Spaces:
Running
Running
add more limit
Browse files
app.py
CHANGED
@@ -64,9 +64,10 @@ class BackendStatus:
|
|
64 |
self.status = "completed"
|
65 |
self.progress = 100
|
66 |
self.end_time = time.time()
|
67 |
-
self.history.append(
|
68 |
-
|
69 |
-
|
|
|
70 |
|
71 |
def fail(self):
|
72 |
self.status = "failed"
|
@@ -93,10 +94,8 @@ class SessionManager:
|
|
93 |
with cls._lock:
|
94 |
to_remove = []
|
95 |
for session_id, manager in cls._instances.items():
|
96 |
-
if (
|
97 |
-
|
98 |
-
and current_time - manager.last_activity > max_age
|
99 |
-
):
|
100 |
to_remove.append(session_id)
|
101 |
|
102 |
for session_id in to_remove:
|
@@ -106,7 +105,10 @@ class SessionManager:
|
|
106 |
class GenerationManager:
|
107 |
|
108 |
def __init__(self):
|
109 |
-
self.backend_statuses = {
|
|
|
|
|
|
|
110 |
self.last_activity = time.time()
|
111 |
self.request_timestamps = [] # Track timestamps of requests
|
112 |
|
@@ -116,7 +118,8 @@ class GenerationManager:
|
|
116 |
def add_request_timestamp(self):
|
117 |
self.request_timestamps.append(time.time())
|
118 |
|
119 |
-
def has_exceeded_limit(self,
|
|
|
120 |
current_time = time.time()
|
121 |
# Filter timestamps to only include those within the last hour
|
122 |
self.request_timestamps = [
|
@@ -144,17 +147,14 @@ class GenerationManager:
|
|
144 |
text=[f"{avg_duration:.2f}s"], # Show time in seconds
|
145 |
textposition="auto",
|
146 |
width=[0.5], # Make bars narrower
|
147 |
-
)
|
148 |
-
)
|
149 |
|
150 |
# Set a minimum y-axis range if we have data
|
151 |
if has_data:
|
152 |
-
max_duration = max(
|
153 |
-
[
|
154 |
-
|
155 |
-
|
156 |
-
]
|
157 |
-
)
|
158 |
# Add 20% padding to the top
|
159 |
y_max = max_duration * 1.2
|
160 |
# Ensure the y-axis always starts at 0
|
@@ -209,15 +209,19 @@ class GenerationManager:
|
|
209 |
|
210 |
# Use aiohttp instead of requests for async
|
211 |
async with aiohttp.ClientSession() as session:
|
212 |
-
async with session.post(url, headers=headers,
|
|
|
213 |
if response.status == 200:
|
214 |
result = await response.json()
|
215 |
request_id = result["data"]["id"]
|
216 |
-
print(
|
|
|
|
|
217 |
return request_id
|
218 |
else:
|
219 |
text = await response.text()
|
220 |
-
raise Exception(
|
|
|
221 |
|
222 |
except Exception as e:
|
223 |
status.fail()
|
@@ -231,6 +235,59 @@ class GenerationManager:
|
|
231 |
return self
|
232 |
|
233 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
234 |
# Helper function to create error images as data URIs
|
235 |
def create_error_image(backend, error_message):
|
236 |
try:
|
@@ -305,9 +362,9 @@ async def poll_once(manager, backend, request_id):
|
|
305 |
# It's base64 data - format it as a data URI if needed
|
306 |
try:
|
307 |
# Format as data URI for Gradio to display directly
|
308 |
-
if isinstance(
|
309 |
-
|
310 |
-
):
|
311 |
# Convert raw base64 to data URI format
|
312 |
return f"data:image/jpeg;base64,{output}"
|
313 |
else:
|
@@ -315,7 +372,8 @@ async def poll_once(manager, backend, request_id):
|
|
315 |
return output
|
316 |
except Exception as e:
|
317 |
print(f"Error processing base64 image: {e}")
|
318 |
-
raise Exception(
|
|
|
319 |
|
320 |
elif current_status == "failed":
|
321 |
manager.backend_statuses[backend].fail()
|
@@ -347,17 +405,13 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
347 |
gr.Markdown("# 🌊 WaveSpeedAI HiDream Arena")
|
348 |
|
349 |
# Add the introduction with link to WaveSpeedAI
|
350 |
-
gr.Markdown(
|
351 |
-
"""
|
352 |
[WaveSpeedAI](https://wavespeed.ai/) is the global pioneer in accelerating AI-powered video and image generation.
|
353 |
Our in-house inference accelerator provides lossless speedup on image & video generation based on our rich inference optimization software stack, including our in-house inference compiler, CUDA kernel libraries and parallel computing libraries.
|
354 |
-
"""
|
355 |
-
|
356 |
-
gr.Markdown(
|
357 |
-
"""
|
358 |
This demo showcases the performance and outputs of leading image generation models, including HiDream and Flux, on our accelerated inference platform.
|
359 |
-
"""
|
360 |
-
)
|
361 |
|
362 |
with gr.Row():
|
363 |
with gr.Column(scale=3):
|
@@ -375,18 +429,20 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
375 |
with gr.Column(scale=1):
|
376 |
generate_btn = gr.Button("Generate", variant="primary")
|
377 |
|
378 |
-
example_dropdown.change(
|
379 |
-
|
380 |
-
|
381 |
|
382 |
# Two status boxes - small (default) and big (during generation)
|
383 |
-
small_status_box = gr.Markdown("Ready to generate images",
|
|
|
384 |
|
385 |
# Big status box in its own row with styling
|
386 |
with gr.Row(elem_id="big-status-row"):
|
387 |
-
big_status_box = gr.Markdown(
|
388 |
-
|
389 |
-
|
|
|
390 |
|
391 |
with gr.Row():
|
392 |
with gr.Column():
|
@@ -399,27 +455,27 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
399 |
performance_plot = gr.Plot(label="Performance Metrics")
|
400 |
|
401 |
with gr.Accordion("Recent Generations (last 16)", open=False):
|
402 |
-
recent_gallery = gr.Gallery(
|
403 |
-
|
404 |
-
|
405 |
|
406 |
def get_recent_gallery_items():
|
407 |
gallery_items = []
|
408 |
for r in reversed(recent_generations):
|
409 |
gallery_items.append((r["flux-dev"], f"FLUX-dev: {r['prompt']}"))
|
410 |
-
gallery_items.append(
|
411 |
-
|
|
|
|
|
412 |
return gallery_items
|
413 |
|
414 |
def update_recent_gallery(prompt, results):
|
415 |
-
recent_generations.append(
|
416 |
-
|
417 |
-
|
418 |
-
|
419 |
-
|
420 |
-
|
421 |
-
}
|
422 |
-
)
|
423 |
if len(recent_generations) > 16:
|
424 |
recent_generations.pop(0)
|
425 |
gallery_items = get_recent_gallery_items()
|
@@ -470,7 +526,30 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
470 |
gr.HTML(f"<style>{css}</style>")
|
471 |
|
472 |
# Update the generation function to use session manager
|
473 |
-
async def generate_all_backends_with_status_boxes(prompt,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
474 |
"""Generate images with big status box during generation"""
|
475 |
# Get or create a session manager
|
476 |
session_id, manager = SessionManager.get_manager(current_session_id)
|
@@ -478,8 +557,7 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
478 |
|
479 |
# Check if the user has exceeded the request limit
|
480 |
if manager.has_exceeded_limit(
|
481 |
-
|
482 |
-
): # Set the limit to 10 requests per hour
|
483 |
error_message = "❌ You have exceeded the limit of 10 requests per hour. Please try again later."
|
484 |
yield (
|
485 |
error_message,
|
@@ -557,7 +635,8 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
557 |
poll_attempt = 0
|
558 |
|
559 |
# Main polling loop
|
560 |
-
while len(completed_backends
|
|
|
561 |
poll_attempt += 1
|
562 |
|
563 |
# Poll each pending backend
|
@@ -569,9 +648,8 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
569 |
# Only do actual API calls every few attempts to reduce load
|
570 |
if poll_attempt % 2 == 0 or backend == "flux-dev":
|
571 |
# Use the session manager instead of global manager
|
572 |
-
result = await poll_once(
|
573 |
-
|
574 |
-
)
|
575 |
if result: # Backend completed
|
576 |
results[backend] = result
|
577 |
completed_backends.add(backend)
|
@@ -585,11 +663,8 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
585 |
results["flux-dev"],
|
586 |
results["hidream-dev"],
|
587 |
results["hidream-full"],
|
588 |
-
(
|
589 |
-
|
590 |
-
if any(completed_backends)
|
591 |
-
else None
|
592 |
-
),
|
593 |
session_id,
|
594 |
None,
|
595 |
)
|
@@ -600,11 +675,9 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
600 |
await asyncio.sleep(0.1)
|
601 |
|
602 |
# Final status
|
603 |
-
final_status = (
|
604 |
-
|
605 |
-
|
606 |
-
else "⚠️ Some generations timed out"
|
607 |
-
)
|
608 |
|
609 |
gallery_update = update_recent_gallery(prompt, results)
|
610 |
|
@@ -641,6 +714,7 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
641 |
# Schedule periodic cleanup of old sessions
|
642 |
def cleanup_task():
|
643 |
SessionManager.cleanup_old_sessions()
|
|
|
644 |
# Schedule the next cleanup
|
645 |
threading.Timer(3600, cleanup_task).start() # Run every hour
|
646 |
|
|
|
64 |
self.status = "completed"
|
65 |
self.progress = 100
|
66 |
self.end_time = time.time()
|
67 |
+
self.history.append({
|
68 |
+
"timestamp": datetime.now(),
|
69 |
+
"duration": self.end_time - self.start_time
|
70 |
+
})
|
71 |
|
72 |
def fail(self):
|
73 |
self.status = "failed"
|
|
|
94 |
with cls._lock:
|
95 |
to_remove = []
|
96 |
for session_id, manager in cls._instances.items():
|
97 |
+
if (hasattr(manager, "last_activity")
|
98 |
+
and current_time - manager.last_activity > max_age):
|
|
|
|
|
99 |
to_remove.append(session_id)
|
100 |
|
101 |
for session_id in to_remove:
|
|
|
105 |
class GenerationManager:
|
106 |
|
107 |
def __init__(self):
|
108 |
+
self.backend_statuses = {
|
109 |
+
backend: BackendStatus()
|
110 |
+
for backend in BACKENDS
|
111 |
+
}
|
112 |
self.last_activity = time.time()
|
113 |
self.request_timestamps = [] # Track timestamps of requests
|
114 |
|
|
|
118 |
def add_request_timestamp(self):
|
119 |
self.request_timestamps.append(time.time())
|
120 |
|
121 |
+
def has_exceeded_limit(self,
|
122 |
+
limit=10): # Default limit: 10 requests per hour
|
123 |
current_time = time.time()
|
124 |
# Filter timestamps to only include those within the last hour
|
125 |
self.request_timestamps = [
|
|
|
147 |
text=[f"{avg_duration:.2f}s"], # Show time in seconds
|
148 |
textposition="auto",
|
149 |
width=[0.5], # Make bars narrower
|
150 |
+
))
|
|
|
151 |
|
152 |
# Set a minimum y-axis range if we have data
|
153 |
if has_data:
|
154 |
+
max_duration = max([
|
155 |
+
max([h["duration"] for h in status.history] or [0])
|
156 |
+
for status in self.backend_statuses.values()
|
157 |
+
])
|
|
|
|
|
158 |
# Add 20% padding to the top
|
159 |
y_max = max_duration * 1.2
|
160 |
# Ensure the y-axis always starts at 0
|
|
|
209 |
|
210 |
# Use aiohttp instead of requests for async
|
211 |
async with aiohttp.ClientSession() as session:
|
212 |
+
async with session.post(url, headers=headers,
|
213 |
+
json=payload) as response:
|
214 |
if response.status == 200:
|
215 |
result = await response.json()
|
216 |
request_id = result["data"]["id"]
|
217 |
+
print(
|
218 |
+
f"Task submitted successfully. Request ID: {request_id}"
|
219 |
+
)
|
220 |
return request_id
|
221 |
else:
|
222 |
text = await response.text()
|
223 |
+
raise Exception(
|
224 |
+
f"API error: {response.status}, {text}")
|
225 |
|
226 |
except Exception as e:
|
227 |
status.fail()
|
|
|
235 |
return self
|
236 |
|
237 |
|
238 |
+
class ClientManager:
|
239 |
+
_instances = {}
|
240 |
+
_lock = threading.Lock()
|
241 |
+
|
242 |
+
@classmethod
|
243 |
+
def get_manager(cls, client_id=None):
|
244 |
+
if not client_id:
|
245 |
+
client_id = str(uuid.uuid4())
|
246 |
+
|
247 |
+
with cls._lock:
|
248 |
+
if client_id not in cls._instances:
|
249 |
+
cls._instances[client_id] = ClientGenerationManager()
|
250 |
+
return cls._instances[client_id]
|
251 |
+
|
252 |
+
@classmethod
|
253 |
+
def cleanup_old_clients(cls, max_age=3600): # 1 hour default
|
254 |
+
current_time = time.time()
|
255 |
+
with cls._lock:
|
256 |
+
to_remove = []
|
257 |
+
for client_id, manager in cls._instances.items():
|
258 |
+
if (hasattr(manager, "last_activity")
|
259 |
+
and current_time - manager.last_activity > max_age):
|
260 |
+
to_remove.append(client_id)
|
261 |
+
|
262 |
+
for client_id in to_remove:
|
263 |
+
del cls._instances[client_id]
|
264 |
+
|
265 |
+
|
266 |
+
class ClientGenerationManager:
|
267 |
+
|
268 |
+
def __init__(self):
|
269 |
+
self.lock = threading.Lock()
|
270 |
+
|
271 |
+
def update_activity(self):
|
272 |
+
with self.lock:
|
273 |
+
self.last_activity = time.time()
|
274 |
+
|
275 |
+
def add_request_timestamp(self):
|
276 |
+
with self.lock:
|
277 |
+
self.request_timestamps.append(time.time())
|
278 |
+
|
279 |
+
def has_exceeded_limit(self,
|
280 |
+
limit=100): # Default limit: 50 requests per hour
|
281 |
+
with self.lock:
|
282 |
+
current_time = time.time()
|
283 |
+
# Filter timestamps to only include those within the last hour
|
284 |
+
self.request_timestamps = [
|
285 |
+
ts for ts in self.request_timestamps
|
286 |
+
if current_time - ts <= 3600
|
287 |
+
]
|
288 |
+
return len(self.request_timestamps) >= limit
|
289 |
+
|
290 |
+
|
291 |
# Helper function to create error images as data URIs
|
292 |
def create_error_image(backend, error_message):
|
293 |
try:
|
|
|
362 |
# It's base64 data - format it as a data URI if needed
|
363 |
try:
|
364 |
# Format as data URI for Gradio to display directly
|
365 |
+
if isinstance(
|
366 |
+
output, str
|
367 |
+
) and not output.startswith("data:image"):
|
368 |
# Convert raw base64 to data URI format
|
369 |
return f"data:image/jpeg;base64,{output}"
|
370 |
else:
|
|
|
372 |
return output
|
373 |
except Exception as e:
|
374 |
print(f"Error processing base64 image: {e}")
|
375 |
+
raise Exception(
|
376 |
+
f"Failed to process base64 image: {str(e)}")
|
377 |
|
378 |
elif current_status == "failed":
|
379 |
manager.backend_statuses[backend].fail()
|
|
|
405 |
gr.Markdown("# 🌊 WaveSpeedAI HiDream Arena")
|
406 |
|
407 |
# Add the introduction with link to WaveSpeedAI
|
408 |
+
gr.Markdown("""
|
|
|
409 |
[WaveSpeedAI](https://wavespeed.ai/) is the global pioneer in accelerating AI-powered video and image generation.
|
410 |
Our in-house inference accelerator provides lossless speedup on image & video generation based on our rich inference optimization software stack, including our in-house inference compiler, CUDA kernel libraries and parallel computing libraries.
|
411 |
+
""")
|
412 |
+
gr.Markdown("""
|
|
|
|
|
413 |
This demo showcases the performance and outputs of leading image generation models, including HiDream and Flux, on our accelerated inference platform.
|
414 |
+
""")
|
|
|
415 |
|
416 |
with gr.Row():
|
417 |
with gr.Column(scale=3):
|
|
|
429 |
with gr.Column(scale=1):
|
430 |
generate_btn = gr.Button("Generate", variant="primary")
|
431 |
|
432 |
+
example_dropdown.change(lambda ex: ex,
|
433 |
+
inputs=[example_dropdown],
|
434 |
+
outputs=[input_text])
|
435 |
|
436 |
# Two status boxes - small (default) and big (during generation)
|
437 |
+
small_status_box = gr.Markdown("Ready to generate images",
|
438 |
+
elem_id="small-status")
|
439 |
|
440 |
# Big status box in its own row with styling
|
441 |
with gr.Row(elem_id="big-status-row"):
|
442 |
+
big_status_box = gr.Markdown("",
|
443 |
+
elem_id="big-status",
|
444 |
+
visible=False,
|
445 |
+
elem_classes="big-status-box")
|
446 |
|
447 |
with gr.Row():
|
448 |
with gr.Column():
|
|
|
455 |
performance_plot = gr.Plot(label="Performance Metrics")
|
456 |
|
457 |
with gr.Accordion("Recent Generations (last 16)", open=False):
|
458 |
+
recent_gallery = gr.Gallery(label="Prompt and Output",
|
459 |
+
columns=3,
|
460 |
+
interactive=False)
|
461 |
|
462 |
def get_recent_gallery_items():
|
463 |
gallery_items = []
|
464 |
for r in reversed(recent_generations):
|
465 |
gallery_items.append((r["flux-dev"], f"FLUX-dev: {r['prompt']}"))
|
466 |
+
gallery_items.append(
|
467 |
+
(r["hidream-dev"], f"HiDream-dev: {r['prompt']}"))
|
468 |
+
gallery_items.append(
|
469 |
+
(r["hidream-full"], f"HiDream-full: {r['prompt']}"))
|
470 |
return gallery_items
|
471 |
|
472 |
def update_recent_gallery(prompt, results):
|
473 |
+
recent_generations.append({
|
474 |
+
"prompt": prompt,
|
475 |
+
"flux-dev": results["flux-dev"],
|
476 |
+
"hidream-dev": results["hidream-dev"],
|
477 |
+
"hidream-full": results["hidream-full"],
|
478 |
+
})
|
|
|
|
|
479 |
if len(recent_generations) > 16:
|
480 |
recent_generations.pop(0)
|
481 |
gallery_items = get_recent_gallery_items()
|
|
|
526 |
gr.HTML(f"<style>{css}</style>")
|
527 |
|
528 |
# Update the generation function to use session manager
|
529 |
+
async def generate_all_backends_with_status_boxes(prompt,
|
530 |
+
current_session_id,
|
531 |
+
request: gr.Request):
|
532 |
+
client_ip = request.client.host
|
533 |
+
print(f"Client IP: {client_ip}")
|
534 |
+
client_generation_manager = ClientManager.get_manager(client_ip)
|
535 |
+
client_generation_manager.update_activity()
|
536 |
+
if client_generation_manager.has_exceeded_limit(limit=100):
|
537 |
+
error_message = "❌ Your network has exceeded the limit of 100 requests per hour. Please try again later."
|
538 |
+
yield (
|
539 |
+
error_message,
|
540 |
+
error_message,
|
541 |
+
gr.update(visible=False),
|
542 |
+
gr.update(visible=True),
|
543 |
+
None,
|
544 |
+
None,
|
545 |
+
None,
|
546 |
+
None,
|
547 |
+
current_session_id, # Return the session ID
|
548 |
+
None,
|
549 |
+
)
|
550 |
+
return
|
551 |
+
|
552 |
+
client_generation_manager.add_request_timestamp()
|
553 |
"""Generate images with big status box during generation"""
|
554 |
# Get or create a session manager
|
555 |
session_id, manager = SessionManager.get_manager(current_session_id)
|
|
|
557 |
|
558 |
# Check if the user has exceeded the request limit
|
559 |
if manager.has_exceeded_limit(
|
560 |
+
limit=10): # Set the limit to 10 requests per hour
|
|
|
561 |
error_message = "❌ You have exceeded the limit of 10 requests per hour. Please try again later."
|
562 |
yield (
|
563 |
error_message,
|
|
|
635 |
poll_attempt = 0
|
636 |
|
637 |
# Main polling loop
|
638 |
+
while len(completed_backends
|
639 |
+
) < 3 and poll_attempt < max_poll_attempts:
|
640 |
poll_attempt += 1
|
641 |
|
642 |
# Poll each pending backend
|
|
|
648 |
# Only do actual API calls every few attempts to reduce load
|
649 |
if poll_attempt % 2 == 0 or backend == "flux-dev":
|
650 |
# Use the session manager instead of global manager
|
651 |
+
result = await poll_once(manager, backend,
|
652 |
+
request_ids[backend])
|
|
|
653 |
if result: # Backend completed
|
654 |
results[backend] = result
|
655 |
completed_backends.add(backend)
|
|
|
663 |
results["flux-dev"],
|
664 |
results["hidream-dev"],
|
665 |
results["hidream-full"],
|
666 |
+
(manager.get_performance_plot()
|
667 |
+
if any(completed_backends) else None),
|
|
|
|
|
|
|
668 |
session_id,
|
669 |
None,
|
670 |
)
|
|
|
675 |
await asyncio.sleep(0.1)
|
676 |
|
677 |
# Final status
|
678 |
+
final_status = ("✅ All generations completed!"
|
679 |
+
if len(completed_backends) == 3 else
|
680 |
+
"⚠️ Some generations timed out")
|
|
|
|
|
681 |
|
682 |
gallery_update = update_recent_gallery(prompt, results)
|
683 |
|
|
|
714 |
# Schedule periodic cleanup of old sessions
|
715 |
def cleanup_task():
|
716 |
SessionManager.cleanup_old_sessions()
|
717 |
+
ClientManager.cleanup_old_clients()
|
718 |
# Schedule the next cleanup
|
719 |
threading.Timer(3600, cleanup_task).start() # Run every hour
|
720 |
|