Model identity and execution limits for a diffusion network. More...
#include <dipla_planner.hpp>
Public Attributes | |
| std::string | descriptor |
| Optional integrity-checked model descriptor path. | |
| unsigned | threads = 1 |
| CPU inference threads, bounded to 1..32; one avoids competing with other MORA workers by default. | |
| unsigned | timeout_ms = 0 |
| Whole-plan cooperative deadline in milliseconds; zero leaves supervision to the caller. | |
| std::string | conditioning_cache |
| Optional verified encoder/denoiser bundle; empty retains the original graph. | |
Detailed Description
Model identity and execution limits for a diffusion network.
A graph and its preprocessing/tensor contract form one assignment. The timeout applies to the complete 20-pass sampler, not separately to each pass; otherwise a nominal budget could expand twentyfold.
Member Data Documentation
◆ conditioning_cache
| std::string dipla::NetworkOptions::conditioning_cache |
Optional verified encoder/denoiser bundle; empty retains the original graph.
◆ descriptor
| std::string dipla::NetworkOptions::descriptor |
Optional integrity-checked model descriptor path.
◆ threads
| unsigned dipla::NetworkOptions::threads = 1 |
CPU inference threads, bounded to 1..32; one avoids competing with other MORA workers by default.
◆ timeout_ms
| unsigned dipla::NetworkOptions::timeout_ms = 0 |
Whole-plan cooperative deadline in milliseconds; zero leaves supervision to the caller.
The documentation for this struct was generated from the following file: