Spaces:
Running
Running
Commit
·
d4dcfc5
1
Parent(s):
ffb7037
增强GPU错误处理和添加CPU回退模式,解决ZeroGPU worker error
Browse files
app.py
CHANGED
@@ -105,13 +105,39 @@ import math
|
|
105 |
# 检查是否在Hugging Face Space环境中
|
106 |
IN_HF_SPACE = os.environ.get('SPACE_ID') is not None
|
107 |
|
|
|
|
|
|
|
|
|
|
|
108 |
# 如果在Hugging Face Space中,导入spaces模块
|
109 |
if IN_HF_SPACE:
|
110 |
try:
|
111 |
import spaces
|
112 |
print("在Hugging Face Space环境中运行,已导入spaces模块")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
113 |
except ImportError:
|
114 |
print("未能导入spaces模块,可能不在Hugging Face Space环境中")
|
|
|
115 |
|
116 |
from PIL import Image
|
117 |
from diffusers import AutoencoderKLHunyuanVideo
|
@@ -149,95 +175,194 @@ if not IN_HF_SPACE:
|
|
149 |
else:
|
150 |
# 在Spaces环境中使用默认值
|
151 |
print("在Spaces环境中使用默认内存设置")
|
152 |
-
|
153 |
-
|
154 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
155 |
|
156 |
# 使用models变量存储全局模型引用
|
157 |
models = {}
|
|
|
158 |
|
159 |
# 使用加载模型的函数
|
160 |
def load_models():
|
161 |
-
global models
|
162 |
|
163 |
-
|
164 |
-
|
165 |
-
|
166 |
-
text_encoder = LlamaModel.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='text_encoder', torch_dtype=torch.float16).cpu()
|
167 |
-
text_encoder_2 = CLIPTextModel.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='text_encoder_2', torch_dtype=torch.float16).cpu()
|
168 |
-
tokenizer = LlamaTokenizerFast.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='tokenizer')
|
169 |
-
tokenizer_2 = CLIPTokenizer.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='tokenizer_2')
|
170 |
-
vae = AutoencoderKLHunyuanVideo.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='vae', torch_dtype=torch.float16).cpu()
|
171 |
-
|
172 |
-
feature_extractor = SiglipImageProcessor.from_pretrained("lllyasviel/flux_redux_bfl", subfolder='feature_extractor')
|
173 |
-
image_encoder = SiglipVisionModel.from_pretrained("lllyasviel/flux_redux_bfl", subfolder='image_encoder', torch_dtype=torch.float16).cpu()
|
174 |
-
|
175 |
-
transformer = HunyuanVideoTransformer3DModelPacked.from_pretrained('lllyasviel/FramePackI2V_HY', torch_dtype=torch.bfloat16).cpu()
|
176 |
-
|
177 |
-
vae.eval()
|
178 |
-
text_encoder.eval()
|
179 |
-
text_encoder_2.eval()
|
180 |
-
image_encoder.eval()
|
181 |
-
transformer.eval()
|
182 |
-
|
183 |
-
if not high_vram:
|
184 |
-
vae.enable_slicing()
|
185 |
-
vae.enable_tiling()
|
186 |
-
|
187 |
-
transformer.high_quality_fp32_output_for_inference = True
|
188 |
-
print('transformer.high_quality_fp32_output_for_inference = True')
|
189 |
-
|
190 |
-
transformer.to(dtype=torch.bfloat16)
|
191 |
-
vae.to(dtype=torch.float16)
|
192 |
-
image_encoder.to(dtype=torch.float16)
|
193 |
-
text_encoder.to(dtype=torch.float16)
|
194 |
-
text_encoder_2.to(dtype=torch.float16)
|
195 |
-
|
196 |
-
vae.requires_grad_(False)
|
197 |
-
text_encoder.requires_grad_(False)
|
198 |
-
text_encoder_2.requires_grad_(False)
|
199 |
-
image_encoder.requires_grad_(False)
|
200 |
-
transformer.requires_grad_(False)
|
201 |
-
|
202 |
-
if torch.cuda.is_available():
|
203 |
-
if not high_vram:
|
204 |
-
# DynamicSwapInstaller is same as huggingface's enable_sequential_offload but 3x faster
|
205 |
-
DynamicSwapInstaller.install_model(transformer, device=gpu)
|
206 |
-
DynamicSwapInstaller.install_model(text_encoder, device=gpu)
|
207 |
-
else:
|
208 |
-
text_encoder.to(gpu)
|
209 |
-
text_encoder_2.to(gpu)
|
210 |
-
image_encoder.to(gpu)
|
211 |
-
vae.to(gpu)
|
212 |
-
transformer.to(gpu)
|
213 |
|
214 |
-
|
215 |
-
models = {
|
216 |
-
'text_encoder': text_encoder,
|
217 |
-
'text_encoder_2': text_encoder_2,
|
218 |
-
'tokenizer': tokenizer,
|
219 |
-
'tokenizer_2': tokenizer_2,
|
220 |
-
'vae': vae,
|
221 |
-
'feature_extractor': feature_extractor,
|
222 |
-
'image_encoder': image_encoder,
|
223 |
-
'transformer': transformer
|
224 |
-
}
|
225 |
|
226 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
227 |
|
228 |
|
229 |
# 使用Hugging Face Spaces GPU装饰器
|
230 |
-
if IN_HF_SPACE and 'spaces' in globals():
|
231 |
-
|
232 |
-
|
233 |
-
|
234 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
235 |
|
236 |
|
237 |
# 以下函数内部会延迟获取模型
|
238 |
def get_models():
|
239 |
"""获取模型,如果尚未加载则加载模型"""
|
240 |
-
global models
|
241 |
|
242 |
# 添加模型加载锁,防止并发加载
|
243 |
model_loading_key = "__model_loading__"
|
@@ -248,20 +373,37 @@ def get_models():
|
|
248 |
print("模型正在加载中,等待...")
|
249 |
# 等待模型加载完成
|
250 |
import time
|
|
|
251 |
while not models and model_loading_key in globals():
|
252 |
time.sleep(0.5)
|
253 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
254 |
|
255 |
try:
|
256 |
# 设置加载标记
|
257 |
globals()[model_loading_key] = True
|
258 |
|
259 |
-
if IN_HF_SPACE and 'spaces' in globals():
|
260 |
-
|
261 |
-
|
|
|
|
|
|
|
|
|
|
|
262 |
else:
|
263 |
print("直接加载模型")
|
264 |
-
load_models()
|
|
|
|
|
|
|
|
|
|
|
265 |
finally:
|
266 |
# 无论成功与否,都移除加载标记
|
267 |
if model_loading_key in globals():
|
@@ -275,16 +417,46 @@ stream = AsyncStream()
|
|
275 |
|
276 |
@torch.no_grad()
|
277 |
def worker(input_image, prompt, n_prompt, seed, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, use_teacache):
|
|
|
|
|
|
|
278 |
# 获取模型
|
279 |
-
|
280 |
-
|
281 |
-
|
282 |
-
|
283 |
-
|
284 |
-
|
285 |
-
|
286 |
-
|
287 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
288 |
|
289 |
total_latent_sections = (total_second_length * 30) / (latent_window_size * 4)
|
290 |
total_latent_sections = int(max(round(total_latent_sections), 1))
|
@@ -299,79 +471,136 @@ def worker(input_image, prompt, n_prompt, seed, total_second_length, latent_wind
|
|
299 |
|
300 |
try:
|
301 |
# Clean GPU
|
302 |
-
if not high_vram:
|
303 |
-
|
304 |
-
|
305 |
-
|
|
|
|
|
|
|
|
|
306 |
|
307 |
# Text encoding
|
308 |
-
|
309 |
stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'Text encoding ...'))))
|
310 |
|
311 |
-
|
312 |
-
|
313 |
-
|
|
|
314 |
|
315 |
-
|
316 |
|
317 |
-
|
318 |
-
|
319 |
-
|
320 |
-
|
321 |
|
322 |
-
|
323 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
324 |
|
325 |
# Processing input image
|
326 |
-
|
327 |
stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'Image processing ...'))))
|
328 |
|
329 |
-
|
330 |
-
|
331 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
332 |
|
333 |
-
|
334 |
|
335 |
-
|
336 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
337 |
|
338 |
# VAE encoding
|
339 |
-
|
340 |
stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'VAE encoding ...'))))
|
341 |
|
342 |
-
|
343 |
-
|
|
|
344 |
|
345 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
346 |
|
347 |
# CLIP Vision
|
348 |
-
|
349 |
stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'CLIP Vision encoding ...'))))
|
350 |
|
351 |
-
|
352 |
-
|
|
|
353 |
|
354 |
-
|
355 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
356 |
|
357 |
# Dtype
|
358 |
-
|
359 |
-
|
360 |
-
|
361 |
-
|
362 |
-
|
363 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
364 |
|
365 |
# Sampling
|
366 |
-
|
367 |
stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'Start sampling ...'))))
|
368 |
|
369 |
rnd = torch.Generator("cpu").manual_seed(seed)
|
370 |
num_frames = latent_window_size * 4 - 3
|
371 |
|
372 |
-
|
373 |
-
|
374 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
375 |
|
376 |
latent_paddings = reversed(range(total_latent_sections))
|
377 |
|
@@ -383,6 +612,7 @@ def worker(input_image, prompt, n_prompt, seed, total_second_length, latent_wind
|
|
383 |
latent_paddings = [3] + [2] * (total_latent_sections - 3) + [1, 0]
|
384 |
|
385 |
for latent_padding in latent_paddings:
|
|
|
386 |
is_last_section = latent_padding == 0
|
387 |
latent_padding_size = latent_padding * latent_window_size
|
388 |
|
@@ -401,42 +631,70 @@ def worker(input_image, prompt, n_prompt, seed, total_second_length, latent_wind
|
|
401 |
|
402 |
print(f'latent_padding_size = {latent_padding_size}, is_last_section = {is_last_section}')
|
403 |
|
404 |
-
|
405 |
-
|
406 |
-
|
|
|
407 |
|
408 |
-
|
409 |
-
|
410 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
411 |
|
412 |
-
if not high_vram:
|
413 |
-
|
414 |
-
|
|
|
|
|
|
|
|
|
415 |
|
416 |
-
if use_teacache:
|
417 |
-
|
|
|
|
|
|
|
|
|
|
|
418 |
else:
|
419 |
transformer.initialize_teacache(enable_teacache=False)
|
420 |
|
421 |
def callback(d):
|
422 |
-
|
423 |
-
|
|
|
|
|
|
|
|
|
424 |
|
425 |
-
|
426 |
-
|
427 |
|
428 |
-
|
429 |
-
|
430 |
-
|
431 |
|
432 |
-
|
433 |
-
|
434 |
-
|
435 |
-
|
436 |
-
|
|
|
|
|
|
|
437 |
return
|
438 |
|
439 |
try:
|
|
|
|
|
|
|
440 |
generated_latents = sample_hunyuan(
|
441 |
transformer=transformer,
|
442 |
sampler='unipc',
|
@@ -455,8 +713,8 @@ def worker(input_image, prompt, n_prompt, seed, total_second_length, latent_wind
|
|
455 |
negative_prompt_embeds=llama_vec_n,
|
456 |
negative_prompt_embeds_mask=llama_attention_mask_n,
|
457 |
negative_prompt_poolers=clip_l_pooler_n,
|
458 |
-
device=
|
459 |
-
dtype=
|
460 |
image_embeddings=image_encoder_last_hidden_state,
|
461 |
latent_indices=latent_indices,
|
462 |
clean_latents=clean_latents,
|
@@ -467,6 +725,8 @@ def worker(input_image, prompt, n_prompt, seed, total_second_length, latent_wind
|
|
467 |
clean_latent_4x_indices=clean_latent_4x_indices,
|
468 |
callback=callback,
|
469 |
)
|
|
|
|
|
470 |
except Exception as e:
|
471 |
print(f"采样过程中出错: {e}")
|
472 |
traceback.print_exc()
|
@@ -474,23 +734,57 @@ def worker(input_image, prompt, n_prompt, seed, total_second_length, latent_wind
|
|
474 |
# 如果已经有生成的视频,返回最后生成的视频
|
475 |
if last_output_filename:
|
476 |
stream.output_queue.push(('file', last_output_filename))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
477 |
|
478 |
stream.output_queue.push(('end', None))
|
479 |
return
|
480 |
|
481 |
-
|
482 |
-
|
|
|
483 |
|
484 |
-
|
485 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
486 |
|
487 |
-
if not high_vram:
|
488 |
-
|
489 |
-
|
|
|
|
|
|
|
|
|
490 |
|
491 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
492 |
|
493 |
try:
|
|
|
|
|
|
|
494 |
if history_pixels is None:
|
495 |
history_pixels = vae_decode(real_history_latents, vae).cpu()
|
496 |
else:
|
@@ -500,12 +794,19 @@ def worker(input_image, prompt, n_prompt, seed, total_second_length, latent_wind
|
|
500 |
current_pixels = vae_decode(real_history_latents[:, :, :section_latent_frames], vae).cpu()
|
501 |
history_pixels = soft_append_bcthw(current_pixels, history_pixels, overlapped_frames)
|
502 |
|
503 |
-
|
504 |
-
|
|
|
|
|
|
|
|
|
|
|
505 |
|
506 |
output_filename = os.path.join(outputs_folder, f'{job_id}_{total_generated_latent_frames}.mp4')
|
507 |
|
|
|
508 |
save_bcthw_as_mp4(history_pixels, output_filename, fps=30)
|
|
|
509 |
|
510 |
print(f'Decoded. Current latent shape {real_history_latents.shape}; pixel shape {history_pixels.shape}')
|
511 |
|
@@ -519,6 +820,10 @@ def worker(input_image, prompt, n_prompt, seed, total_second_length, latent_wind
|
|
519 |
if last_output_filename:
|
520 |
stream.output_queue.push(('file', last_output_filename))
|
521 |
|
|
|
|
|
|
|
|
|
522 |
# 尝试继续下一次迭代
|
523 |
continue
|
524 |
|
@@ -528,7 +833,7 @@ def worker(input_image, prompt, n_prompt, seed, total_second_length, latent_wind
|
|
528 |
print(f"处理过程中出现错误: {e}")
|
529 |
traceback.print_exc()
|
530 |
|
531 |
-
if not high_vram:
|
532 |
try:
|
533 |
unload_complete_models(
|
534 |
text_encoder, text_encoder_2, image_encoder, vae, transformer
|
@@ -539,6 +844,10 @@ def worker(input_image, prompt, n_prompt, seed, total_second_length, latent_wind
|
|
539 |
# 如果已经有生成的视频,返回最后生成的视频
|
540 |
if last_output_filename:
|
541 |
stream.output_queue.push(('file', last_output_filename))
|
|
|
|
|
|
|
|
|
542 |
|
543 |
# 确保总是返回end信号
|
544 |
stream.output_queue.push(('end', None))
|
@@ -563,6 +872,7 @@ if IN_HF_SPACE and 'spaces' in globals():
|
|
563 |
|
564 |
output_filename = None
|
565 |
prev_output_filename = None
|
|
|
566 |
|
567 |
# 持续检查worker的输出
|
568 |
while True:
|
@@ -577,13 +887,23 @@ if IN_HF_SPACE and 'spaces' in globals():
|
|
577 |
if flag == 'progress':
|
578 |
preview, desc, html = data
|
579 |
yield gr.update(), gr.update(visible=True, value=preview), desc, html, gr.update(interactive=False), gr.update(interactive=True)
|
|
|
|
|
|
|
|
|
|
|
580 |
|
581 |
if flag == 'end':
|
582 |
# 如果有最后的视频文件,确保返回
|
583 |
if output_filename is None and prev_output_filename is not None:
|
584 |
output_filename = prev_output_filename
|
585 |
-
|
586 |
-
|
|
|
|
|
|
|
|
|
|
|
587 |
break
|
588 |
except Exception as e:
|
589 |
print(f"处理输出时出错: {e}")
|
@@ -594,52 +914,10 @@ if IN_HF_SPACE and 'spaces' in globals():
|
|
594 |
|
595 |
# 如果有部分生成的视频,返回
|
596 |
if prev_output_filename:
|
597 |
-
|
598 |
-
|
599 |
-
<div id="partial-video-container">
|
600 |
-
<div class="msg-en" data-lang="en">Processing error, but partial video has been generated</div>
|
601 |
-
<div class="msg-zh" data-lang="zh">处理过程中出现错误,但已生成部分视频</div>
|
602 |
-
</div>
|
603 |
-
<script>
|
604 |
-
// 根据当前语言显示相应的消息
|
605 |
-
(function() {{
|
606 |
-
const container = document.getElementById('partial-video-container');
|
607 |
-
if (container) {{
|
608 |
-
const currentLang = window.currentLang || 'en'; // 默认英语
|
609 |
-
const msgs = container.querySelectorAll('[data-lang]');
|
610 |
-
msgs.forEach(msg => {{
|
611 |
-
msg.style.display = msg.getAttribute('data-lang') === currentLang ? 'block' : 'none';
|
612 |
-
}});
|
613 |
-
}}
|
614 |
-
}})();
|
615 |
-
</script>
|
616 |
-
"""
|
617 |
-
yield prev_output_filename, gr.update(visible=False), gr.update(), partial_video_msg, gr.update(interactive=True), gr.update(interactive=False)
|
618 |
else:
|
619 |
-
|
620 |
-
error_msg = str(e)
|
621 |
-
en_msg = f"Processing error: {error_msg}"
|
622 |
-
zh_msg = f"处理过程中出现错误: {error_msg}"
|
623 |
-
|
624 |
-
error_html = f"""
|
625 |
-
<div id="error-msg-container">
|
626 |
-
<div class="error-msg-en" data-lang="en">{en_msg}</div>
|
627 |
-
<div class="error-msg-zh" data-lang="zh">{zh_msg}</div>
|
628 |
-
</div>
|
629 |
-
<script>
|
630 |
-
// 根据当前语言显示相应的错误消息
|
631 |
-
(function() {{
|
632 |
-
const errorContainer = document.getElementById('error-msg-container');
|
633 |
-
if (errorContainer) {{
|
634 |
-
const currentLang = window.currentLang || 'en'; // 默认英语
|
635 |
-
const errMsgs = errorContainer.querySelectorAll('[data-lang]');
|
636 |
-
errMsgs.forEach(msg => {{
|
637 |
-
msg.style.display = msg.getAttribute('data-lang') === currentLang ? 'block' : 'none';
|
638 |
-
}});
|
639 |
-
}}
|
640 |
-
}})();
|
641 |
-
</script>
|
642 |
-
"""
|
643 |
yield None, gr.update(visible=False), gr.update(), error_html, gr.update(interactive=True), gr.update(interactive=False)
|
644 |
break
|
645 |
|
@@ -647,47 +925,9 @@ if IN_HF_SPACE and 'spaces' in globals():
|
|
647 |
print(f"启动处理时出错: {e}")
|
648 |
traceback.print_exc()
|
649 |
error_msg = str(e)
|
650 |
-
user_friendly_msg = f'处理过程出错: {error_msg}'
|
651 |
-
|
652 |
-
# 提供更友好的中英文双语错误信息
|
653 |
-
en_msg = ""
|
654 |
-
zh_msg = ""
|
655 |
|
656 |
-
|
657 |
-
|
658 |
-
zh_msg = "网络连接不稳定,模型下载超时。请稍后再试。"
|
659 |
-
elif "GPU内存不足" in error_msg or "CUDA out of memory" in error_msg or "OutOfMemoryError" in error_msg:
|
660 |
-
en_msg = "GPU memory insufficient, please try increasing GPU memory preservation value or reduce video length."
|
661 |
-
zh_msg = "GPU内存不足,请尝试增加GPU推理保留内存值或降低视频长度。"
|
662 |
-
elif "无法加载模型" in error_msg:
|
663 |
-
en_msg = "Failed to load model, possibly due to network issues or high server load. Please try again later."
|
664 |
-
zh_msg = "模型加载失败,可能是网络问题或服务器负载过高。请稍后再试。"
|
665 |
-
else:
|
666 |
-
en_msg = f"Processing error: {error_msg}"
|
667 |
-
zh_msg = f"处理过程出错: {error_msg}"
|
668 |
-
|
669 |
-
# 创建双语错误消息HTML
|
670 |
-
bilingual_error = f"""
|
671 |
-
<div id="error-container">
|
672 |
-
<div class="error-msg-en" data-lang="en">{en_msg}</div>
|
673 |
-
<div class="error-msg-zh" data-lang="zh">{zh_msg}</div>
|
674 |
-
</div>
|
675 |
-
<script>
|
676 |
-
// 根据当前语言显示相应的错误消息
|
677 |
-
(function() {{
|
678 |
-
const errorContainer = document.getElementById('error-container');
|
679 |
-
if (errorContainer) {{
|
680 |
-
const currentLang = window.currentLang || 'en'; // 默认英语
|
681 |
-
const errMsgs = errorContainer.querySelectorAll('[data-lang]');
|
682 |
-
errMsgs.forEach(msg => {{
|
683 |
-
msg.style.display = msg.getAttribute('data-lang') === currentLang ? 'block' : 'none';
|
684 |
-
}});
|
685 |
-
}}
|
686 |
-
}})();
|
687 |
-
</script>
|
688 |
-
"""
|
689 |
-
|
690 |
-
yield None, gr.update(visible=False), gr.update(), bilingual_error, gr.update(interactive=True), gr.update(interactive=False)
|
691 |
|
692 |
process = process_with_gpu
|
693 |
else:
|
@@ -706,6 +946,7 @@ else:
|
|
706 |
|
707 |
output_filename = None
|
708 |
prev_output_filename = None
|
|
|
709 |
|
710 |
# 持续检查worker的输出
|
711 |
while True:
|
@@ -720,13 +961,23 @@ else:
|
|
720 |
if flag == 'progress':
|
721 |
preview, desc, html = data
|
722 |
yield gr.update(), gr.update(visible=True, value=preview), desc, html, gr.update(interactive=False), gr.update(interactive=True)
|
|
|
|
|
|
|
|
|
|
|
723 |
|
724 |
if flag == 'end':
|
725 |
# 如果有最后的视频文件,确保返回
|
726 |
if output_filename is None and prev_output_filename is not None:
|
727 |
output_filename = prev_output_filename
|
728 |
-
|
729 |
-
|
|
|
|
|
|
|
|
|
|
|
730 |
break
|
731 |
except Exception as e:
|
732 |
print(f"处理输出时出错: {e}")
|
@@ -737,74 +988,20 @@ else:
|
|
737 |
|
738 |
# 如果有部分生成的视频,返回
|
739 |
if prev_output_filename:
|
740 |
-
|
741 |
-
|
742 |
-
|
743 |
-
|
744 |
-
|
745 |
-
|
746 |
-
<script>
|
747 |
-
// 根据当前语言显示相应的消息
|
748 |
-
(function() {{
|
749 |
-
const container = document.getElementById('interrupt-container');
|
750 |
-
if (container) {{
|
751 |
-
const currentLang = window.currentLang || 'en'; // 默认英语
|
752 |
-
const msgs = container.querySelectorAll('[data-lang]');
|
753 |
-
msgs.forEach(msg => {{
|
754 |
-
msg.style.display = msg.getAttribute('data-lang') === currentLang ? 'block' : 'none';
|
755 |
-
}});
|
756 |
-
}}
|
757 |
-
}})();
|
758 |
-
</script>
|
759 |
-
"""
|
760 |
-
yield prev_output_filename, gr.update(visible=False), gr.update(), interrupt_msg, gr.update(interactive=True), gr.update(interactive=False)
|
761 |
-
break
|
762 |
|
763 |
except Exception as e:
|
764 |
print(f"启动处理时出错: {e}")
|
765 |
traceback.print_exc()
|
766 |
error_msg = str(e)
|
767 |
-
user_friendly_msg = f'处理过程出错: {error_msg}'
|
768 |
-
|
769 |
-
# 提供更友好的中英文双语错误信息
|
770 |
-
en_msg = ""
|
771 |
-
zh_msg = ""
|
772 |
|
773 |
-
|
774 |
-
|
775 |
-
zh_msg = "网络连接不稳定,模型下载超时。请稍后再试。"
|
776 |
-
elif "GPU内存不足" in error_msg or "CUDA out of memory" in error_msg or "OutOfMemoryError" in error_msg:
|
777 |
-
en_msg = "GPU memory insufficient, please try increasing GPU memory preservation value or reduce video length."
|
778 |
-
zh_msg = "GPU内存不足,请尝试增加GPU推理保留内存值或降低视频长度。"
|
779 |
-
elif "无法加载模型" in error_msg:
|
780 |
-
en_msg = "Failed to load model, possibly due to network issues or high server load. Please try again later."
|
781 |
-
zh_msg = "模型加载失败,可能是网络问题或服务器负载过高。请稍后再试。"
|
782 |
-
else:
|
783 |
-
en_msg = f"Processing error: {error_msg}"
|
784 |
-
zh_msg = f"处理过程出错: {error_msg}"
|
785 |
-
|
786 |
-
# 创建双语错误消息HTML
|
787 |
-
bilingual_error = f"""
|
788 |
-
<div id="error-container">
|
789 |
-
<div class="error-msg-en" data-lang="en">{en_msg}</div>
|
790 |
-
<div class="error-msg-zh" data-lang="zh">{zh_msg}</div>
|
791 |
-
</div>
|
792 |
-
<script>
|
793 |
-
// 根据当前语言显示相应的错误消息
|
794 |
-
(function() {{
|
795 |
-
const errorContainer = document.getElementById('error-container');
|
796 |
-
if (errorContainer) {{
|
797 |
-
const currentLang = window.currentLang || 'en'; // 默认英语
|
798 |
-
const errMsgs = errorContainer.querySelectorAll('[data-lang]');
|
799 |
-
errMsgs.forEach(msg => {{
|
800 |
-
msg.style.display = msg.getAttribute('data-lang') === currentLang ? 'block' : 'none';
|
801 |
-
}});
|
802 |
-
}}
|
803 |
-
}})();
|
804 |
-
</script>
|
805 |
-
"""
|
806 |
-
|
807 |
-
yield None, gr.update(visible=False), gr.update(), bilingual_error, gr.update(interactive=True), gr.update(interactive=False)
|
808 |
|
809 |
|
810 |
def end_process():
|
@@ -1268,4 +1465,58 @@ with block:
|
|
1268 |
end_button.click(fn=end_process)
|
1269 |
|
1270 |
|
1271 |
-
block.launch()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
105 |
# 检查是否在Hugging Face Space环境中
|
106 |
IN_HF_SPACE = os.environ.get('SPACE_ID') is not None
|
107 |
|
108 |
+
# 添加变量跟踪GPU可用性
|
109 |
+
GPU_AVAILABLE = False
|
110 |
+
GPU_INITIALIZED = False
|
111 |
+
last_update_time = time.time()
|
112 |
+
|
113 |
# 如果在Hugging Face Space中,导入spaces模块
|
114 |
if IN_HF_SPACE:
|
115 |
try:
|
116 |
import spaces
|
117 |
print("在Hugging Face Space环境中运行,已导入spaces模块")
|
118 |
+
|
119 |
+
# 检查GPU可用性
|
120 |
+
try:
|
121 |
+
GPU_AVAILABLE = torch.cuda.is_available()
|
122 |
+
print(f"GPU available: {GPU_AVAILABLE}")
|
123 |
+
if GPU_AVAILABLE:
|
124 |
+
print(f"GPU device name: {torch.cuda.get_device_name(0)}")
|
125 |
+
print(f"GPU memory: {torch.cuda.get_device_properties(0).total_memory / 1e9} GB")
|
126 |
+
|
127 |
+
# 尝试进行小型GPU操作,确认GPU实际可用
|
128 |
+
test_tensor = torch.zeros(1, device='cuda')
|
129 |
+
test_tensor = test_tensor + 1
|
130 |
+
del test_tensor
|
131 |
+
print("成功进行GPU测试操作")
|
132 |
+
else:
|
133 |
+
print("警告: CUDA报告可用,但未检测到GPU设备")
|
134 |
+
except Exception as e:
|
135 |
+
GPU_AVAILABLE = False
|
136 |
+
print(f"检查GPU时出错: {e}")
|
137 |
+
print("将使用CPU模式运行")
|
138 |
except ImportError:
|
139 |
print("未能导入spaces模块,可能不在Hugging Face Space环境中")
|
140 |
+
GPU_AVAILABLE = torch.cuda.is_available()
|
141 |
|
142 |
from PIL import Image
|
143 |
from diffusers import AutoencoderKLHunyuanVideo
|
|
|
175 |
else:
|
176 |
# 在Spaces环境中使用默认值
|
177 |
print("在Spaces环境中使用默认内存设置")
|
178 |
+
try:
|
179 |
+
if GPU_AVAILABLE:
|
180 |
+
free_mem_gb = torch.cuda.get_device_properties(0).total_memory / 1e9 * 0.9 # 使用90%的GPU内存
|
181 |
+
high_vram = free_mem_gb > 10 # 更保守的条件
|
182 |
+
else:
|
183 |
+
free_mem_gb = 6.0 # 默认值
|
184 |
+
high_vram = False
|
185 |
+
except Exception as e:
|
186 |
+
print(f"获取GPU内存时出错: {e}")
|
187 |
+
free_mem_gb = 6.0 # 默认值
|
188 |
+
high_vram = False
|
189 |
+
|
190 |
+
print(f'GPU内存: {free_mem_gb:.2f} GB, High-VRAM Mode: {high_vram}')
|
191 |
|
192 |
# 使用models变量存储全局模型引用
|
193 |
models = {}
|
194 |
+
cpu_fallback_mode = not GPU_AVAILABLE # 如果GPU不可用,使用CPU回退模式
|
195 |
|
196 |
# 使用加载模型的函数
|
197 |
def load_models():
|
198 |
+
global models, cpu_fallback_mode, GPU_INITIALIZED
|
199 |
|
200 |
+
if GPU_INITIALIZED:
|
201 |
+
print("模型已加载,跳过重复加载")
|
202 |
+
return models
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
203 |
|
204 |
+
print("开始加载模型...")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
205 |
|
206 |
+
try:
|
207 |
+
# 设置设备,根据GPU可用性确定
|
208 |
+
device = 'cuda' if GPU_AVAILABLE and not cpu_fallback_mode else 'cpu'
|
209 |
+
model_device = 'cpu' # 初始加载到CPU
|
210 |
+
|
211 |
+
# 降低精度以节省内存
|
212 |
+
dtype = torch.float16 if GPU_AVAILABLE else torch.float32
|
213 |
+
transformer_dtype = torch.bfloat16 if GPU_AVAILABLE else torch.float32
|
214 |
+
|
215 |
+
print(f"使用设备: {device}, 模型精度: {dtype}, Transformer精度: {transformer_dtype}")
|
216 |
+
|
217 |
+
# 加载模型
|
218 |
+
try:
|
219 |
+
text_encoder = LlamaModel.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='text_encoder', torch_dtype=dtype).to(model_device)
|
220 |
+
text_encoder_2 = CLIPTextModel.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='text_encoder_2', torch_dtype=dtype).to(model_device)
|
221 |
+
tokenizer = LlamaTokenizerFast.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='tokenizer')
|
222 |
+
tokenizer_2 = CLIPTokenizer.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='tokenizer_2')
|
223 |
+
vae = AutoencoderKLHunyuanVideo.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='vae', torch_dtype=dtype).to(model_device)
|
224 |
+
|
225 |
+
feature_extractor = SiglipImageProcessor.from_pretrained("lllyasviel/flux_redux_bfl", subfolder='feature_extractor')
|
226 |
+
image_encoder = SiglipVisionModel.from_pretrained("lllyasviel/flux_redux_bfl", subfolder='image_encoder', torch_dtype=dtype).to(model_device)
|
227 |
+
|
228 |
+
transformer = HunyuanVideoTransformer3DModelPacked.from_pretrained('lllyasviel/FramePackI2V_HY', torch_dtype=transformer_dtype).to(model_device)
|
229 |
+
|
230 |
+
print("成功加载所有模型")
|
231 |
+
except Exception as e:
|
232 |
+
print(f"加载模型时出错: {e}")
|
233 |
+
print("尝试降低精度重新加载...")
|
234 |
+
|
235 |
+
# 降低精度重试
|
236 |
+
dtype = torch.float32
|
237 |
+
transformer_dtype = torch.float32
|
238 |
+
cpu_fallback_mode = True
|
239 |
+
|
240 |
+
text_encoder = LlamaModel.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='text_encoder', torch_dtype=dtype).to('cpu')
|
241 |
+
text_encoder_2 = CLIPTextModel.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='text_encoder_2', torch_dtype=dtype).to('cpu')
|
242 |
+
tokenizer = LlamaTokenizerFast.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='tokenizer')
|
243 |
+
tokenizer_2 = CLIPTokenizer.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='tokenizer_2')
|
244 |
+
vae = AutoencoderKLHunyuanVideo.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder='vae', torch_dtype=dtype).to('cpu')
|
245 |
+
|
246 |
+
feature_extractor = SiglipImageProcessor.from_pretrained("lllyasviel/flux_redux_bfl", subfolder='feature_extractor')
|
247 |
+
image_encoder = SiglipVisionModel.from_pretrained("lllyasviel/flux_redux_bfl", subfolder='image_encoder', torch_dtype=dtype).to('cpu')
|
248 |
+
|
249 |
+
transformer = HunyuanVideoTransformer3DModelPacked.from_pretrained('lllyasviel/FramePackI2V_HY', torch_dtype=transformer_dtype).to('cpu')
|
250 |
+
|
251 |
+
print("使用CPU模式成功加载所有模型")
|
252 |
+
|
253 |
+
vae.eval()
|
254 |
+
text_encoder.eval()
|
255 |
+
text_encoder_2.eval()
|
256 |
+
image_encoder.eval()
|
257 |
+
transformer.eval()
|
258 |
+
|
259 |
+
if not high_vram or cpu_fallback_mode:
|
260 |
+
vae.enable_slicing()
|
261 |
+
vae.enable_tiling()
|
262 |
+
|
263 |
+
transformer.high_quality_fp32_output_for_inference = True
|
264 |
+
print('transformer.high_quality_fp32_output_for_inference = True')
|
265 |
+
|
266 |
+
# 设置模型精度
|
267 |
+
if not cpu_fallback_mode:
|
268 |
+
transformer.to(dtype=transformer_dtype)
|
269 |
+
vae.to(dtype=dtype)
|
270 |
+
image_encoder.to(dtype=dtype)
|
271 |
+
text_encoder.to(dtype=dtype)
|
272 |
+
text_encoder_2.to(dtype=dtype)
|
273 |
+
|
274 |
+
vae.requires_grad_(False)
|
275 |
+
text_encoder.requires_grad_(False)
|
276 |
+
text_encoder_2.requires_grad_(False)
|
277 |
+
image_encoder.requires_grad_(False)
|
278 |
+
transformer.requires_grad_(False)
|
279 |
+
|
280 |
+
if torch.cuda.is_available() and not cpu_fallback_mode:
|
281 |
+
try:
|
282 |
+
if not high_vram:
|
283 |
+
# DynamicSwapInstaller is same as huggingface's enable_sequential_offload but 3x faster
|
284 |
+
DynamicSwapInstaller.install_model(transformer, device=device)
|
285 |
+
DynamicSwapInstaller.install_model(text_encoder, device=device)
|
286 |
+
else:
|
287 |
+
text_encoder.to(device)
|
288 |
+
text_encoder_2.to(device)
|
289 |
+
image_encoder.to(device)
|
290 |
+
vae.to(device)
|
291 |
+
transformer.to(device)
|
292 |
+
print(f"成功将模型移动到{device}设备")
|
293 |
+
except Exception as e:
|
294 |
+
print(f"移动模型到{device}时出错: {e}")
|
295 |
+
print("回退到CPU模式")
|
296 |
+
cpu_fallback_mode = True
|
297 |
+
|
298 |
+
# 保存到全局变量
|
299 |
+
models = {
|
300 |
+
'text_encoder': text_encoder,
|
301 |
+
'text_encoder_2': text_encoder_2,
|
302 |
+
'tokenizer': tokenizer,
|
303 |
+
'tokenizer_2': tokenizer_2,
|
304 |
+
'vae': vae,
|
305 |
+
'feature_extractor': feature_extractor,
|
306 |
+
'image_encoder': image_encoder,
|
307 |
+
'transformer': transformer
|
308 |
+
}
|
309 |
+
|
310 |
+
GPU_INITIALIZED = True
|
311 |
+
print(f"模型加载完成,运行模式: {'CPU' if cpu_fallback_mode else 'GPU'}")
|
312 |
+
return models
|
313 |
+
except Exception as e:
|
314 |
+
print(f"加载模型过程中发生错误: {e}")
|
315 |
+
traceback.print_exc()
|
316 |
+
|
317 |
+
# 记录更详细的错误信息
|
318 |
+
error_info = {
|
319 |
+
"error": str(e),
|
320 |
+
"traceback": traceback.format_exc(),
|
321 |
+
"cuda_available": torch.cuda.is_available(),
|
322 |
+
"device": "cpu" if cpu_fallback_mode else "cuda",
|
323 |
+
}
|
324 |
+
|
325 |
+
# 保存错误信息到文件,方便排查
|
326 |
+
try:
|
327 |
+
with open(os.path.join(outputs_folder, "error_log.txt"), "w") as f:
|
328 |
+
f.write(str(error_info))
|
329 |
+
except:
|
330 |
+
pass
|
331 |
+
|
332 |
+
# 返回空字典,允许应用继续尝试运行
|
333 |
+
cpu_fallback_mode = True
|
334 |
+
return {}
|
335 |
|
336 |
|
337 |
# 使用Hugging Face Spaces GPU装饰器
|
338 |
+
if IN_HF_SPACE and 'spaces' in globals() and GPU_AVAILABLE:
|
339 |
+
try:
|
340 |
+
@spaces.GPU
|
341 |
+
def initialize_models():
|
342 |
+
"""在@spaces.GPU装饰器内初始化模型"""
|
343 |
+
global GPU_INITIALIZED
|
344 |
+
try:
|
345 |
+
result = load_models()
|
346 |
+
GPU_INITIALIZED = True
|
347 |
+
return result
|
348 |
+
except Exception as e:
|
349 |
+
print(f"使用spaces.GPU初始化模型时出错: {e}")
|
350 |
+
traceback.print_exc()
|
351 |
+
global cpu_fallback_mode
|
352 |
+
cpu_fallback_mode = True
|
353 |
+
# 不使用装饰器再次尝试
|
354 |
+
return load_models()
|
355 |
+
except Exception as e:
|
356 |
+
print(f"创建spaces.GPU装饰器时出错: {e}")
|
357 |
+
# 如果装饰器出错,直接使用非装饰器版本
|
358 |
+
def initialize_models():
|
359 |
+
return load_models()
|
360 |
|
361 |
|
362 |
# 以下函数内部会延迟获取模型
|
363 |
def get_models():
|
364 |
"""获取模型,如果尚未加载则加载模型"""
|
365 |
+
global models, GPU_INITIALIZED
|
366 |
|
367 |
# 添加模型加载锁,防止并发加载
|
368 |
model_loading_key = "__model_loading__"
|
|
|
373 |
print("模型正在加载中,等待...")
|
374 |
# 等待模型加载完成
|
375 |
import time
|
376 |
+
start_wait = time.time()
|
377 |
while not models and model_loading_key in globals():
|
378 |
time.sleep(0.5)
|
379 |
+
# 超过60秒认为加载失败
|
380 |
+
if time.time() - start_wait > 60:
|
381 |
+
print("等待模型加载超时")
|
382 |
+
break
|
383 |
+
|
384 |
+
if models:
|
385 |
+
return models
|
386 |
|
387 |
try:
|
388 |
# 设置加载标记
|
389 |
globals()[model_loading_key] = True
|
390 |
|
391 |
+
if IN_HF_SPACE and 'spaces' in globals() and GPU_AVAILABLE and not cpu_fallback_mode:
|
392 |
+
try:
|
393 |
+
print("使用@spaces.GPU装饰器加载模型")
|
394 |
+
models = initialize_models()
|
395 |
+
except Exception as e:
|
396 |
+
print(f"使用GPU装饰器加载模型失败: {e}")
|
397 |
+
print("尝试直接加载模型")
|
398 |
+
models = load_models()
|
399 |
else:
|
400 |
print("直接加载模型")
|
401 |
+
models = load_models()
|
402 |
+
except Exception as e:
|
403 |
+
print(f"加载模型时发生未预期的错误: {e}")
|
404 |
+
traceback.print_exc()
|
405 |
+
# 确保有一个空字典
|
406 |
+
models = {}
|
407 |
finally:
|
408 |
# 无论成功与否,都移除加载标记
|
409 |
if model_loading_key in globals():
|
|
|
417 |
|
418 |
@torch.no_grad()
|
419 |
def worker(input_image, prompt, n_prompt, seed, total_second_length, latent_window_size, steps, cfg, gs, rs, gpu_memory_preservation, use_teacache):
|
420 |
+
global last_update_time
|
421 |
+
last_update_time = time.time()
|
422 |
+
|
423 |
# 获取模型
|
424 |
+
try:
|
425 |
+
models = get_models()
|
426 |
+
if not models:
|
427 |
+
error_msg = "模型加载失败,请检查日志获取详细信息"
|
428 |
+
print(error_msg)
|
429 |
+
stream.output_queue.push(('error', error_msg))
|
430 |
+
stream.output_queue.push(('end', None))
|
431 |
+
return
|
432 |
+
|
433 |
+
text_encoder = models['text_encoder']
|
434 |
+
text_encoder_2 = models['text_encoder_2']
|
435 |
+
tokenizer = models['tokenizer']
|
436 |
+
tokenizer_2 = models['tokenizer_2']
|
437 |
+
vae = models['vae']
|
438 |
+
feature_extractor = models['feature_extractor']
|
439 |
+
image_encoder = models['image_encoder']
|
440 |
+
transformer = models['transformer']
|
441 |
+
except Exception as e:
|
442 |
+
error_msg = f"获取模型时出错: {e}"
|
443 |
+
print(error_msg)
|
444 |
+
traceback.print_exc()
|
445 |
+
stream.output_queue.push(('error', error_msg))
|
446 |
+
stream.output_queue.push(('end', None))
|
447 |
+
return
|
448 |
+
|
449 |
+
# 确定设备
|
450 |
+
device = 'cuda' if GPU_AVAILABLE and not cpu_fallback_mode else 'cpu'
|
451 |
+
print(f"使用设备: {device} 进行推理")
|
452 |
+
|
453 |
+
# 调整参数以适应CPU模式
|
454 |
+
if cpu_fallback_mode:
|
455 |
+
print("CPU模式下使用更精简的参数")
|
456 |
+
# 减小处理大小以加快CPU处理
|
457 |
+
latent_window_size = min(latent_window_size, 5)
|
458 |
+
steps = min(steps, 15) # 减少步数
|
459 |
+
total_second_length = min(total_second_length, 2.0) # 限制视频长度
|
460 |
|
461 |
total_latent_sections = (total_second_length * 30) / (latent_window_size * 4)
|
462 |
total_latent_sections = int(max(round(total_latent_sections), 1))
|
|
|
471 |
|
472 |
try:
|
473 |
# Clean GPU
|
474 |
+
if not high_vram and not cpu_fallback_mode:
|
475 |
+
try:
|
476 |
+
unload_complete_models(
|
477 |
+
text_encoder, text_encoder_2, image_encoder, vae, transformer
|
478 |
+
)
|
479 |
+
except Exception as e:
|
480 |
+
print(f"卸载模型时出错: {e}")
|
481 |
+
# 继续执行,不中断流程
|
482 |
|
483 |
# Text encoding
|
484 |
+
last_update_time = time.time()
|
485 |
stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'Text encoding ...'))))
|
486 |
|
487 |
+
try:
|
488 |
+
if not high_vram and not cpu_fallback_mode:
|
489 |
+
fake_diffusers_current_device(text_encoder, device)
|
490 |
+
load_model_as_complete(text_encoder_2, target_device=device)
|
491 |
|
492 |
+
llama_vec, clip_l_pooler = encode_prompt_conds(prompt, text_encoder, text_encoder_2, tokenizer, tokenizer_2)
|
493 |
|
494 |
+
if cfg == 1:
|
495 |
+
llama_vec_n, clip_l_pooler_n = torch.zeros_like(llama_vec), torch.zeros_like(clip_l_pooler)
|
496 |
+
else:
|
497 |
+
llama_vec_n, clip_l_pooler_n = encode_prompt_conds(n_prompt, text_encoder, text_encoder_2, tokenizer, tokenizer_2)
|
498 |
|
499 |
+
llama_vec, llama_attention_mask = crop_or_pad_yield_mask(llama_vec, length=512)
|
500 |
+
llama_vec_n, llama_attention_mask_n = crop_or_pad_yield_mask(llama_vec_n, length=512)
|
501 |
+
except Exception as e:
|
502 |
+
error_msg = f"文本编码过程出错: {e}"
|
503 |
+
print(error_msg)
|
504 |
+
traceback.print_exc()
|
505 |
+
stream.output_queue.push(('error', error_msg))
|
506 |
+
stream.output_queue.push(('end', None))
|
507 |
+
return
|
508 |
|
509 |
# Processing input image
|
510 |
+
last_update_time = time.time()
|
511 |
stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'Image processing ...'))))
|
512 |
|
513 |
+
try:
|
514 |
+
H, W, C = input_image.shape
|
515 |
+
height, width = find_nearest_bucket(H, W, resolution=640)
|
516 |
+
|
517 |
+
# 如果是CPU模式,缩小处理尺寸
|
518 |
+
if cpu_fallback_mode:
|
519 |
+
height = min(height, 320)
|
520 |
+
width = min(width, 320)
|
521 |
+
|
522 |
+
input_image_np = resize_and_center_crop(input_image, target_width=width, target_height=height)
|
523 |
|
524 |
+
Image.fromarray(input_image_np).save(os.path.join(outputs_folder, f'{job_id}.png'))
|
525 |
|
526 |
+
input_image_pt = torch.from_numpy(input_image_np).float() / 127.5 - 1
|
527 |
+
input_image_pt = input_image_pt.permute(2, 0, 1)[None, :, None]
|
528 |
+
except Exception as e:
|
529 |
+
error_msg = f"图像处理过程出错: {e}"
|
530 |
+
print(error_msg)
|
531 |
+
traceback.print_exc()
|
532 |
+
stream.output_queue.push(('error', error_msg))
|
533 |
+
stream.output_queue.push(('end', None))
|
534 |
+
return
|
535 |
|
536 |
# VAE encoding
|
537 |
+
last_update_time = time.time()
|
538 |
stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'VAE encoding ...'))))
|
539 |
|
540 |
+
try:
|
541 |
+
if not high_vram and not cpu_fallback_mode:
|
542 |
+
load_model_as_complete(vae, target_device=device)
|
543 |
|
544 |
+
start_latent = vae_encode(input_image_pt, vae)
|
545 |
+
except Exception as e:
|
546 |
+
error_msg = f"VAE编码过程出错: {e}"
|
547 |
+
print(error_msg)
|
548 |
+
traceback.print_exc()
|
549 |
+
stream.output_queue.push(('error', error_msg))
|
550 |
+
stream.output_queue.push(('end', None))
|
551 |
+
return
|
552 |
|
553 |
# CLIP Vision
|
554 |
+
last_update_time = time.time()
|
555 |
stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'CLIP Vision encoding ...'))))
|
556 |
|
557 |
+
try:
|
558 |
+
if not high_vram and not cpu_fallback_mode:
|
559 |
+
load_model_as_complete(image_encoder, target_device=device)
|
560 |
|
561 |
+
image_encoder_output = hf_clip_vision_encode(input_image_np, feature_extractor, image_encoder)
|
562 |
+
image_encoder_last_hidden_state = image_encoder_output.last_hidden_state
|
563 |
+
except Exception as e:
|
564 |
+
error_msg = f"CLIP Vision编码过程出错: {e}"
|
565 |
+
print(error_msg)
|
566 |
+
traceback.print_exc()
|
567 |
+
stream.output_queue.push(('error', error_msg))
|
568 |
+
stream.output_queue.push(('end', None))
|
569 |
+
return
|
570 |
|
571 |
# Dtype
|
572 |
+
try:
|
573 |
+
llama_vec = llama_vec.to(transformer.dtype)
|
574 |
+
llama_vec_n = llama_vec_n.to(transformer.dtype)
|
575 |
+
clip_l_pooler = clip_l_pooler.to(transformer.dtype)
|
576 |
+
clip_l_pooler_n = clip_l_pooler_n.to(transformer.dtype)
|
577 |
+
image_encoder_last_hidden_state = image_encoder_last_hidden_state.to(transformer.dtype)
|
578 |
+
except Exception as e:
|
579 |
+
error_msg = f"数据类型转换出错: {e}"
|
580 |
+
print(error_msg)
|
581 |
+
traceback.print_exc()
|
582 |
+
stream.output_queue.push(('error', error_msg))
|
583 |
+
stream.output_queue.push(('end', None))
|
584 |
+
return
|
585 |
|
586 |
# Sampling
|
587 |
+
last_update_time = time.time()
|
588 |
stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'Start sampling ...'))))
|
589 |
|
590 |
rnd = torch.Generator("cpu").manual_seed(seed)
|
591 |
num_frames = latent_window_size * 4 - 3
|
592 |
|
593 |
+
try:
|
594 |
+
history_latents = torch.zeros(size=(1, 16, 1 + 2 + 16, height // 8, width // 8), dtype=torch.float32).cpu()
|
595 |
+
history_pixels = None
|
596 |
+
total_generated_latent_frames = 0
|
597 |
+
except Exception as e:
|
598 |
+
error_msg = f"初始化历史状态出错: {e}"
|
599 |
+
print(error_msg)
|
600 |
+
traceback.print_exc()
|
601 |
+
stream.output_queue.push(('error', error_msg))
|
602 |
+
stream.output_queue.push(('end', None))
|
603 |
+
return
|
604 |
|
605 |
latent_paddings = reversed(range(total_latent_sections))
|
606 |
|
|
|
612 |
latent_paddings = [3] + [2] * (total_latent_sections - 3) + [1, 0]
|
613 |
|
614 |
for latent_padding in latent_paddings:
|
615 |
+
last_update_time = time.time()
|
616 |
is_last_section = latent_padding == 0
|
617 |
latent_padding_size = latent_padding * latent_window_size
|
618 |
|
|
|
631 |
|
632 |
print(f'latent_padding_size = {latent_padding_size}, is_last_section = {is_last_section}')
|
633 |
|
634 |
+
try:
|
635 |
+
indices = torch.arange(0, sum([1, latent_padding_size, latent_window_size, 1, 2, 16])).unsqueeze(0)
|
636 |
+
clean_latent_indices_pre, blank_indices, latent_indices, clean_latent_indices_post, clean_latent_2x_indices, clean_latent_4x_indices = indices.split([1, latent_padding_size, latent_window_size, 1, 2, 16], dim=1)
|
637 |
+
clean_latent_indices = torch.cat([clean_latent_indices_pre, clean_latent_indices_post], dim=1)
|
638 |
|
639 |
+
clean_latents_pre = start_latent.to(history_latents)
|
640 |
+
clean_latents_post, clean_latents_2x, clean_latents_4x = history_latents[:, :, :1 + 2 + 16, :, :].split([1, 2, 16], dim=2)
|
641 |
+
clean_latents = torch.cat([clean_latents_pre, clean_latents_post], dim=2)
|
642 |
+
except Exception as e:
|
643 |
+
error_msg = f"准备采样数据时出错: {e}"
|
644 |
+
print(error_msg)
|
645 |
+
traceback.print_exc()
|
646 |
+
# 尝试继续下一轮迭代而不是完全终止
|
647 |
+
if last_output_filename:
|
648 |
+
stream.output_queue.push(('file', last_output_filename))
|
649 |
+
continue
|
650 |
|
651 |
+
if not high_vram and not cpu_fallback_mode:
|
652 |
+
try:
|
653 |
+
unload_complete_models()
|
654 |
+
move_model_to_device_with_memory_preservation(transformer, target_device=device, preserved_memory_gb=gpu_memory_preservation)
|
655 |
+
except Exception as e:
|
656 |
+
print(f"移动transformer到GPU时出错: {e}")
|
657 |
+
# 继续执行,可能影响性能但不必终止
|
658 |
|
659 |
+
if use_teacache and not cpu_fallback_mode:
|
660 |
+
try:
|
661 |
+
transformer.initialize_teacache(enable_teacache=True, num_steps=steps)
|
662 |
+
except Exception as e:
|
663 |
+
print(f"初始化teacache时出错: {e}")
|
664 |
+
# 禁用teacache并继续
|
665 |
+
transformer.initialize_teacache(enable_teacache=False)
|
666 |
else:
|
667 |
transformer.initialize_teacache(enable_teacache=False)
|
668 |
|
669 |
def callback(d):
|
670 |
+
global last_update_time
|
671 |
+
last_update_time = time.time()
|
672 |
+
|
673 |
+
try:
|
674 |
+
preview = d['denoised']
|
675 |
+
preview = vae_decode_fake(preview)
|
676 |
|
677 |
+
preview = (preview * 255.0).detach().cpu().numpy().clip(0, 255).astype(np.uint8)
|
678 |
+
preview = einops.rearrange(preview, 'b c t h w -> (b h) (t w) c')
|
679 |
|
680 |
+
if stream.input_queue.top() == 'end':
|
681 |
+
stream.output_queue.push(('end', None))
|
682 |
+
raise KeyboardInterrupt('User ends the task.')
|
683 |
|
684 |
+
current_step = d['i'] + 1
|
685 |
+
percentage = int(100.0 * current_step / steps)
|
686 |
+
hint = f'Sampling {current_step}/{steps}'
|
687 |
+
desc = f'Total generated frames: {int(max(0, total_generated_latent_frames * 4 - 3))}, Video length: {max(0, (total_generated_latent_frames * 4 - 3) / 30) :.2f} seconds (FPS-30). The video is being extended now ...'
|
688 |
+
stream.output_queue.push(('progress', (preview, desc, make_progress_bar_html(percentage, hint))))
|
689 |
+
except Exception as e:
|
690 |
+
print(f"回调函数中出错: {e}")
|
691 |
+
# 不中断采样过程
|
692 |
return
|
693 |
|
694 |
try:
|
695 |
+
sampling_start_time = time.time()
|
696 |
+
print(f"开始采样,设备: {device}, 数据类型: {transformer.dtype}, 使用TeaCache: {use_teacache and not cpu_fallback_mode}")
|
697 |
+
|
698 |
generated_latents = sample_hunyuan(
|
699 |
transformer=transformer,
|
700 |
sampler='unipc',
|
|
|
713 |
negative_prompt_embeds=llama_vec_n,
|
714 |
negative_prompt_embeds_mask=llama_attention_mask_n,
|
715 |
negative_prompt_poolers=clip_l_pooler_n,
|
716 |
+
device=device,
|
717 |
+
dtype=transformer.dtype,
|
718 |
image_embeddings=image_encoder_last_hidden_state,
|
719 |
latent_indices=latent_indices,
|
720 |
clean_latents=clean_latents,
|
|
|
725 |
clean_latent_4x_indices=clean_latent_4x_indices,
|
726 |
callback=callback,
|
727 |
)
|
728 |
+
|
729 |
+
print(f"采样完成,用时: {time.time() - sampling_start_time:.2f}秒")
|
730 |
except Exception as e:
|
731 |
print(f"采样过程中出错: {e}")
|
732 |
traceback.print_exc()
|
|
|
734 |
# 如果已经有生成的视频,返回最后生成的视频
|
735 |
if last_output_filename:
|
736 |
stream.output_queue.push(('file', last_output_filename))
|
737 |
+
|
738 |
+
# 创建错误信息
|
739 |
+
error_msg = f"采样过程中出错,但已返回部分生成的视频: {e}"
|
740 |
+
stream.output_queue.push(('error', error_msg))
|
741 |
+
else:
|
742 |
+
# 如果没有生成的视频,返回错误信息
|
743 |
+
error_msg = f"采样过程中出错,无法生成视频: {e}"
|
744 |
+
stream.output_queue.push(('error', error_msg))
|
745 |
|
746 |
stream.output_queue.push(('end', None))
|
747 |
return
|
748 |
|
749 |
+
try:
|
750 |
+
if is_last_section:
|
751 |
+
generated_latents = torch.cat([start_latent.to(generated_latents), generated_latents], dim=2)
|
752 |
|
753 |
+
total_generated_latent_frames += int(generated_latents.shape[2])
|
754 |
+
history_latents = torch.cat([generated_latents.to(history_latents), history_latents], dim=2)
|
755 |
+
except Exception as e:
|
756 |
+
error_msg = f"处理生成的潜变量时出错: {e}"
|
757 |
+
print(error_msg)
|
758 |
+
traceback.print_exc()
|
759 |
+
|
760 |
+
if last_output_filename:
|
761 |
+
stream.output_queue.push(('file', last_output_filename))
|
762 |
+
stream.output_queue.push(('error', error_msg))
|
763 |
+
stream.output_queue.push(('end', None))
|
764 |
+
return
|
765 |
|
766 |
+
if not high_vram and not cpu_fallback_mode:
|
767 |
+
try:
|
768 |
+
offload_model_from_device_for_memory_preservation(transformer, target_device=device, preserved_memory_gb=8)
|
769 |
+
load_model_as_complete(vae, target_device=device)
|
770 |
+
except Exception as e:
|
771 |
+
print(f"管理模型内存时出错: {e}")
|
772 |
+
# 继续执行
|
773 |
|
774 |
+
try:
|
775 |
+
real_history_latents = history_latents[:, :, :total_generated_latent_frames, :, :]
|
776 |
+
except Exception as e:
|
777 |
+
error_msg = f"处理历史潜变量时出错: {e}"
|
778 |
+
print(error_msg)
|
779 |
+
|
780 |
+
if last_output_filename:
|
781 |
+
stream.output_queue.push(('file', last_output_filename))
|
782 |
+
continue
|
783 |
|
784 |
try:
|
785 |
+
vae_start_time = time.time()
|
786 |
+
print(f"开始VAE解码,潜变量形状: {real_history_latents.shape}")
|
787 |
+
|
788 |
if history_pixels is None:
|
789 |
history_pixels = vae_decode(real_history_latents, vae).cpu()
|
790 |
else:
|
|
|
794 |
current_pixels = vae_decode(real_history_latents[:, :, :section_latent_frames], vae).cpu()
|
795 |
history_pixels = soft_append_bcthw(current_pixels, history_pixels, overlapped_frames)
|
796 |
|
797 |
+
print(f"VAE解码完成,用时: {time.time() - vae_start_time:.2f}秒")
|
798 |
+
|
799 |
+
if not high_vram and not cpu_fallback_mode:
|
800 |
+
try:
|
801 |
+
unload_complete_models()
|
802 |
+
except Exception as e:
|
803 |
+
print(f"卸载模型时出错: {e}")
|
804 |
|
805 |
output_filename = os.path.join(outputs_folder, f'{job_id}_{total_generated_latent_frames}.mp4')
|
806 |
|
807 |
+
save_start_time = time.time()
|
808 |
save_bcthw_as_mp4(history_pixels, output_filename, fps=30)
|
809 |
+
print(f"保存视频完成,用时: {time.time() - save_start_time:.2f}秒")
|
810 |
|
811 |
print(f'Decoded. Current latent shape {real_history_latents.shape}; pixel shape {history_pixels.shape}')
|
812 |
|
|
|
820 |
if last_output_filename:
|
821 |
stream.output_queue.push(('file', last_output_filename))
|
822 |
|
823 |
+
# 记录错误信息
|
824 |
+
error_msg = f"视频解码或保存过程中出错: {e}"
|
825 |
+
stream.output_queue.push(('error', error_msg))
|
826 |
+
|
827 |
# 尝试继续下一次迭代
|
828 |
continue
|
829 |
|
|
|
833 |
print(f"处理过程中出现错误: {e}")
|
834 |
traceback.print_exc()
|
835 |
|
836 |
+
if not high_vram and not cpu_fallback_mode:
|
837 |
try:
|
838 |
unload_complete_models(
|
839 |
text_encoder, text_encoder_2, image_encoder, vae, transformer
|
|
|
844 |
# 如果已经有生成的视频,返回最后生成的视频
|
845 |
if last_output_filename:
|
846 |
stream.output_queue.push(('file', last_output_filename))
|
847 |
+
|
848 |
+
# 返回错误信息
|
849 |
+
error_msg = f"处理过程中出现错误: {e}"
|
850 |
+
stream.output_queue.push(('error', error_msg))
|
851 |
|
852 |
# 确保总是返回end信号
|
853 |
stream.output_queue.push(('end', None))
|
|
|
872 |
|
873 |
output_filename = None
|
874 |
prev_output_filename = None
|
875 |
+
error_message = None
|
876 |
|
877 |
# 持续检查worker的输出
|
878 |
while True:
|
|
|
887 |
if flag == 'progress':
|
888 |
preview, desc, html = data
|
889 |
yield gr.update(), gr.update(visible=True, value=preview), desc, html, gr.update(interactive=False), gr.update(interactive=True)
|
890 |
+
|
891 |
+
if flag == 'error':
|
892 |
+
error_message = data
|
893 |
+
print(f"收到错误消息: {error_message}")
|
894 |
+
# 不立即显示,等待end信号
|
895 |
|
896 |
if flag == 'end':
|
897 |
# 如果有最后的视频文件,确保返回
|
898 |
if output_filename is None and prev_output_filename is not None:
|
899 |
output_filename = prev_output_filename
|
900 |
+
|
901 |
+
# 如果有错误消息,创建友好的错误显示
|
902 |
+
if error_message:
|
903 |
+
error_html = create_error_html(error_message)
|
904 |
+
yield output_filename, gr.update(visible=False), gr.update(), error_html, gr.update(interactive=True), gr.update(interactive=False)
|
905 |
+
else:
|
906 |
+
yield output_filename, gr.update(visible=False), gr.update(), '', gr.update(interactive=True), gr.update(interactive=False)
|
907 |
break
|
908 |
except Exception as e:
|
909 |
print(f"处理输出时出错: {e}")
|
|
|
914 |
|
915 |
# 如果有部分生成的视频,返回
|
916 |
if prev_output_filename:
|
917 |
+
error_html = create_error_html("处理超时,但已生成部分视频", is_timeout=True)
|
918 |
+
yield prev_output_filename, gr.update(visible=False), gr.update(), error_html, gr.update(interactive=True), gr.update(interactive=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
919 |
else:
|
920 |
+
error_html = create_error_html(f"处理超时: {e}", is_timeout=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
921 |
yield None, gr.update(visible=False), gr.update(), error_html, gr.update(interactive=True), gr.update(interactive=False)
|
922 |
break
|
923 |
|
|
|
925 |
print(f"启动处理时出错: {e}")
|
926 |
traceback.print_exc()
|
927 |
error_msg = str(e)
|
|
|
|
|
|
|
|
|
|
|
928 |
|
929 |
+
error_html = create_error_html(error_msg)
|
930 |
+
yield None, gr.update(visible=False), gr.update(), error_html, gr.update(interactive=True), gr.update(interactive=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
931 |
|
932 |
process = process_with_gpu
|
933 |
else:
|
|
|
946 |
|
947 |
output_filename = None
|
948 |
prev_output_filename = None
|
949 |
+
error_message = None
|
950 |
|
951 |
# 持续检查worker的输出
|
952 |
while True:
|
|
|
961 |
if flag == 'progress':
|
962 |
preview, desc, html = data
|
963 |
yield gr.update(), gr.update(visible=True, value=preview), desc, html, gr.update(interactive=False), gr.update(interactive=True)
|
964 |
+
|
965 |
+
if flag == 'error':
|
966 |
+
error_message = data
|
967 |
+
print(f"收到错误消息: {error_message}")
|
968 |
+
# 不立即显示,等待end信号
|
969 |
|
970 |
if flag == 'end':
|
971 |
# 如果有最后的视频文件,确保返回
|
972 |
if output_filename is None and prev_output_filename is not None:
|
973 |
output_filename = prev_output_filename
|
974 |
+
|
975 |
+
# 如果有错误消息,创建友好的错误显示
|
976 |
+
if error_message:
|
977 |
+
error_html = create_error_html(error_message)
|
978 |
+
yield output_filename, gr.update(visible=False), gr.update(), error_html, gr.update(interactive=True), gr.update(interactive=False)
|
979 |
+
else:
|
980 |
+
yield output_filename, gr.update(visible=False), gr.update(), '', gr.update(interactive=True), gr.update(interactive=False)
|
981 |
break
|
982 |
except Exception as e:
|
983 |
print(f"处理输出时出错: {e}")
|
|
|
988 |
|
989 |
# 如果有部分生成的视频,返回
|
990 |
if prev_output_filename:
|
991 |
+
error_html = create_error_html("处理超时,但已生成部分视频", is_timeout=True)
|
992 |
+
yield prev_output_filename, gr.update(visible=False), gr.update(), error_html, gr.update(interactive=True), gr.update(interactive=False)
|
993 |
+
else:
|
994 |
+
error_html = create_error_html(f"处理超时: {e}", is_timeout=True)
|
995 |
+
yield None, gr.update(visible=False), gr.update(), error_html, gr.update(interactive=True), gr.update(interactive=False)
|
996 |
+
break
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
997 |
|
998 |
except Exception as e:
|
999 |
print(f"启动处理时出错: {e}")
|
1000 |
traceback.print_exc()
|
1001 |
error_msg = str(e)
|
|
|
|
|
|
|
|
|
|
|
1002 |
|
1003 |
+
error_html = create_error_html(error_msg)
|
1004 |
+
yield None, gr.update(visible=False), gr.update(), error_html, gr.update(interactive=True), gr.update(interactive=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1005 |
|
1006 |
|
1007 |
def end_process():
|
|
|
1465 |
end_button.click(fn=end_process)
|
1466 |
|
1467 |
|
1468 |
+
block.launch()
|
1469 |
+
|
1470 |
+
# 创建友好的错误显示HTML
|
1471 |
+
def create_error_html(error_msg, is_timeout=False):
|
1472 |
+
"""创建双语错误消息HTML"""
|
1473 |
+
# 提供更友好的中英文双语错误信息
|
1474 |
+
en_msg = ""
|
1475 |
+
zh_msg = ""
|
1476 |
+
|
1477 |
+
if is_timeout:
|
1478 |
+
en_msg = "Processing timed out, but partial video may have been generated" if "部分视频" in error_msg else f"Processing timed out: {error_msg}"
|
1479 |
+
zh_msg = "处理超时,但已生成部分视频" if "部分视频" in error_msg else f"处理超时: {error_msg}"
|
1480 |
+
elif "模型加载失败" in error_msg:
|
1481 |
+
en_msg = "Failed to load models. The Space may be experiencing high traffic or GPU issues."
|
1482 |
+
zh_msg = "模型加载失败,可能是Space流量过高或GPU资源不足。"
|
1483 |
+
elif "GPU" in error_msg or "CUDA" in error_msg or "内存" in error_msg or "memory" in error_msg:
|
1484 |
+
en_msg = "GPU memory insufficient or GPU error. Try increasing GPU memory preservation value or reduce video length."
|
1485 |
+
zh_msg = "GPU内存不足或GPU错误,请尝试增加GPU推理保留内存值或降低视频长度。"
|
1486 |
+
elif "采样过程中出错" in error_msg:
|
1487 |
+
if "部分" in error_msg:
|
1488 |
+
en_msg = "Error during sampling process, but partial video has been generated."
|
1489 |
+
zh_msg = "采样过程中出错,但已生成部分视频。"
|
1490 |
+
else:
|
1491 |
+
en_msg = "Error during sampling process. Unable to generate video."
|
1492 |
+
zh_msg = "采样过程中出错,无法生成视频。"
|
1493 |
+
elif "模型下载超时" in error_msg or "网络连接不稳定" in error_msg or "ReadTimeoutError" in error_msg or "ConnectionError" in error_msg:
|
1494 |
+
en_msg = "Network connection is unstable, model download timed out. Please try again later."
|
1495 |
+
zh_msg = "网络连接不稳定,模型下载超时。请稍后再试。"
|
1496 |
+
elif "VAE" in error_msg or "解码" in error_msg or "decode" in error_msg:
|
1497 |
+
en_msg = "Error during video decoding or saving process. Try again with a different seed."
|
1498 |
+
zh_msg = "视频解码或保存过程中出错,请尝试使用不同的随机种子。"
|
1499 |
+
else:
|
1500 |
+
en_msg = f"Processing error: {error_msg}"
|
1501 |
+
zh_msg = f"处理过程出错: {error_msg}"
|
1502 |
+
|
1503 |
+
# 创建双语错误消息HTML
|
1504 |
+
return f"""
|
1505 |
+
<div id="error-container" class="error-message">
|
1506 |
+
<div class="error-msg-en" data-lang="en">{en_msg}</div>
|
1507 |
+
<div class="error-msg-zh" data-lang="zh">{zh_msg}</div>
|
1508 |
+
</div>
|
1509 |
+
<script>
|
1510 |
+
// 根据当前语言显示相应的错误消息
|
1511 |
+
(function() {{
|
1512 |
+
const errorContainer = document.getElementById('error-container');
|
1513 |
+
if (errorContainer) {{
|
1514 |
+
const currentLang = window.currentLang || 'en'; // 默认英语
|
1515 |
+
const errMsgs = errorContainer.querySelectorAll('[data-lang]');
|
1516 |
+
errMsgs.forEach(msg => {{
|
1517 |
+
msg.style.display = msg.getAttribute('data-lang') === currentLang ? 'block' : 'none';
|
1518 |
+
}});
|
1519 |
+
}}
|
1520 |
+
}})();
|
1521 |
+
</script>
|
1522 |
+
"""
|