jaretburkett
commented
2 months ago the 'flowmatch' sampler is a euler version that supports flowmatching. The sampler cannot handle other samplers currently as none of the other ones in diffusers support flowmatching. Just stick to flowmatching for now for both sampling and training.
The initial sampling fails when sampler is set to euler. It's fine when reverted back to to flowmatch. BTW what is this flowmatch sampler? Haven't seen it anywhere before.
Error msg:
Traceback (most recent call last): File "/ssd/62_only/bjhonson/flux_training/ai-toolkit/run.py", line 90, in
main()
File "/ssd/62_only/bjhonson/flux_training/ai-toolkit/run.py", line 86, in main
raise e
File "/ssd/62_only/bjhonson/flux_training/ai-toolkit/run.py", line 78, in main
job.run()
File "/ssd/62_only/bjhonson/flux_training/ai-toolkit/jobs/ExtensionJob.py", line 22, in run
process.run()
File "/ssd/62_only/bjhonson/flux_training/ai-toolkit/jobs/process/BaseSDTrainProcess.py", line 1582, in run
self.sample(self.step_num)
File "/ssd/62_only/bjhonson/flux_training/ai-toolkit/jobs/process/BaseSDTrainProcess.py", line 269, in sample
self.sd.generate_images(gen_img_config_list, sampler=sample_config.sampler)
File "/ssd/62_only/bjhonson/miniconda3/envs/lora_training/lib/python3.11/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
return func(*args, *kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/ssd/62_only/bjhonson/flux_training/ai-toolkit/toolkit/stable_diffusion_model.py", line 1140, in generate_images
img = pipeline(
^^^^^^^^^
File "/ssd/62_only/bjhonson/miniconda3/envs/lora_training/lib/python3.11/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
return func(args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/ssd/62_only/bjhonson/miniconda3/envs/lora_training/lib/python3.11/site-packages/diffusers/pipelines/flux/pipeline_flux.py", line 664, in call
self.scheduler.config.base_image_seq_len,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: 'FrozenDict' object has no attribute 'base_image_seq_len'