Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse fileschanged model loading
app.py
CHANGED
|
@@ -18,7 +18,10 @@ device = "cuda" if torch.cuda.is_available() else "cpu"
|
|
| 18 |
# device = "cpu"
|
| 19 |
# model_type = 'SD3'
|
| 20 |
|
| 21 |
-
|
|
|
|
|
|
|
|
|
|
| 22 |
# scheduler = pipe.scheduler
|
| 23 |
# pipe = pipe.to(device)
|
| 24 |
loaded_model = 'None'
|
|
@@ -63,7 +66,7 @@ def get_examples():
|
|
| 63 |
return case
|
| 64 |
|
| 65 |
|
| 66 |
-
@spaces.GPU(duration=
|
| 67 |
def FlowEditRun(
|
| 68 |
image_src: str,
|
| 69 |
model_type: str,
|
|
@@ -84,22 +87,21 @@ def FlowEditRun(
|
|
| 84 |
if not len(tar_prompt):
|
| 85 |
raise gr.Error("target prompt cannot be empty")
|
| 86 |
|
| 87 |
-
global
|
| 88 |
-
global scheduler
|
| 89 |
-
global loaded_model
|
| 90 |
|
| 91 |
# reload model only if different from the loaded model
|
| 92 |
-
if loaded_model != model_type:
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
|
| 97 |
-
|
| 98 |
-
|
| 99 |
-
|
| 100 |
-
|
| 101 |
-
|
| 102 |
-
raise NotImplementedError(f"Model type {model_type} not implemented")
|
| 103 |
|
| 104 |
scheduler = pipe.scheduler
|
| 105 |
pipe = pipe.to(device)
|
|
|
|
| 18 |
# device = "cpu"
|
| 19 |
# model_type = 'SD3'
|
| 20 |
|
| 21 |
+
pipe_sd3 = StableDiffusion3Pipeline.from_pretrained("stabilityai/stable-diffusion-3-medium-diffusers", torch_dtype=torch.float16, token=os.getenv('HF_ACCESS_TOK'))
|
| 22 |
+
pipe_flux = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.float16, token=os.getenv('HF_ACCESS_TOK'))
|
| 23 |
+
|
| 24 |
+
|
| 25 |
# scheduler = pipe.scheduler
|
| 26 |
# pipe = pipe.to(device)
|
| 27 |
loaded_model = 'None'
|
|
|
|
| 66 |
return case
|
| 67 |
|
| 68 |
|
| 69 |
+
@spaces.GPU(duration=60)
|
| 70 |
def FlowEditRun(
|
| 71 |
image_src: str,
|
| 72 |
model_type: str,
|
|
|
|
| 87 |
if not len(tar_prompt):
|
| 88 |
raise gr.Error("target prompt cannot be empty")
|
| 89 |
|
| 90 |
+
# global pipe_sd3
|
| 91 |
+
# global scheduler
|
| 92 |
+
# global loaded_model
|
| 93 |
|
| 94 |
# reload model only if different from the loaded model
|
| 95 |
+
# if loaded_model != model_type:
|
| 96 |
+
|
| 97 |
+
if model_type == 'FLUX':
|
| 98 |
+
# pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.float16, token=os.getenv('HF_ACCESS_TOK'))
|
| 99 |
+
pipe = pipe_flux.clone() # still on CPU
|
| 100 |
+
elif model_type == 'SD3':
|
| 101 |
+
# pipe = StableDiffusion3Pipeline.from_pretrained("stabilityai/stable-diffusion-3-medium-diffusers", torch_dtype=torch.float16, token=os.getenv('HF_ACCESS_TOK'))
|
| 102 |
+
pipe = pipe_sd3.clone() # still on CPU
|
| 103 |
+
else:
|
| 104 |
+
raise NotImplementedError(f"Model type {model_type} not implemented")
|
|
|
|
| 105 |
|
| 106 |
scheduler = pipe.scheduler
|
| 107 |
pipe = pipe.to(device)
|