAI spans, metrics, and events related to AWS Bedrock should have the gen_ai.provider.name set to aws.bedrock and include applicable aws.bedrock.* attributes and are not expected to include openai.* attributes.

[3] error.type: The error.type SHOULD match the error code returned by the Generative AI provider or the client library, the canonical name of exception that occurred, or another low-cardinality error identifier. Instrumentations SHOULD document the list of errors they report.

[4] server.port: When observed from the client side, and when communicating through an intermediary, server.port SHOULD represent the server port behind any intermediaries, for example proxies, if it's available.

[5] server.address: When observed from the client side, and when communicating through an intermediary, server.address SHOULD represent the server address behind any intermediaries, for example proxies, if it's available.

error.type has the following list of well-known values. If one of them applies, then the respective value MUST be used; otherwise, a custom value MAY be used.

Value Description Stability
_OTHER A fallback error value to be used when the instrumentation doesn't define a custom value. Stable

gen_ai.operation.name has the following list of well-known values. If one of them applies, then the respective value MUST be used; otherwise, a custom value MAY be used.

Value Description Stability
chat Chat completion operation such as OpenAI Chat API Development
create_agent Create GenAI agent Development
embeddings Embeddings operation such as OpenAI Create embeddings API Development
execute_tool Execute a tool Development
generate_content Multimodal content generation operation such as Gemini Generate Content Development
invoke_agent Invoke GenAI agent Development
text_completion Text completions operation such as OpenAI Completions API (Legacy) Development

gen_ai.provider.name has the following list of well-known values. If one of them applies, then the respective value MUST be used; otherwise, a custom value MAY be used.

Value Description Stability
anthropic Anthropic Development
aws.bedrock AWS Bedrock Development
azure.ai.inference Azure AI Inference Development
azure.ai.openai Azure OpenAI Development
cohere Cohere Development
deepseek DeepSeek Development
gcp.gemini Gemini [6] Development
gcp.gen_ai Any Google generative AI endpoint [7] Development
gcp.vertex_ai Vertex AI [8] Development
groq Groq Development
ibm.watsonx.ai IBM Watsonx AI Development
mistral_ai Mistral AI Development
openai OpenAI Development
perplexity Perplexity Development
x_ai xAI Development

[6]: Used when accessing the 'generativelanguage.googleapis.com' endpoint. Also known as the AI Studio API.

[7]: May be used when specific backend is unknown.

[8]: Used when accessing the 'aiplatform.googleapis.com' endpoint.

Generative AI model server metrics

The following metric instruments describe Generative AI model servers' operational metrics. It includes both functional and performance metrics.

Metric: gen_ai.server.request.duration

This metric is recommended to report the model server latency in terms of time spent per request.

This metric SHOULD be specified with ExplicitBucketBoundaries of [0.01, 0.02, 0.04, 0.08, 0.16, 0.32, 0.64, 1.28, 2.56, 5.12, 10.24, 20.48, 40.96, 81.92].

Name Instrument Type Unit (UCUM) Description Stability Entity Associations
gen_ai.server.request.duration Histogram s Generative AI server request duration such as time-to-last byte or last output token. Development
Attribute Type Description Examples Requirement Level Stability
gen_ai.operation.name string The name of the operation being performed. [1] chat; generate_content; text_completion Required Development
gen_ai.provider.name string The Generative AI provider as identified by the client or server instrumentation. [2] openai; gcp.gen_ai; gcp.vertex_ai Required Development
error.type string Describes a class of error the operation ended with. [3] timeout; java.net.UnknownHostException; server_certificate_invalid; 500 Conditionally Required if the operation ended in an error Stable
gen_ai.request.model string The name of the GenAI model a request is being made to. gpt-4 Conditionally Required If available. Development
server.port int GenAI server port. [4] 80; 8080; 443 Conditionally Required If server.address is set. Stable
gen_ai.response.model string The name of the model that generated the response. gpt-4-0613 Recommended Development
server.address string GenAI server address. [5] example.com; 10.1.2.80; /tmp/my.sock Recommended Stable

[1] gen_ai.operation.name: If one of the predefined values applies, but specific system uses a different name it's RECOMMENDED to document it in the semantic conventions for specific GenAI system and use system-specific name in the instrumentation. If a different name is not documented, instrumentation libraries SHOULD use applicable predefined value.

[2] gen_ai.provider.name: The attribute SHOULD be set based on the instrumentation's best knowledge and may differ from the actual model provider.

Multiple providers, including Azure OpenAI, Gemini, and AI hosting platforms are accessible using the OpenAI REST API and corresponding client libraries, but may proxy or host models from different providers.

The gen_ai.request.model, gen_ai.response.model, and server.address attributes may help identify the actual system in use.

The gen_ai.provider.name attribute acts as a discriminator that identifies the GenAI telemetry format flavor specific to that provider within GenAI semantic conventions. It SHOULD be set consistently with provider-specific attributes and signals. For example, GenAI spans, metrics, and events related to AWS Bedrock should have the gen_ai.provider.name set to aws.bedrock and include applicable aws.bedrock.* attributes and are not expected to include openai.* attributes.

[3] error.type: The error.type SHOULD match the error code returned by the Generative AI service, the canonical name of exception that occurred, or another low-cardinality error identifier. Instrumentations SHOULD document the list of errors they report.

[4] server.port: When observed from the client side, and when communicating through an intermediary, server.port SHOULD represent the server port behind any intermediaries, for example proxies, if it's available.

[5] server.address: When observed from the client side, and when communicating through an intermediary, server.address SHOULD represent the server address behind any intermediaries, for example proxies, if it's available.

error.type has the following list of well-known values. If one of them applies, then the respective value MUST be used; otherwise, a custom value MAY be used.

Value Description Stability
_OTHER A fallback error value to be used when the instrumentation doesn't define a custom value. Stable

gen_ai.operation.name has the following list of well-known values. If one of them applies, then the respective value MUST be used; otherwise, a custom value MAY be used.

Value Description Stability
chat Chat completion operation such as OpenAI Chat API Development
create_agent Create GenAI agent Development
embeddings Embeddings operation such as OpenAI Create embeddings API Development
execute_tool Execute a tool Development
generate_content Multimodal content generation operation such as Gemini Generate Content Development
invoke_agent Invoke GenAI agent Development
text_completion Text completions operation such as OpenAI Completions API (Legacy) Development

gen_ai.provider.name has the following list of well-known values. If one of them applies, then the respective value MUST be used; otherwise, a custom value MAY be used.

Value Description Stability
anthropic Anthropic Development
aws.bedrock AWS Bedrock Development
azure.ai.inference Azure AI Inference Development
azure.ai.openai Azure OpenAI Development
cohere Cohere Development
deepseek DeepSeek Development
gcp.gemini Gemini [6] Development
gcp.gen_ai Any Google generative AI endpoint [7] Development
gcp.vertex_ai Vertex AI [8] Development
groq Groq Development
ibm.watsonx.ai IBM Watsonx AI Development
mistral_ai Mistral AI Development
openai OpenAI Development
perplexity Perplexity Development
x_ai xAI Development

[6]: Used when accessing the 'generativelanguage.googleapis.com' endpoint. Also known as the AI Studio API.

[7]: May be used when specific backend is unknown.

[8]: Used when accessing the 'aiplatform.googleapis.com' endpoint.

Metric: gen_ai.server.time_per_output_token

This metric is recommended to report the model server latency in terms of time per token generated after the first token for any model servers which support serving LLMs. It is measured by subtracting the time taken to generate the first output token from the request duration and dividing the rest of the duration by the number of output tokens generated after the first token. This is important in measuring the performance of the decode phase of LLM inference.

This metric SHOULD be specified with ExplicitBucketBoundaries of [0.01, 0.025, 0.05, 0.075, 0.1, 0.15, 0.2, 0.3, 0.4, 0.5, 0.75, 1.0, 2.5].

Name Instrument Type Unit (UCUM) Description Stability Entity Associations
gen_ai.server.time_per_output_token Histogram s Time per output token generated after the first token for successful responses. Development
Attribute Type Description Examples Requirement Level Stability
gen_ai.operation.name string The name of the operation being performed. [1] chat; generate_content; text_completion Required Development
gen_ai.provider.name string The Generative AI provider as identified by the client or server instrumentation. [2] openai; gcp.gen_ai; gcp.vertex_ai Required Development
gen_ai.request.model string The name of the GenAI model a request is being made to. gpt-4 Conditionally Required If available. Development
server.port int GenAI server port. [3] 80; 8080; 443 Conditionally Required If server.address is set. Stable
gen_ai.response.model string The name of the model that generated the response. gpt-4-0613 Recommended Development
server.address string GenAI server address. [4] example.com; 10.1.2.80; /tmp/my.sock Recommended Stable

[1] gen_ai.operation.name: If one of the predefined values applies, but specific system uses a different name it's RECOMMENDED to document it in the semantic conventions for specific GenAI system and use system-specific name in the instrumentation. If a different name is not documented, instrumentation libraries SHOULD use applicable predefined value.

[2] gen_ai.provider.name: The attribute SHOULD be set based on the instrumentation's best knowledge and may differ from the actual model provider.

Multiple providers, including Azure OpenAI, Gemini, and AI hosting platforms are accessible using the OpenAI REST API and corresponding client libraries, but may proxy or host models from different providers.

The gen_ai.request.model, gen_ai.response.model, and server.address attributes may help identify the actual system in use.

The gen_ai.provider.name attribute acts as a discriminator that identifies the GenAI telemetry format flavor specific to that provider within GenAI semantic conventions. It SHOULD be set consistently with provider-specific attributes and signals. For example, GenAI spans, metrics, and events related to AWS Bedrock should have the gen_ai.provider.name set to aws.bedrock and include applicable aws.bedrock.* attributes and are not expected to include openai.* attributes.

[3] server.port: When observed from the client side, and when communicating through an intermediary, server.port SHOULD represent the server port behind any intermediaries, for example proxies, if it's available.

[4] server.address: When observed from the client side, and when communicating through an intermediary, server.address SHOULD represent the server address behind any intermediaries, for example proxies, if it's available.

gen_ai.operation.name has the following list of well-known values. If one of them applies, then the respective value MUST be used; otherwise, a custom value MAY be used.

Value Description Stability
chat Chat completion operation such as OpenAI Chat API Development
create_agent Create GenAI agent Development
embeddings Embeddings operation such as OpenAI Create embeddings API Development
execute_tool Execute a tool Development
generate_content Multimodal content generation operation such as Gemini Generate Content Development
invoke_agent Invoke GenAI agent Development
text_completion Text completions operation such as OpenAI Completions API (Legacy) Development

gen_ai.provider.name has the following list of well-known values. If one of them applies, then the respective value MUST be used; otherwise, a custom value MAY be used.

Value Description Stability
anthropic Anthropic Development
aws.bedrock AWS Bedrock Development
azure.ai.inference Azure AI Inference Development
azure.ai.openai Azure OpenAI Development
cohere Cohere Development
deepseek DeepSeek Development
gcp.gemini Gemini [5] Development
gcp.gen_ai Any Google generative AI endpoint [6] Development
gcp.vertex_ai Vertex AI [7] Development
groq Groq Development
ibm.watsonx.ai IBM Watsonx AI Development
mistral_ai Mistral AI Development
openai OpenAI Development
perplexity Perplexity Development
x_ai xAI Development

[5]: Used when accessing the 'generativelanguage.googleapis.com' endpoint. Also known as the AI Studio API.

[6]: May be used when specific backend is unknown.

[7]: Used when accessing the 'aiplatform.googleapis.com' endpoint.

Metric: gen_ai.server.time_to_first_token

This metric is recommended to report the model server latency in terms of time spent to generate the first token of the response for any model servers which support serving LLMs. It helps measure the time spent in the queue and the prefill phase. It is important especially for streaming requests. It is calculated at a request level and is reported as a histogram using the buckets mentioned below.

This metric SHOULD be specified with ExplicitBucketBoundaries of [0.001, 0.005, 0.01, 0.02, 0.04, 0.06, 0.08, 0.1, 0.25, 0.5, 0.75, 1.0, 2.5, 5.0, 7.5, 10.0].

Name Instrument Type Unit (UCUM) Description Stability Entity Associations
gen_ai.server.time_to_first_token Histogram s Time to generate first token for successful responses. Development
Attribute Type Description Examples Requirement Level Stability
gen_ai.operation.name string The name of the operation being performed. [1] chat; generate_content; text_completion Required Development
gen_ai.provider.name string The Generative AI provider as identified by the client or server instrumentation. [2] openai; gcp.gen_ai; gcp.vertex_ai Required Development
gen_ai.request.model string The name of the GenAI model a request is being made to. gpt-4 Conditionally Required If available. Development
server.port int GenAI server port. [3] 80; 8080; 443 Conditionally Required If server.address is set. Stable
gen_ai.response.model string The name of the model that generated the response. gpt-4-0613 Recommended Development
server.address string GenAI server address. [4] example.com; 10.1.2.80; /tmp/my.sock Recommended Stable

[1] gen_ai.operation.name: If one of the predefined values applies, but specific system uses a different name it's RECOMMENDED to document it in the semantic conventions for specific GenAI system and use system-specific name in the instrumentation. If a different name is not documented, instrumentation libraries SHOULD use applicable predefined value.

[2] gen_ai.provider.name: The attribute SHOULD be set based on the instrumentation's best knowledge and may differ from the actual model provider.

Multiple providers, including Azure OpenAI, Gemini, and AI hosting platforms are accessible using the OpenAI REST API and corresponding client libraries, but may proxy or host models from different providers.

The gen_ai.request.model, gen_ai.response.model, and server.address attributes may help identify the actual system in use.

The gen_ai.provider.name attribute acts as a discriminator that identifies the GenAI telemetry format flavor specific to that provider within GenAI semantic conventions. It SHOULD be set consistently with provider-specific attributes and signals. For example, GenAI spans, metrics, and events related to AWS Bedrock should have the gen_ai.provider.name set to aws.bedrock and include applicable aws.bedrock.* attributes and are not expected to include openai.* attributes.

[3] server.port: When observed from the client side, and when communicating through an intermediary, server.port SHOULD represent the server port behind any intermediaries, for example proxies, if it's available.

[4] server.address: When observed from the client side, and when communicating through an intermediary, server.address SHOULD represent the server address behind any intermediaries, for example proxies, if it's available.

gen_ai.operation.name has the following list of well-known values. If one of them applies, then the respective value MUST be used; otherwise, a custom value MAY be used.

Value Description Stability
chat Chat completion operation such as OpenAI Chat API Development
create_agent Create GenAI agent Development
embeddings Embeddings operation such as OpenAI Create embeddings API Development
execute_tool Execute a tool Development
generate_content Multimodal content generation operation such as Gemini Generate Content Development
invoke_agent Invoke GenAI agent Development
text_completion Text completions operation such as OpenAI Completions API (Legacy) Development

gen_ai.provider.name has the following list of well-known values. If one of them applies, then the respective value MUST be used; otherwise, a custom value MAY be used.

Value Description Stability
anthropic Anthropic Development
aws.bedrock AWS Bedrock Development
azure.ai.inference Azure AI Inference Development
azure.ai.openai Azure OpenAI Development
cohere Cohere Development
deepseek DeepSeek Development
gcp.gemini Gemini [5] Development
gcp.gen_ai Any Google generative AI endpoint [6] Development
gcp.vertex_ai Vertex AI [7] Development
groq Groq Development
ibm.watsonx.ai IBM Watsonx AI Development
mistral_ai Mistral AI Development
openai OpenAI Development
perplexity Perplexity Development
x_ai xAI Development

[5]: Used when accessing the 'generativelanguage.googleapis.com' endpoint. Also known as the AI Studio API.

[6]: May be used when specific backend is unknown.

[7]: Used when accessing the 'aiplatform.googleapis.com' endpoint.