youngkingdom/dify
1
0
Fork 0
mirror of https://github.com/langgenius/dify.git synced 2026-05-05 09:58:04 +08:00
Code Issues Packages Projects Releases Wiki Activity

Expose LLM usage in workflows (#21766)

Browse Source 
Co-authored-by: crazywoola <100913391+crazywoola@users.noreply.github.com>
This commit is contained in:
Davide Delbianco
2025-07-08 15:18:00 +02:00
committed by GitHub
parent f925869f61
commit 816210d744
32 changed files with 201 additions and 81 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
api/core/ops/aliyun_trace/aliyun_trace.py
View File

@ -372,6 +372,7 @@ class AliyunDataTrace(BaseTraceInstance):
) -> SpanData:
process_data = node_execution.process_data or {}
outputs = node_execution.outputs or {}
usage_data = process_data.get("usage", {}) if "usage" in process_data else outputs.get("usage", {})
return SpanData(
trace_id=trace_id,
parent_span_id=workflow_span_id,
@ -385,9 +386,9 @@ class AliyunDataTrace(BaseTraceInstance):
GEN_AI_FRAMEWORK: "dify",
GEN_AI_MODEL_NAME: process_data.get("model_name", ""),
GEN_AI_SYSTEM: process_data.get("model_provider", ""),
GEN_AI_USAGE_INPUT_TOKENS: str(outputs.get("usage", {}).get("prompt_tokens", 0)),
GEN_AI_USAGE_OUTPUT_TOKENS: str(outputs.get("usage", {}).get("completion_tokens", 0)),
GEN_AI_USAGE_TOTAL_TOKENS: str(outputs.get("usage", {}).get("total_tokens", 0)),
GEN_AI_USAGE_INPUT_TOKENS: str(usage_data.get("prompt_tokens", 0)),
GEN_AI_USAGE_OUTPUT_TOKENS: str(usage_data.get("completion_tokens", 0)),
GEN_AI_USAGE_TOTAL_TOKENS: str(usage_data.get("total_tokens", 0)),
GEN_AI_PROMPT: json.dumps(process_data.get("prompts", []), ensure_ascii=False),
GEN_AI_COMPLETION: str(outputs.get("text", "")),
GEN_AI_RESPONSE_FINISH_REASON: outputs.get("finish_reason", ""),

26
api/core/ops/arize_phoenix_trace/arize_phoenix_trace.py
View File

@ -213,11 +213,12 @@ class ArizePhoenixDataTrace(BaseTraceInstance):
if model:
node_metadata["ls_model_name"] = model
usage = json.loads(node_execution.outputs).get("usage", {}) if node_execution.outputs else {}
if usage:
node_metadata["total_tokens"] = usage.get("total_tokens", 0)
node_metadata["prompt_tokens"] = usage.get("prompt_tokens", 0)
node_metadata["completion_tokens"] = usage.get("completion_tokens", 0)
outputs = json.loads(node_execution.outputs).get("usage", {})
usage_data = process_data.get("usage", {}) if "usage" in process_data else outputs.get("usage", {})
if usage_data:
node_metadata["total_tokens"] = usage_data.get("total_tokens", 0)
node_metadata["prompt_tokens"] = usage_data.get("prompt_tokens", 0)
node_metadata["completion_tokens"] = usage_data.get("completion_tokens", 0)
elif node_execution.node_type == "dataset_retrieval":
span_kind = OpenInferenceSpanKindValues.RETRIEVER.value
elif node_execution.node_type == "tool":
@ -246,14 +247,19 @@ class ArizePhoenixDataTrace(BaseTraceInstance):
if model:
node_span.set_attribute(SpanAttributes.LLM_MODEL_NAME, model)
usage = json.loads(node_execution.outputs).get("usage", {}) if node_execution.outputs else {}
if usage:
node_span.set_attribute(SpanAttributes.LLM_TOKEN_COUNT_TOTAL, usage.get("total_tokens", 0))
outputs = json.loads(node_execution.outputs).get("usage", {})
usage_data = (
process_data.get("usage", {}) if "usage" in process_data else outputs.get("usage", {})
)
if usage_data:
node_span.set_attribute(
SpanAttributes.LLM_TOKEN_COUNT_PROMPT, usage.get("prompt_tokens", 0)
SpanAttributes.LLM_TOKEN_COUNT_TOTAL, usage_data.get("total_tokens", 0)
)
node_span.set_attribute(
SpanAttributes.LLM_TOKEN_COUNT_COMPLETION, usage.get("completion_tokens", 0)
SpanAttributes.LLM_TOKEN_COUNT_PROMPT, usage_data.get("prompt_tokens", 0)
)
node_span.set_attribute(
SpanAttributes.LLM_TOKEN_COUNT_COMPLETION, usage_data.get("completion_tokens", 0)
)
finally:
node_span.end(end_time=datetime_to_nanos(finished_at))

9
api/core/ops/langfuse_trace/langfuse_trace.py
View File

@ -181,12 +181,9 @@ class LangFuseDataTrace(BaseTraceInstance):
prompt_tokens = 0
completion_tokens = 0
try:
if outputs.get("usage"):
prompt_tokens = outputs.get("usage", {}).get("prompt_tokens", 0)
completion_tokens = outputs.get("usage", {}).get("completion_tokens", 0)
else:
prompt_tokens = process_data.get("usage", {}).get("prompt_tokens", 0)
completion_tokens = process_data.get("usage", {}).get("completion_tokens", 0)
usage_data = process_data.get("usage", {}) if "usage" in process_data else outputs.get("usage", {})
prompt_tokens = usage_data.get("prompt_tokens", 0)
completion_tokens = usage_data.get("completion_tokens", 0)
except Exception:
logger.error("Failed to extract usage", exc_info=True)

9
api/core/ops/langsmith_trace/langsmith_trace.py
View File

@ -206,12 +206,9 @@ class LangSmithDataTrace(BaseTraceInstance):
prompt_tokens = 0
completion_tokens = 0
try:
if outputs.get("usage"):
prompt_tokens = outputs.get("usage", {}).get("prompt_tokens", 0)
completion_tokens = outputs.get("usage", {}).get("completion_tokens", 0)
else:
prompt_tokens = process_data.get("usage", {}).get("prompt_tokens", 0)
completion_tokens = process_data.get("usage", {}).get("completion_tokens", 0)
usage_data = process_data.get("usage", {}) if "usage" in process_data else outputs.get("usage", {})
prompt_tokens = usage_data.get("prompt_tokens", 0)
completion_tokens = usage_data.get("completion_tokens", 0)
except Exception:
logger.error("Failed to extract usage", exc_info=True)

8
api/core/ops/opik_trace/opik_trace.py
View File

@ -222,10 +222,10 @@ class OpikDataTrace(BaseTraceInstance):
)
try:
if outputs.get("usage"):
total_tokens = outputs["usage"].get("total_tokens", 0)
prompt_tokens = outputs["usage"].get("prompt_tokens", 0)
completion_tokens = outputs["usage"].get("completion_tokens", 0)
usage_data = process_data.get("usage", {}) if "usage" in process_data else outputs.get("usage", {})
total_tokens = usage_data.get("total_tokens", 0)
prompt_tokens = usage_data.get("prompt_tokens", 0)
completion_tokens = usage_data.get("completion_tokens", 0)
except Exception:
logger.error("Failed to extract usage", exc_info=True)

20
api/core/workflow/nodes/llm/node.py
View File

@ -221,15 +221,6 @@ class LLMNode(BaseNode[LLMNodeData]):
jinja2_variables=self.node_data.prompt_config.jinja2_variables,
)
process_data = {
"model_mode": model_config.mode,
"prompts": PromptMessageUtil.prompt_messages_to_prompt_for_saving(
model_mode=model_config.mode, prompt_messages=prompt_messages
),
"model_provider": model_config.provider,
"model_name": model_config.model,
}
# handle invoke result
generator = self._invoke_llm(
node_data_model=self.node_data.model,
@ -253,6 +244,17 @@ class LLMNode(BaseNode[LLMNodeData]):
elif isinstance(event, LLMStructuredOutput):
structured_output = event
process_data = {
"model_mode": model_config.mode,
"prompts": PromptMessageUtil.prompt_messages_to_prompt_for_saving(
model_mode=model_config.mode, prompt_messages=prompt_messages
),
"usage": jsonable_encoder(usage),
"finish_reason": finish_reason,
"model_provider": model_config.provider,
"model_name": model_config.model,
}
outputs = {"text": result_text, "usage": jsonable_encoder(usage), "finish_reason": finish_reason}
if structured_output:
outputs["structured_output"] = structured_output.structured_output

7
api/core/workflow/nodes/parameter_extractor/parameter_extractor_node.py
View File

@ -253,7 +253,12 @@ class ParameterExtractorNode(BaseNode):
status=WorkflowNodeExecutionStatus.SUCCEEDED,
inputs=inputs,
process_data=process_data,
outputs={"__is_success": 1 if not error else 0, "__reason": error, **result},
outputs={
"__is_success": 1 if not error else 0,
"__reason": error,
"__usage": jsonable_encoder(usage),
**result,
},
metadata={
WorkflowNodeExecutionMetadataKey.TOTAL_TOKENS: usage.total_tokens,
WorkflowNodeExecutionMetadataKey.TOTAL_PRICE: usage.total_price,

6
api/core/workflow/nodes/question_classifier/question_classifier_node.py
View File

@ -145,7 +145,11 @@ class QuestionClassifierNode(LLMNode):
"model_provider": model_config.provider,
"model_name": model_config.model,
}
outputs = {"class_name": category_name, "class_id": category_id}
outputs = {
"class_name": category_name,
"class_id": category_id,
"usage": jsonable_encoder(usage),
}
return NodeRunResult(
status=WorkflowNodeExecutionStatus.SUCCEEDED,