GenAI Trace Field Definitions

LLM Trace fields are defined by TingYun with reference to the OpenTelemetry standard and concepts in the large language model (LLM) application domain. By extending Attributes, they are used to describe the semantics of LLM application call chain data, reflecting key operations such as LLM input/output requests and token consumption. They provide rich, context-related semantic data for scenarios such as Completion, Chat, RAG, Agent, and Tool, facilitating data tracing and reporting. These semantic fields will be continuously updated and optimized as the community evolves.

Attributes

General

Attribute	Type	Description	Required	Example
gen_ai.session.id	String	Session ID	No	"ddde34343-f93a-4477-33333-sdfsdaf"
gen_ai.user.id	String	Application user ID	No	"e1b330b0-d177-4dce-9407-f18ff7f9d600" or "tingyun"
gen_ai.span.kind	String	Operation type detail	Yes	"LLM"
gen_ai.framework	String	Framework type	No	"dify" or "langchain" ..

Span Kind

In practical LLM interaction scenarios, multiple operation steps are often involved. Based on the LLM application domain and development frameworks, the following operation types are currently defined. These types support extension, and different operation types have different Attributes fields.

• WORKFLOW: A tool that connects LLM and other components to accomplish complex tasks, possibly including Retrieval, Embedding, LLM calls, etc.
• LLM: Identifies calls to large models, such as reasoning or text generation via SDK or OpenAPI.
• AGENT: Intelligent agent scenario, a complex CHAIN that may involve multiple LLM and Tool calls, with step-by-step decision-making to reach the final answer.

WORKFLOW

Attribute	Type	Description	Required	Example
gen_ai.span.kind	String	Operation type	Yes	"WORKFLOW"
gen_ai.framework	String	Application framework	Yes	"dify"
gen_ai.session.id	String	Session ID	No	"ddde34343-f93a-4477-33333-sdfsdaf"
gen_ai.user.id	String	Application user ID	No	"e1b330b0-d177-4dce-9407-f18ff7f9d600" or "tingyun"
gen_ai.workflow.total_steps	String	Number of nodes in WORKFLOW	No	"12"
gen_ai.workflow.node_type	String	Node type in WORKFLOW	No	"llm"
gen_ai.workflow.index	String	Node index in WORKFLOW	No	"6"
gen_ai.status	String	Final execution status	No	"success"
gen_ai.title	String	Custom name	No	"AI Assistant"
gen_ai.error	String	Original error message	No	"Failed to invoke model, Range of input length should be [1, 14000]"}!"
gen_ai.input_text	String	API request content, usually JSON, with length limit	No	"Who Are You!"
gen_ai.process_data	String	API request processed data, with length limit	No	"Hello, Who Are You!"
gen_ai.output_text	String	API response content, usually JSON, with length limit	No	"I am ChatBot"
gen_ai.usage.input_tokens	String	Number of tokens in prompt	No	"100"
gen_ai.usage.output_tokens	String	Number of tokens in completion	No	"200"
gen_ai.usage.total_tokens	String	Total number of tokens, including prompt and completion	No	"300"

LLM & AGENT

Attribute	Type	Description	Required	Example
gen_ai.span.kind	String	Operation type	Yes	"LLM" or "AGENT"
gen_ai.framework	String	Application framework	Yes	"dify"
gen_ai.system	String	LLM provider	Yes	"OPENAI"
gen_ai.stream	String	Whether response is in stream format	No	"True"
gen_ai.session.id	String	Session ID	No	"ddde34343-f93a-4477-33333-sdfsdaf"
gen_ai.operation.name	String	Secondary operation type	No	"chat" or "completion"
gen_ai.request.id	String	Request ID	Yes	"ddde34343-f93a-4477-33333-sdfsdaf"
gen_ai.request.model	String	Target LLM model name	Yes	"gpt-4"
gen_ai.request.stop_sequences	String	Stop sequence strings	No	"stop"
gen_ai.request.frequency_penalty	String	Frequency penalty	No	"1.0"
gen_ai.request.presence_penalty	String	Presence penalty	No	"1.0"
gen_ai.request.repetition_penalty	String	Repetition penalty	No	"1.0"
gen_ai.request.max_tokens	String	Max tokens to generate	No	"8192"
gen_ai.request.seed	String	Random seed	No	"1234"
gen_ai.request.temperature	String	LLM request temperature	No	"0.1"
gen_ai.request.top_k	String	LLM request top_k setting	No	"1"
gen_ai.request.top_p	String	LLM request top_p setting	No	"1"
gen_ai.request.response_format	String	LLM request response format	No	"json"
gen_ai.request.incremental_output	String	LLM request incremental output	No	"True"
gen_ai.response.id	String	Response ID	No	"ddde34343-f93a-4477-33333-sdfsdaf"
gen_ai.response.model	String	LLM model name for response	Yes	"gpt-4"
gen_ai.response.finish_reason	String	Model stop reason	No	"stop"
gen_ai.workflow.node_type	String	Node type in WORKFLOW	No	"llm"
gen_ai.workflow.index	String	Node index in WORKFLOW	No	"6"
gen_ai.status	String	Final status of LLM call	No	"success"
gen_ai.title	String	Custom name for LLM call	No	"AI Assistant"
gen_ai.error	String	Original error message from API	No	"Failed to invoke model"
gen_ai.request.input_text	String	API request content, usually JSON, with length limit	No	"Who Are You!"
gen_ai.response.output_text	String	API response content, usually JSON, with length limit	No	"I am ChatBot"
gen_ai.usage.input_tokens	String	Number of tokens in prompt	No	"100"
gen_ai.usage.output_tokens	String	Number of tokens in completion	No	"200"
gen_ai.usage.total_tokens	String	Total number of tokens, including prompt and completion	No	"300"
gen_ai.response.first_pack_duration	String	Time from LLM request to first response packet (for stream scenarios)	No	"10"