Update app.py
Browse files
app.py
CHANGED
@@ -91,9 +91,6 @@ class ModelWrapper:
|
|
91 |
|
92 |
DTYPE = prompt_embed.dtype
|
93 |
print(DTYPE)
|
94 |
-
print(type(noise))
|
95 |
-
print(type(current_timesteps))
|
96 |
-
print(type(unet_added_conditions))
|
97 |
|
98 |
for constant in all_timesteps:
|
99 |
current_timesteps = torch.ones(len(prompt_embed), device="cuda", dtype=torch.long) * constant
|
@@ -124,7 +121,7 @@ class ModelWrapper:
|
|
124 |
|
125 |
add_time_ids = self.build_condition_input(height, width).repeat(num_images, 1)
|
126 |
|
127 |
-
noise = torch.randn(num_images, 4, height // self.vae_downsample_ratio, width // self.vae_downsample_ratio, generator=generator).to(device="cuda", dtype=
|
128 |
|
129 |
prompt_inputs = self._encode_prompt(prompt)
|
130 |
|
@@ -161,7 +158,7 @@ def get_x0_from_noise(sample, model_output, alphas_cumprod, timestep):
|
|
161 |
return pred_original_sample
|
162 |
|
163 |
class SDXLTextEncoder(torch.nn.Module):
|
164 |
-
def __init__(self, model_id, revision, accelerator, dtype=torch.
|
165 |
super().__init__()
|
166 |
|
167 |
self.text_encoder_one = CLIPTextModel.from_pretrained(model_id, subfolder="text_encoder", revision=revision).to(0).to(dtype=dtype)
|
|
|
91 |
|
92 |
DTYPE = prompt_embed.dtype
|
93 |
print(DTYPE)
|
|
|
|
|
|
|
94 |
|
95 |
for constant in all_timesteps:
|
96 |
current_timesteps = torch.ones(len(prompt_embed), device="cuda", dtype=torch.long) * constant
|
|
|
121 |
|
122 |
add_time_ids = self.build_condition_input(height, width).repeat(num_images, 1)
|
123 |
|
124 |
+
noise = torch.randn(num_images, 4, height // self.vae_downsample_ratio, width // self.vae_downsample_ratio, generator=generator).to(device="cuda", dtype=float16)
|
125 |
|
126 |
prompt_inputs = self._encode_prompt(prompt)
|
127 |
|
|
|
158 |
return pred_original_sample
|
159 |
|
160 |
class SDXLTextEncoder(torch.nn.Module):
|
161 |
+
def __init__(self, model_id, revision, accelerator, dtype=torch.float16):
|
162 |
super().__init__()
|
163 |
|
164 |
self.text_encoder_one = CLIPTextModel.from_pretrained(model_id, subfolder="text_encoder", revision=revision).to(0).to(dtype=dtype)
|