diff options
| -rw-r--r-- | modules/devices.py | 28 | ||||
| -rw-r--r-- | modules/processing.py | 3 | ||||
| -rw-r--r-- | modules/sd_samplers.py | 2 |
3 files changed, 33 insertions, 0 deletions
diff --git a/modules/devices.py b/modules/devices.py index caeb0276..6f034948 100644 --- a/modules/devices.py +++ b/modules/devices.py @@ -106,6 +106,33 @@ def autocast(disable=False): return torch.autocast("cuda") +class NansException(Exception): + pass + + +def test_for_nans(x, where): + from modules import shared + + if not torch.all(torch.isnan(x)).item(): + return + + if where == "unet": + message = "A tensor with all NaNs was produced in Unet." + + if not shared.cmd_opts.no_half: + message += " This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try using --no-half commandline argument to fix this." + + elif where == "vae": + message = "A tensor with all NaNs was produced in VAE." + + if not shared.cmd_opts.no_half and not shared.cmd_opts.no_half_vae: + message += " This could be because there's not enough precision to represent the picture. Try adding --no-half-vae commandline argument to fix this." + else: + message = "A tensor with all NaNs was produced." + + raise NansException(message) + + # MPS workaround for https://github.com/pytorch/pytorch/issues/79383 orig_tensor_to = torch.Tensor.to def tensor_to_fix(self, *args, **kwargs): @@ -156,3 +183,4 @@ if has_mps(): torch.Tensor.cumsum = lambda self, *args, **kwargs: ( cumsum_fix(self, orig_Tensor_cumsum, *args, **kwargs) ) orig_narrow = torch.narrow torch.narrow = lambda *args, **kwargs: ( orig_narrow(*args, **kwargs).clone() ) + diff --git a/modules/processing.py b/modules/processing.py index 849f6b19..ab7b3b7d 100644 --- a/modules/processing.py +++ b/modules/processing.py @@ -608,6 +608,9 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed: samples_ddim = p.sample(conditioning=c, unconditional_conditioning=uc, seeds=seeds, subseeds=subseeds, subseed_strength=p.subseed_strength, prompts=prompts)
x_samples_ddim = [decode_first_stage(p.sd_model, samples_ddim[i:i+1].to(dtype=devices.dtype_vae))[0].cpu() for i in range(samples_ddim.size(0))]
+ for x in x_samples_ddim:
+ devices.test_for_nans(x, "vae")
+
x_samples_ddim = torch.stack(x_samples_ddim).float()
x_samples_ddim = torch.clamp((x_samples_ddim + 1.0) / 2.0, min=0.0, max=1.0)
diff --git a/modules/sd_samplers.py b/modules/sd_samplers.py index 76e0e0d5..6261d1f7 100644 --- a/modules/sd_samplers.py +++ b/modules/sd_samplers.py @@ -351,6 +351,8 @@ class CFGDenoiser(torch.nn.Module): x_out[-uncond.shape[0]:] = self.inner_model(x_in[-uncond.shape[0]:], sigma_in[-uncond.shape[0]:], cond={"c_crossattn": [uncond], "c_concat": [image_cond_in[-uncond.shape[0]:]]})
+ devices.test_for_nans(x_out, "unet")
+
if opts.live_preview_content == "Prompt":
store_latent(x_out[0:uncond.shape[0]])
elif opts.live_preview_content == "Negative prompt":
|