Spaces:
Running
on
Zero
Running
on
Zero
PommesPeter
commited on
Commit
•
3c34a7f
1
Parent(s):
f18c66a
Update app.py
Browse files
app.py
CHANGED
@@ -241,21 +241,11 @@ def infer_ode(args, infer_args, text_encoder, tokenizer, vae, model):
|
|
241 |
)
|
242 |
# end sampler
|
243 |
|
|
|
244 |
resolution = resolution.split(" ")[-1]
|
245 |
w, h = resolution.split("x")
|
246 |
w, h = int(w), int(h)
|
247 |
|
248 |
-
res_cat = (w * h) ** 0.5
|
249 |
-
seq_len = res_cat // 16
|
250 |
-
|
251 |
-
scaling_method = "ntk"
|
252 |
-
train_seq_len = 64
|
253 |
-
if scaling_method == "ntk":
|
254 |
-
scale_factor = seq_len / train_seq_len
|
255 |
-
else:
|
256 |
-
raise NotImplementedError
|
257 |
-
|
258 |
-
print(f"> scale factor: {scale_factor}")
|
259 |
|
260 |
latent_w, latent_h = w // 8, h // 8
|
261 |
if int(seed) != 0:
|
@@ -284,9 +274,18 @@ def infer_ode(args, infer_args, text_encoder, tokenizer, vae, model):
|
|
284 |
cap_feats=cap_feats,
|
285 |
cap_mask=cap_mask,
|
286 |
cfg_scale=cfg_scale,
|
287 |
-
scale_factor=scale_factor,
|
288 |
)
|
289 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
290 |
print("> start sample")
|
291 |
samples = sample_fn(z, model.forward_with_cfg, **model_kwargs)[-1]
|
292 |
samples = samples[:1]
|
@@ -511,9 +510,9 @@ def main():
|
|
511 |
)
|
512 |
with gr.Row():
|
513 |
scale_methods = gr.Dropdown(
|
514 |
-
value="
|
515 |
-
choices=["
|
516 |
-
label="
|
517 |
)
|
518 |
proportional_attn = gr.Checkbox(
|
519 |
value=True,
|
|
|
241 |
)
|
242 |
# end sampler
|
243 |
|
244 |
+
do_extrapolation = "Extrapolation" in resolution
|
245 |
resolution = resolution.split(" ")[-1]
|
246 |
w, h = resolution.split("x")
|
247 |
w, h = int(w), int(h)
|
248 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
249 |
|
250 |
latent_w, latent_h = w // 8, h // 8
|
251 |
if int(seed) != 0:
|
|
|
274 |
cap_feats=cap_feats,
|
275 |
cap_mask=cap_mask,
|
276 |
cfg_scale=cfg_scale,
|
|
|
277 |
)
|
278 |
|
279 |
+
if proportional_attn:
|
280 |
+
model_kwargs["proportional_attn"] = True
|
281 |
+
model_kwargs["base_seqlen"] = (train_args.image_size // 16) ** 2
|
282 |
+
if do_extrapolation and scaling_method == "Time-aware":
|
283 |
+
model_kwargs["scale_factor"] = math.sqrt(w * h / train_args.image_size ** 2)
|
284 |
+
else:
|
285 |
+
model_kwargs["scale_factor"] = 1.0
|
286 |
+
|
287 |
+
print(f"> scale factor: {model_kwargs["scale_factor"]}")
|
288 |
+
|
289 |
print("> start sample")
|
290 |
samples = sample_fn(z, model.forward_with_cfg, **model_kwargs)[-1]
|
291 |
samples = samples[:1]
|
|
|
510 |
)
|
511 |
with gr.Row():
|
512 |
scale_methods = gr.Dropdown(
|
513 |
+
value="Time-aware",
|
514 |
+
choices=["Time-aware", "None"],
|
515 |
+
label="Rope scaling method",
|
516 |
)
|
517 |
proportional_attn = gr.Checkbox(
|
518 |
value=True,
|