Commit Wan Trainer Version 1.1.0:
* Added Sampler node to generate images every n steps or epochs (Images are in the folder "sample" in your output Lora path).
* BlockSwap node fix triggering it to the launcher.
* Network weights fix triggering the assigned path (now you are able to train from a previous lora setting the her's path location).
* Network args trigger fix.
* Optimizer args trigger fix.
* Tom argument image extension setted up to ".txt" as default.
* Enable bucket no upscale setting trigger fix. (no works correctly to prevent upscale).
* Max train step setted up to 2048 as default (taking in account this set is multiplied by epochs and images).
* Setted up timestep_sampling to "shit" as default.
* Setted up epoch to 128 as default.
* Added 2 example workflows for 14B and 1.3B models (located in the folder "workflows", you can also download form the website).
* Code improvements.
Jaime Pascual Gonzalez
8b6bb81d48