Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
|
@@ -31,29 +31,29 @@ def generate(slider_x, slider_y, prompt, seed, iterations, steps,
|
|
| 31 |
print("x_concept_1", x_concept_1, "x_concept_2", x_concept_2)
|
| 32 |
if not sorted(slider_x) == sorted([x_concept_1, x_concept_2]):
|
| 33 |
avg_diff = clip_slider.find_latent_direction(slider_x[0], slider_x[1], num_iterations=iterations)
|
| 34 |
-
|
| 35 |
-
|
| 36 |
x_concept_1, x_concept_2 = slider_x[0], slider_x[1]
|
| 37 |
|
| 38 |
print("avg_diff[0].dtype", avg_diff[0].dtype)
|
| 39 |
if not sorted(slider_y) == sorted([y_concept_1, y_concept_2]):
|
| 40 |
avg_diff_2nd = clip_slider.find_latent_direction(slider_y[0], slider_y[1], num_iterations=iterations)
|
| 41 |
-
|
| 42 |
-
|
| 43 |
y_concept_1, y_concept_2 = slider_y[0], slider_y[1]
|
| 44 |
end_time = time.time()
|
| 45 |
print(f"direction time: {end_time - start_time:.2f} ms")
|
| 46 |
start_time = time.time()
|
| 47 |
-
image = clip_slider.generate(prompt, scale=0, scale_2nd=0, seed=seed, num_inference_steps=steps, avg_diff=
|
| 48 |
end_time = time.time()
|
| 49 |
print(f"generation time: {end_time - start_time:.2f} ms")
|
| 50 |
comma_concepts_x = ', '.join(slider_x)
|
| 51 |
comma_concepts_y = ', '.join(slider_y)
|
| 52 |
|
| 53 |
-
avg_diff_x_1 =
|
| 54 |
-
avg_diff_x_2 =
|
| 55 |
-
avg_diff_y_1 =
|
| 56 |
-
avg_diff_y_2 =
|
| 57 |
|
| 58 |
return gr.update(label=comma_concepts_x, interactive=True),gr.update(label=comma_concepts_y, interactive=True), x_concept_1, x_concept_2, y_concept_1, y_concept_2, avg_diff_x_1, avg_diff_x_2, avg_diff_y_1, avg_diff_y_2, image
|
| 59 |
|
|
|
|
| 31 |
print("x_concept_1", x_concept_1, "x_concept_2", x_concept_2)
|
| 32 |
if not sorted(slider_x) == sorted([x_concept_1, x_concept_2]):
|
| 33 |
avg_diff = clip_slider.find_latent_direction(slider_x[0], slider_x[1], num_iterations=iterations)
|
| 34 |
+
avg_diff_0 = avg_diff[0].to(torch.float16)
|
| 35 |
+
avg_diff_1 = avg_diff[1].to(torch.float16)
|
| 36 |
x_concept_1, x_concept_2 = slider_x[0], slider_x[1]
|
| 37 |
|
| 38 |
print("avg_diff[0].dtype", avg_diff[0].dtype)
|
| 39 |
if not sorted(slider_y) == sorted([y_concept_1, y_concept_2]):
|
| 40 |
avg_diff_2nd = clip_slider.find_latent_direction(slider_y[0], slider_y[1], num_iterations=iterations)
|
| 41 |
+
avg_diff_2nd_0 = avg_diff_2nd[0].to(torch.float16)
|
| 42 |
+
avg_diff_2nd_1 = avg_diff_2nd[1].to(torch.float16)
|
| 43 |
y_concept_1, y_concept_2 = slider_y[0], slider_y[1]
|
| 44 |
end_time = time.time()
|
| 45 |
print(f"direction time: {end_time - start_time:.2f} ms")
|
| 46 |
start_time = time.time()
|
| 47 |
+
image = clip_slider.generate(prompt, scale=0, scale_2nd=0, seed=seed, num_inference_steps=steps, avg_diff=(avg_diff_0,avg_diff_1), avg_diff_2nd=(avg_diff_2nd_0,avg_diff_2nd_1))
|
| 48 |
end_time = time.time()
|
| 49 |
print(f"generation time: {end_time - start_time:.2f} ms")
|
| 50 |
comma_concepts_x = ', '.join(slider_x)
|
| 51 |
comma_concepts_y = ', '.join(slider_y)
|
| 52 |
|
| 53 |
+
avg_diff_x_1 = avg_diff_0.cpu()
|
| 54 |
+
avg_diff_x_2 = avg_diff_1.cpu()
|
| 55 |
+
avg_diff_y_1 = avg_diff_2nd_0.cpu()
|
| 56 |
+
avg_diff_y_2 = avg_diff_2nd_1.cpu()
|
| 57 |
|
| 58 |
return gr.update(label=comma_concepts_x, interactive=True),gr.update(label=comma_concepts_y, interactive=True), x_concept_1, x_concept_2, y_concept_1, y_concept_2, avg_diff_x_1, avg_diff_x_2, avg_diff_y_1, avg_diff_y_2, image
|
| 59 |
|