ReasoningTokenShare metric
Bases: LMOperationalMetric
Fraction of completion tokens spent on reasoning: reasoning_tokens / completion_tokens. Signals whether a thinking model is actually thinking on the workload.
Example:
Bases: LMOperationalMetric
Fraction of completion tokens spent on reasoning: reasoning_tokens / completion_tokens. Signals whether a thinking model is actually thinking on the workload.
Example:
synalinks/src/metrics/lm_metrics.py