Define client spans for Generative AI agents #3
Conversation
…or the future/customizations
| @@ -0,0 +1,5 @@ | |||
| change_type: 'enhancement' | |||
| component: gen_ai | |||
| note: Yamlify gen_ai events and clean up examples. | |||
| | [`gen_ai.request.presence_penalty`](/docs/attributes-registry/gen-ai.md) | double | The presence penalty setting for the GenAI request. | `0.1` | `Recommended` |  | | ||
| | [`gen_ai.request.stop_sequences`](/docs/attributes-registry/gen-ai.md) | string[] | List of sequences that the model will use to stop generating further tokens. | `["forest", "lived"]` | `Recommended` |  | | ||
| | [`gen_ai.request.temperature`](/docs/attributes-registry/gen-ai.md) | double | The temperature setting for the GenAI request. | `0.0` | `Recommended` |  | | ||
| | [`gen_ai.request.top_p`](/docs/attributes-registry/gen-ai.md) | double | The top_p sampling setting for the GenAI request. | `1.0` | `Recommended` |  | | ||
| | [`gen_ai.response.finish_reasons`](/docs/attributes-registry/gen-ai.md) | string[] | Array of reasons the model stopped generating tokens, corresponding to each generation received. | `["stop"]`; `["stop", "length"]` | `Recommended` |  | | ||
| | [`gen_ai.response.id`](/docs/attributes-registry/gen-ai.md) | string | The unique identifier for the completion. | `chatcmpl-123` | `Recommended` |  | |
There was a problem hiding this comment.
we probably want to keep it for the time being (for correlation with evals) and see how it can me merged with message id in the future
| @@ -1,6 +1,7 @@ | |||
| groups: | |||
There was a problem hiding this comment.
@lmolkova this agent semantic convention looks great, but seems this is mainly based on openai assistent. I did some research work for different agent frameworks at https://gyliu513.github.io/jekyll/update/2024/12/17/battle-of-ai-agent-frameworks.html , will think more for the semantic convention.
I was hoping I can probably borrow some idea from your PR and apply to open-telemetry#1657 , may I know why you did not create this PR in the repo https://github.com/open-telemetry/semantic-conventions but under a forked repo? Thanks!
There was a problem hiding this comment.
Generally we need to discuss and figure out how to draw the line between generic agent semantics vs vendor specific ones for different frameworks
There was a problem hiding this comment.
@karthikscale3 +1, we can probably follow same pattern as LLM, we can have a common agent semantic convention, and each agent framework can extend based on the common agent semantic convention
There was a problem hiding this comment.
rule of 3 is bare minimum on specs IMHO (and I know nobody asked me). If a spec doesn't cover 3 diverse impls, it will break until then (probably still break after though ;))
No description provided.