Merge pull request #12515 from catboxanon/fix/gc1
Clear sampler and garbage collect before decoding images to reduce VRAM
This commit is contained in:
commit
d73db17ee3
|
@ -1195,6 +1195,9 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing):
|
|||
|
||||
sd_models.apply_token_merging(self.sd_model, self.get_token_merging_ratio())
|
||||
|
||||
self.sampler = None
|
||||
devices.torch_gc()
|
||||
|
||||
decoded_samples = decode_latent_batch(self.sd_model, samples, target_device=devices.cpu, check_for_nans=True)
|
||||
|
||||
self.is_hr_pass = False
|
||||
|
|
Loading…
Reference in New Issue