Internal
Replace the endpoint's traffic allocation across ALL backends (platform admin)
Percentage-based traffic split over the model's backends. Every backend must be listed; an unlisted backend is rejected rather than silently zeroed. Targets are percentages (they must sum to 100); the server normalizes them to integer weights once, with largest-remainder rounding, so every consumer of this API shares one deterministic rule. A PUT that targets a self-hosted pool whose lane has a live rollout campaign answers 409: pause the rollout, then edit the allocation.
put/v1/admin/inference/catalog/{modelId}/allocation
Path parameters
modelIdstring required
Request body
Response
Applied; echoes the persisted integer weights