DecodeWindowConfig

class DecodeWindowConfig

Configuration for decode window analysis.

Fields:

min_active_requestsint | str = 1

Minimum number of simultaneously generating (decoding) requests required for a time interval to be considered inside the decode window. Use ‘max_observed’ to auto-detect the peak concurrent decoding count.

selection_strategystr = "longest"

Which window(s) to analyze when multiple windows exist. Supported: ‘longest’ (single longest), ‘first’ (single first), ‘all’ (aggregate all qualifying windows).

anchor_to_client_pickupbool = True

If True, anchor per-request token times to client_picked_up_at when available; otherwise use scheduler_dispatched_at.

require_streamingbool = True

If True, only streaming requests contribute to decode window analysis.