update

installer.py

@@ -235,8 +235,10 @@ def check_torch():
         machine = platform.machine()
         if allow_directml and ('arm' not in machine and 'aarch' not in machine and args.use_directml):
             log.info('Using DirectML Backend')
-            torch_command = os.environ.get('TORCH_COMMAND', 'torch==2.0.0 torchvision==0.15.1 torch-directml')
+            torch_command = os.environ.get('TORCH_COMMAND', 'torch==2.0.0 torchaudio torchvision==0.15.1 torch-directml')
             xformers_package = os.environ.get('XFORMERS_PACKAGE', 'none')
+            if 'torch' in torch_command and not args.version:
+                install(torch_command, 'torch torchvision torchaudio')
         else:
             log.info('Using CPU-only Torch')
             torch_command = os.environ.get('TORCH_COMMAND', 'torch torchaudio torchvision==0.15.1')
@@ -480,7 +482,7 @@ def check_version(offline=False, reset=True): # pylint: disable=unused-argument
         log.error('Not a git repository')
         if not args.ignore:
             exit(1)
-    _status = git('status')
+    # status = git('status')
     # if 'branch' not in status:
     #    log.error('Cannot get git repository status')
     #    exit(1)

modules/devices.py

@@ -110,9 +110,10 @@ def set_cuda_params():
             try:
                 if any([torch.cuda.get_device_capability(devid) == (7, 5) for devid in range(0, torch.cuda.device_count())]): # monkey-patch for old nvidia cards
                     torch.backends.cudnn.benchmark = True
+                    torch.backends.cudnn.benchmark_limit = 1
                 else:
                     torch.backends.cudnn.benchmark = shared.opts.cudnn_benchmark
-                torch.backends.cudnn.benchmark_limit = 0
+                    torch.backends.cudnn.benchmark_limit = 0
                 torch.backends.cudnn.allow_tf32 = shared.opts.cuda_allow_tf32
             except:
                 pass

modules/processing.py

@@ -847,14 +847,12 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing):
             for i in range(samples.shape[0]):
                 save_intermediate(samples, i)
             samples = torch.nn.functional.interpolate(samples, size=(target_height // opt_f, target_width // opt_f), mode=latent_scale_mode["mode"], antialias=latent_scale_mode["antialias"])
-            # Avoid making the inpainting conditioning unless necessary as
-            # this does need some extra compute to decode / encode the image again.
             if getattr(self, "inpainting_mask_weight", shared.opts.inpainting_mask_weight) < 1.0:
-                image_conditioning = self.img2img_image_conditioning(decode_first_stage(self.sd_model, samples), samples)
+                image_conditioning = self.img2img_image_conditioning(decode_first_stage(self.sd_model, samples.to(dtype=devices.dtype_vae)), samples)
             else:
-                image_conditioning = self.txt2img_image_conditioning(samples)
+                image_conditioning = self.txt2img_image_conditioning(samples.to(dtype=devices.dtype_vae))
         else:
-            decoded_samples = decode_first_stage(self.sd_model, samples)
+            decoded_samples = decode_first_stage(self.sd_model, samples.to(dtype=devices.dtype_vae))
             lowres_samples = torch.clamp((decoded_samples + 1.0) / 2.0, min=0.0, max=1.0)
             batch_images = []
             for i, x_sample in enumerate(lowres_samples):

requirements.txt

@@ -60,3 +60,4 @@ transformers==4.26.1
 timm==0.6.13
 tomesd==0.1.2
 urllib3==1.26.15
+onnxruntime==1.14.0