patrickvonplaten commited on
Commit
c7d725d
1 Parent(s): 8947670

speed up demo

Browse files
Files changed (3) hide show
  1. README.md +1 -2
  2. app.py +7 -3
  3. requirements.txt +7 -4
README.md CHANGED
@@ -4,10 +4,9 @@ emoji: 👀
4
  colorFrom: pink
5
  colorTo: purple
6
  sdk: gradio
7
- sdk_version: 3.50.2
8
  app_file: app.py
9
  pinned: false
10
- license: agpl-3.0
11
  ---
12
 
13
  Inference Code: https://github.com/PixArt-alpha/PixArt-alpha
 
4
  colorFrom: pink
5
  colorTo: purple
6
  sdk: gradio
7
+ sdk_version: 4.1.1
8
  app_file: app.py
9
  pinned: false
 
10
  ---
11
 
12
  Inference Code: https://github.com/PixArt-alpha/PixArt-alpha
app.py CHANGED
@@ -22,7 +22,7 @@ if not torch.cuda.is_available():
22
  MAX_SEED = np.iinfo(np.int32).max
23
  CACHE_EXAMPLES = torch.cuda.is_available() and os.getenv("CACHE_EXAMPLES", "1") == "1"
24
  MAX_IMAGE_SIZE = int(os.getenv("MAX_IMAGE_SIZE", "1024"))
25
- USE_TORCH_COMPILE = os.getenv("USE_TORCH_COMPILE", "1") == "1"
26
  ENABLE_CPU_OFFLOAD = os.getenv("ENABLE_CPU_OFFLOAD", "0") == "1"
27
 
28
  device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
@@ -96,6 +96,7 @@ if torch.cuda.is_available():
96
  pipe = PixArtAlphaPipeline.from_pretrained(
97
  "PixArt-alpha/PixArt-XL-2-1024-MS",
98
  torch_dtype=torch.float16,
 
99
  use_safetensors=True,
100
  )
101
 
@@ -105,6 +106,9 @@ if torch.cuda.is_available():
105
  pipe.to(device)
106
  print("Loaded on Device!")
107
 
 
 
 
108
  if USE_TORCH_COMPILE:
109
  pipe.transformer = torch.compile(
110
  pipe.transformer, mode="reduce-overhead", fullgraph=True
@@ -254,7 +258,6 @@ with gr.Blocks(css="style.css") as demo:
254
  fn=lambda x: gr.update(visible=x),
255
  inputs=use_negative_prompt,
256
  outputs=negative_prompt,
257
- queue=False,
258
  api_name=False,
259
  )
260
 
@@ -282,4 +285,5 @@ with gr.Blocks(css="style.css") as demo:
282
  )
283
 
284
  if __name__ == "__main__":
285
- demo.queue(max_size=20).launch()
 
 
22
  MAX_SEED = np.iinfo(np.int32).max
23
  CACHE_EXAMPLES = torch.cuda.is_available() and os.getenv("CACHE_EXAMPLES", "1") == "1"
24
  MAX_IMAGE_SIZE = int(os.getenv("MAX_IMAGE_SIZE", "1024"))
25
+ USE_TORCH_COMPILE = os.getenv("USE_TORCH_COMPILE", "0") == "1"
26
  ENABLE_CPU_OFFLOAD = os.getenv("ENABLE_CPU_OFFLOAD", "0") == "1"
27
 
28
  device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
96
  pipe = PixArtAlphaPipeline.from_pretrained(
97
  "PixArt-alpha/PixArt-XL-2-1024-MS",
98
  torch_dtype=torch.float16,
99
+ variant="fp16",
100
  use_safetensors=True,
101
  )
102
 
 
106
  pipe.to(device)
107
  print("Loaded on Device!")
108
 
109
+ # speed-up T5
110
+ pipe.text_encoder.to_bettertransformer()
111
+
112
  if USE_TORCH_COMPILE:
113
  pipe.transformer = torch.compile(
114
  pipe.transformer, mode="reduce-overhead", fullgraph=True
 
258
  fn=lambda x: gr.update(visible=x),
259
  inputs=use_negative_prompt,
260
  outputs=negative_prompt,
 
261
  api_name=False,
262
  )
263
 
 
285
  )
286
 
287
  if __name__ == "__main__":
288
+ # demo.queue(max_size=20).launch()
289
+ demo.launch(share=True)
requirements.txt CHANGED
@@ -1,7 +1,10 @@
1
- diffusers==0.22.0
2
- accelerate==0.24.1
 
 
 
 
3
  gradio==4.1.1
4
  Pillow==10.1.0
5
- torch==2.0.1
6
- transformers==4.35.0
7
  sentencepiece==0.1.99
 
 
1
+ --index-url https://download.pytorch.org/whl/cu118
2
+ torch==2.0.1
3
+
4
+ diffusers==0.22.1
5
+ accelerate
6
+ transformers
7
  gradio==4.1.1
8
  Pillow==10.1.0
 
 
9
  sentencepiece==0.1.99
10
+