Closed PotatoBananaApple closed 10 months ago
It started working after i messed around with the config file.
Do you happen to remember what you changed that made it work?
Sorry i don't know exactly what was causing the issue. I tried to troubleshoot little comparing 2 configs, 1 of them works and one of them does not work. This time i get this kind of error on the other one. I'll attach the configs if you want to have look. I can´t find what is causing the issue.
Running 1 process
create LoRA network. base dim (rank): 8, alpha: 4
neuron dropout: p=None, rank dropout: p=None, module dropout: p=None
create LoRA for Text Encoder: 0 modules.
create LoRA for U-Net: 128 modules.
enable LoRA for U-Net
Prompt tensors not found. Encoding prompts..
Error running job: 'NoneType' object has no attribute 'text_embeds'
========================================
Result:
- 0 completed jobs
- 1 failure
========================================
Traceback (most recent call last):
File "C:\ai-toolkit\run.py", line 75, in <module>
main()
File "C:\ai-toolkit\run.py", line 71, in main
raise e
File "C:\ai-toolkit\run.py", line 63, in main
job.run()
File "C:\ai-toolkit\jobs\TrainJob.py", line 50, in run
process.run()
File "C:\ai-toolkit\jobs\process\BaseSDTrainProcess.py", line 344, in run
self.hook_before_train_loop()
File "C:\ai-toolkit\jobs\process\TrainSliderProcess.py", line 114, in hook_before_train_loop
concat_prompt_pair_batch = concat_prompt_pairs(prompt_pair_batch).to('cpu')
File "C:\ai-toolkit\toolkit\prompt_utils.py", line 88, in concat_prompt_pairs
positive_target = concat_prompt_embeds([p.positive_target for p in prompt_pairs])
File "C:\ai-toolkit\toolkit\prompt_utils.py", line 77, in concat_prompt_embeds
text_embeds = torch.cat([p.text_embeds for p in prompt_embeds], dim=0)
File "C:\ai-toolkit\toolkit\prompt_utils.py", line 77, in <listcomp>
text_embeds = torch.cat([p.text_embeds for p in prompt_embeds], dim=0)
AttributeError: 'NoneType' object has no attribute 'text_embeds'
Another thing i was wondering, is it supposed to be this slow? I have rtx 3060 12gb
#############################################
# Running job: test1
#############################################
Running 1 process
create LoRA network. base dim (rank): 8, alpha: 4
neuron dropout: p=None, rank dropout: p=None, module dropout: p=None
create LoRA for Text Encoder: 0 modules.
create LoRA for U-Net: 128 modules.
enable LoRA for U-Net
Prompt tensors not found. Encoding prompts..
Generating baseline samples before training
test1: 0%|▏ | 2/500 [00:44<2:52:32, 20.79s/it, lr: 1.0e-04 loss: 1.041e-04]
21%|█████████████████▊ | 6/28 [00:03<00:13, 1.60it/s]
@PotatoBananaApple That seems very slow. I have a laptop with that same GPU (except the laptop version) Ill try it out on there to see. I suspect it is overflowing the ram. The new driver for windows won't crash, it just does dynamic CPU offloading which is SSSLLLLOOOOOOWWWWW. What kind of ram usage are you seeing? Are you sure it is utilizing the gpu?
@jaretburkett Hey, the vram does not hit the cap, plenty of room. It is using cuda +80% while training.
Updated driver, did not change anything.
@jaretburkett Did clean install on sandbox, no interference from anything, same 1.5it-1.7it/s. Tried with cu118, similar speed.
What kind of speed you getting with your laptop?
@jaretburkett I also have the same question, does 1.7it/s very slow? I have 2080ti and when I train a normal lora using sd-scripts, the speed is like 1.5-1.7it/s
Trying to run the
LoRA Slider Trainer
example on windows gives me this error