Merge diffusion-auto-policy: request type accepts explicit Dtype off

This commit is contained in:
Daniel Han 2026-07-04 09:46:47 +00:00
commit ba3b521fa7

View file

@ -57,7 +57,7 @@ export interface DiffusionLoadRequest {
cpu_offload?: boolean;
// Advanced (load-time) tuning. All optional; omit for the backend's auto defaults.
speed_mode?: "off" | "eager" | "default" | "max";
transformer_quant?: "auto" | "int8" | "fp8" | "nvfp4" | "mxfp8";
transformer_quant?: "auto" | "none" | "off" | "int8" | "fp8" | "nvfp4" | "mxfp8";
attention_backend?:
| "auto"
| "native"