Technologies/Langfuse/langfuse.generation.tokens.completion
LangfuseLangfuseMetric

langfuse.generation.tokens.completion

Completion token count
Dimensions:None
Available on:DynatraceDynatrace (1)
Interface Metrics (1)
DynatraceDynatrace
Number of tokens in the completion/output
Dimensions:None
Sources
langfuse.generation.tokens.completionlangfuse.com

Technical Annotations (22)

Configuration Parameters (7)
monthly_budget_usdrecommended: 5000.0
Budget threshold for cost alerting (example value from code)
usage_details.inputrecommended: response.usage.input_tokens
map input token count from API response
usage_details.outputrecommended: response.usage.output_tokens
map output token count from API response
usage_details.totalrecommended: response.usage.total_tokens
map total token count from API response
usage.prompt_tokensrecommended: int(usage.prompt_token_count)
Map Gemini prompt tokens to Langfuse format
usage.completion_tokensrecommended: int(usage.candidates_token_count)
Map Gemini completion tokens (not total) to Langfuse format
usage.total_tokensrecommended: int(usage.total_token_count)
Explicitly set total to avoid recalculation errors
Error Signatures (1)
TypeError: Langfuse.update_current_generation() got an unexpected keyword argument 'usage'exception
Technical References (14)
token pricingconceptcost trackingconcepttiktokencomponentOpenTelemetryprotocol@anthropic-ai/tokenizercomponentClaude 3componentPython SDKcomponentBedrockcomponentgen_ai.usage.output_tokenscomponentcandidates_token_countcomponenttotal_token_countcomponentprompt templateconceptsystem promptcomponenttokenconcept
Related Insights (11)
Token usage metrics missing from LLM generation logswarning
▸
Token usage approaching budget limits triggers cost overrun riskwarning
▸
LLM monthly cost exceeds 90% of budgetwarning
▸
Tiktoken tokenization causes high CPU usage on large trace inputswarning
▸
Zero token counts for Claude models via OpenTelemetry after initial traceswarning
▸
Python SDK v2.57.1 reports zero tokens for Bedrock Claude callswarning
▸
Missing trace flush causes lost token counts in automated testswarning
▸
Streaming responses produce zero token counts without manual instrumentationinfo
▸
Model Usage chart displays inconsistent cost calculationswarning
▸
Gemini token counts misattributed via OpenTelemetry mappingwarning
▸
Token-bloated prompts driving excessive API costswarning
▸