Spaces:
Running
on
Zero
Running
on
Zero
wondervictor
commited on
Update model.py
Browse files
model.py
CHANGED
@@ -64,10 +64,14 @@ class Model:
|
|
64 |
return gpt_model
|
65 |
|
66 |
def load_gpt_weight(self, condition_type='edge'):
|
|
|
|
|
67 |
gpt_ckpt = models[condition_type]
|
68 |
model_weight = load_file(gpt_ckpt)
|
69 |
self.gpt_model.load_state_dict(model_weight, strict=False)
|
70 |
self.gpt_model.eval()
|
|
|
|
|
71 |
# print("gpt model is loaded")
|
72 |
|
73 |
def load_t5(self):
|
@@ -193,10 +197,12 @@ class Model:
|
|
193 |
control_strength: float,
|
194 |
preprocessor_name: str
|
195 |
) -> list[PIL.Image.Image]:
|
196 |
-
self.gpt_model_edge.to('cpu')
|
197 |
self.t5_model.model.to(self.device)
|
198 |
-
self.gpt_model_depth.to(self.device)
|
199 |
-
self.
|
|
|
|
|
200 |
self.vq_model.to(self.device)
|
201 |
if isinstance(image, np.ndarray):
|
202 |
image = Image.fromarray(image)
|
@@ -237,7 +243,7 @@ class Model:
|
|
237 |
qzshape = [len(c_indices), 8, H // 16, W // 16]
|
238 |
t1 = time.time()
|
239 |
index_sample = generate(
|
240 |
-
self.
|
241 |
c_indices,
|
242 |
(H // 16) * (W // 16),
|
243 |
c_emb_masks,
|
|
|
64 |
return gpt_model
|
65 |
|
66 |
def load_gpt_weight(self, condition_type='edge'):
|
67 |
+
torch.cuda.empty_cache()
|
68 |
+
gc.collect()
|
69 |
gpt_ckpt = models[condition_type]
|
70 |
model_weight = load_file(gpt_ckpt)
|
71 |
self.gpt_model.load_state_dict(model_weight, strict=False)
|
72 |
self.gpt_model.eval()
|
73 |
+
torch.cuda.empty_cache()
|
74 |
+
gc.collect()
|
75 |
# print("gpt model is loaded")
|
76 |
|
77 |
def load_t5(self):
|
|
|
197 |
control_strength: float,
|
198 |
preprocessor_name: str
|
199 |
) -> list[PIL.Image.Image]:
|
200 |
+
# self.gpt_model_edge.to('cpu')
|
201 |
self.t5_model.model.to(self.device)
|
202 |
+
# self.gpt_model_depth.to(self.device)
|
203 |
+
self.load_gpt_weight('depth')
|
204 |
+
self.gpt_model.to('cuda').to(torch.bfloat16)
|
205 |
+
# self.get_control_depth.model.to(self.device)
|
206 |
self.vq_model.to(self.device)
|
207 |
if isinstance(image, np.ndarray):
|
208 |
image = Image.fromarray(image)
|
|
|
243 |
qzshape = [len(c_indices), 8, H // 16, W // 16]
|
244 |
t1 = time.time()
|
245 |
index_sample = generate(
|
246 |
+
self.gpt_model,
|
247 |
c_indices,
|
248 |
(H // 16) * (W // 16),
|
249 |
c_emb_masks,
|