Commit
•
ad569d5
1
Parent(s):
2a2c118
Update app.py
Browse files
app.py
CHANGED
@@ -2,6 +2,7 @@ import gradio as gr
|
|
2 |
import torch
|
3 |
from diffusers import StableDiffusionXLPipeline, AutoencoderKL
|
4 |
from huggingface_hub import hf_hub_download
|
|
|
5 |
from share_btn import community_icon_html, loading_icon_html, share_js
|
6 |
from cog_sdxl_dataset_and_utils import TokenEmbeddingsHandler
|
7 |
import lora
|
@@ -26,12 +27,19 @@ with open("sdxl_loras.json", "r") as file:
|
|
26 |
}
|
27 |
for item in data
|
28 |
]
|
29 |
-
print(sdxl_loras)
|
30 |
-
saved_names = [
|
31 |
-
hf_hub_download(item["repo"], item["weights"]) for item in sdxl_loras
|
32 |
-
]
|
33 |
|
34 |
-
device = "cuda"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
35 |
|
36 |
vae = AutoencoderKL.from_pretrained(
|
37 |
"madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16
|
@@ -40,14 +48,13 @@ pipe = StableDiffusionXLPipeline.from_pretrained(
|
|
40 |
"stabilityai/stable-diffusion-xl-base-1.0",
|
41 |
vae=vae,
|
42 |
torch_dtype=torch.float16,
|
43 |
-
)
|
44 |
original_pipe = copy.deepcopy(pipe)
|
45 |
pipe.to(device)
|
46 |
|
47 |
last_lora = ""
|
48 |
last_merged = False
|
49 |
|
50 |
-
|
51 |
def update_selection(selected_state: gr.SelectData):
|
52 |
lora_repo = sdxl_loras[selected_state.index]["repo"]
|
53 |
instance_prompt = sdxl_loras[selected_state.index]["trigger_word"]
|
@@ -128,7 +135,7 @@ def merge_incompatible_lora(full_path_lora, lora_scale):
|
|
128 |
del lora_model
|
129 |
gc.collect()
|
130 |
|
131 |
-
def run_lora(prompt, negative, lora_scale, selected_state):
|
132 |
global last_lora, last_merged, pipe
|
133 |
|
134 |
if negative == "":
|
@@ -138,7 +145,8 @@ def run_lora(prompt, negative, lora_scale, selected_state):
|
|
138 |
raise gr.Error("You must select a LoRA")
|
139 |
repo_name = sdxl_loras[selected_state.index]["repo"]
|
140 |
weight_name = sdxl_loras[selected_state.index]["weights"]
|
141 |
-
full_path_lora =
|
|
|
142 |
cross_attention_kwargs = None
|
143 |
if last_lora != repo_name:
|
144 |
if last_merged:
|
@@ -148,17 +156,17 @@ def run_lora(prompt, negative, lora_scale, selected_state):
|
|
148 |
pipe.to(device)
|
149 |
else:
|
150 |
pipe.unload_lora_weights()
|
|
|
151 |
is_compatible = sdxl_loras[selected_state.index]["is_compatible"]
|
152 |
if is_compatible:
|
153 |
-
pipe.load_lora_weights(
|
154 |
-
|
155 |
else:
|
156 |
is_pivotal = sdxl_loras[selected_state.index]["is_pivotal"]
|
157 |
if(is_pivotal):
|
|
|
|
|
158 |
|
159 |
-
pipe.load_lora_weights(full_path_lora)
|
160 |
-
cross_attention_kwargs = {"scale": lora_scale}
|
161 |
-
|
162 |
#Add the textual inversion embeddings from pivotal tuning models
|
163 |
text_embedding_name = sdxl_loras[selected_state.index]["text_embedding_weights"]
|
164 |
text_encoders = [pipe.text_encoder, pipe.text_encoder_2]
|
@@ -177,7 +185,6 @@ def run_lora(prompt, negative, lora_scale, selected_state):
|
|
177 |
height=768,
|
178 |
num_inference_steps=20,
|
179 |
guidance_scale=7.5,
|
180 |
-
cross_attention_kwargs=cross_attention_kwargs,
|
181 |
).images[0]
|
182 |
last_lora = repo_name
|
183 |
gc.collect()
|
|
|
2 |
import torch
|
3 |
from diffusers import StableDiffusionXLPipeline, AutoencoderKL
|
4 |
from huggingface_hub import hf_hub_download
|
5 |
+
from safetensors.torch import load_file
|
6 |
from share_btn import community_icon_html, loading_icon_html, share_js
|
7 |
from cog_sdxl_dataset_and_utils import TokenEmbeddingsHandler
|
8 |
import lora
|
|
|
27 |
}
|
28 |
for item in data
|
29 |
]
|
|
|
|
|
|
|
|
|
30 |
|
31 |
+
device = "cuda"
|
32 |
+
|
33 |
+
for item in sdxl_loras:
|
34 |
+
saved_name = hf_hub_download(item["repo"], item["weights"])
|
35 |
+
|
36 |
+
if not saved_name.endswith('.safetensors'):
|
37 |
+
state_dict = torch.load(saved_name)
|
38 |
+
else:
|
39 |
+
state_dict = load_file(saved_name)
|
40 |
+
|
41 |
+
item["saved_name"] = saved_name
|
42 |
+
item["state_dict"] = state_dict #{k: v.to(device=device, dtype=torch.float16) for k, v in state_dict.items() if torch.is_tensor(v)}
|
43 |
|
44 |
vae = AutoencoderKL.from_pretrained(
|
45 |
"madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16
|
|
|
48 |
"stabilityai/stable-diffusion-xl-base-1.0",
|
49 |
vae=vae,
|
50 |
torch_dtype=torch.float16,
|
51 |
+
)
|
52 |
original_pipe = copy.deepcopy(pipe)
|
53 |
pipe.to(device)
|
54 |
|
55 |
last_lora = ""
|
56 |
last_merged = False
|
57 |
|
|
|
58 |
def update_selection(selected_state: gr.SelectData):
|
59 |
lora_repo = sdxl_loras[selected_state.index]["repo"]
|
60 |
instance_prompt = sdxl_loras[selected_state.index]["trigger_word"]
|
|
|
135 |
del lora_model
|
136 |
gc.collect()
|
137 |
|
138 |
+
def run_lora(prompt, negative, lora_scale, selected_state, progress=gr.Progress(track_tqdm=True)):
|
139 |
global last_lora, last_merged, pipe
|
140 |
|
141 |
if negative == "":
|
|
|
145 |
raise gr.Error("You must select a LoRA")
|
146 |
repo_name = sdxl_loras[selected_state.index]["repo"]
|
147 |
weight_name = sdxl_loras[selected_state.index]["weights"]
|
148 |
+
full_path_lora = sdxl_loras[selected_state.index]["saved_name"]
|
149 |
+
loaded_state_dict = sdxl_loras[selected_state.index]["state_dict"]
|
150 |
cross_attention_kwargs = None
|
151 |
if last_lora != repo_name:
|
152 |
if last_merged:
|
|
|
156 |
pipe.to(device)
|
157 |
else:
|
158 |
pipe.unload_lora_weights()
|
159 |
+
pipe.unfuse_lora()
|
160 |
is_compatible = sdxl_loras[selected_state.index]["is_compatible"]
|
161 |
if is_compatible:
|
162 |
+
pipe.load_lora_weights(loaded_state_dict)
|
163 |
+
pipe.fuse_lora(lora_scale)
|
164 |
else:
|
165 |
is_pivotal = sdxl_loras[selected_state.index]["is_pivotal"]
|
166 |
if(is_pivotal):
|
167 |
+
pipe.load_lora_weights(loaded_state_dict)
|
168 |
+
pipe.fuse_lora(lora_scale)
|
169 |
|
|
|
|
|
|
|
170 |
#Add the textual inversion embeddings from pivotal tuning models
|
171 |
text_embedding_name = sdxl_loras[selected_state.index]["text_embedding_weights"]
|
172 |
text_encoders = [pipe.text_encoder, pipe.text_encoder_2]
|
|
|
185 |
height=768,
|
186 |
num_inference_steps=20,
|
187 |
guidance_scale=7.5,
|
|
|
188 |
).images[0]
|
189 |
last_lora = repo_name
|
190 |
gc.collect()
|