Merge fa87f263ce into d0f3752e33

This is required to support some different types of t5 models.

app/app_settings.py

@@ -1,6 +1,7 @@
 import os
 import json
 from aiohttp import web
+import logging
 
 
 class AppSettings():
@@ -11,8 +12,12 @@ class AppSettings():
         file = self.user_manager.get_request_user_filepath(
             request, "comfy.settings.json")
         if os.path.isfile(file):
+            try:
                 with open(file) as f:
                     return json.load(f)
+            except:
+                logging.error(f"The user settings file is corrupted: {file}")
+                return {}
         else:
             return {}
 

comfy/sample.py

@@ -5,12 +5,15 @@ import comfy.utils
 import numpy as np
 import logging
 
-def prepare_noise(latent_image, seed, noise_inds=None):
+def prepare_noise(latent_image, seed, noise_inds=None, disable_noise=False):
     """
     creates random noise given a latent image and a seed.
     optional arg skip can be used to skip and discard x number of noise generations for a given seed
     """
     generator = torch.manual_seed(seed)
+    if disable_noise:
+        return torch.zeros(latent_image.size(), dtype=latent_image.dtype, layout=latent_image.layout, device="cpu")
+
     if noise_inds is None:
         return torch.randn(latent_image.size(), dtype=latent_image.dtype, layout=latent_image.layout, generator=generator, device="cpu")
 

comfy/text_encoders/t5.py

@@ -227,8 +227,9 @@ class T5(torch.nn.Module):
         super().__init__()
         self.num_layers = config_dict["num_layers"]
         model_dim = config_dict["d_model"]
+        inner_dim = config_dict["d_kv"] * config_dict["num_heads"]
 
-        self.encoder = T5Stack(self.num_layers, model_dim, model_dim, config_dict["d_ff"], config_dict["dense_act_fn"], config_dict["is_gated_act"], config_dict["num_heads"], config_dict["model_type"] != "umt5", dtype, device, operations)
+        self.encoder = T5Stack(self.num_layers, model_dim, inner_dim, config_dict["d_ff"], config_dict["dense_act_fn"], config_dict["is_gated_act"], config_dict["num_heads"], config_dict["model_type"] != "umt5", dtype, device, operations)
         self.dtype = dtype
         self.shared = operations.Embedding(config_dict["vocab_size"], model_dim, device=device, dtype=dtype)
 

comfy_extras/nodes_custom_sampler.py

@@ -418,7 +418,7 @@ class Noise_EmptyNoise:
 
     def generate_noise(self, input_latent):
         latent_image = input_latent["samples"]
-        return torch.zeros(latent_image.shape, dtype=latent_image.dtype, layout=latent_image.layout, device="cpu")
+        return comfy.sample.prepare_noise(latent_image, self.seed, disable_noise=True)
 
 
 class Noise_RandomNoise:

nodes.py

@@ -1485,11 +1485,8 @@ def common_ksampler(model, seed, steps, cfg, sampler_name, scheduler, positive,
     latent_image = latent["samples"]
     latent_image = comfy.sample.fix_empty_latent_channels(model, latent_image)
 
-    if disable_noise:
-        noise = torch.zeros(latent_image.size(), dtype=latent_image.dtype, layout=latent_image.layout, device="cpu")
-    else:
     batch_inds = latent["batch_index"] if "batch_index" in latent else None
-        noise = comfy.sample.prepare_noise(latent_image, seed, batch_inds)
+    noise = comfy.sample.prepare_noise(latent_image, seed, batch_inds, disable_noise)
 
     noise_mask = None
     if "noise_mask" in latent: