mirror of
https://github.com/BerriAI/litellm.git
synced 2026-10-11 03:38:38 +00:00
70cb219dfa
When LiteLLM cannot price a call (model newer than the bundled cost map), the standard logging payload gets response_cost = 0.0 together with response_cost_failure_debug_info. The span writers only skipped None, so OTel v2 spans carried litellm.cost.total = 0 and Arize/Phoenix spans carried llm.cost.total = 0 — indistinguishable from a genuinely free model. Langfuse prefers an ingested cost over its own price knowledge, so every call to an unpriced model showed $0 there (issue #44186). When response_cost_failure_debug_info is present, the cost is now reported as unknown (None) instead of 0 in both the OTel v2 span payload builders and the Arize attribute writer. A genuine 0 without the failure marker (free model) is still emitted as 0. |
||
|---|---|---|
| .. | ||
| __init__.py | ||
| test_arize.py | ||
| test_arize_health_check.py | ||
| test_arize_otel_coexistence.py | ||
| test_arize_phoenix.py | ||
| test_arize_utils.py | ||