Spaces:
Running
on
L40S
Running
on
L40S
updated
Browse files- app.py +209 -13
- requirements.txt +2 -1
app.py
CHANGED
@@ -6,12 +6,20 @@ import random
|
|
6 |
import os
|
7 |
import tempfile
|
8 |
import soundfile as sf
|
|
|
9 |
|
10 |
os.environ['ELASTIC_LOG_LEVEL'] = 'DEBUG'
|
11 |
from transformers import AutoProcessor, pipeline
|
12 |
from elastic_models.transformers import MusicgenForConditionalGeneration
|
13 |
|
14 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
def set_seed(seed: int = 42):
|
16 |
random.seed(seed)
|
17 |
np.random.seed(seed)
|
@@ -23,7 +31,6 @@ def set_seed(seed: int = 42):
|
|
23 |
|
24 |
|
25 |
def cleanup_gpu():
|
26 |
-
"""Clean up GPU memory to avoid TensorRT conflicts."""
|
27 |
if torch.cuda.is_available():
|
28 |
torch.cuda.empty_cache()
|
29 |
torch.cuda.synchronize()
|
@@ -31,7 +38,6 @@ def cleanup_gpu():
|
|
31 |
|
32 |
|
33 |
def cleanup_temp_files():
|
34 |
-
"""Clean up old temporary audio files."""
|
35 |
import glob
|
36 |
import time
|
37 |
temp_dir = tempfile.gettempdir()
|
@@ -47,6 +53,8 @@ def cleanup_temp_files():
|
|
47 |
|
48 |
_generator = None
|
49 |
_processor = None
|
|
|
|
|
50 |
|
51 |
|
52 |
def load_model():
|
@@ -88,6 +96,44 @@ def load_model():
|
|
88 |
return _generator, _processor
|
89 |
|
90 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
91 |
def calculate_max_tokens(duration_seconds):
|
92 |
token_rate = 50
|
93 |
max_new_tokens = int(duration_seconds * token_rate)
|
@@ -160,9 +206,9 @@ def generate_music(text_prompt, duration=10, guidance_scale=3.0):
|
|
160 |
|
161 |
max_val = np.max(np.abs(audio_data))
|
162 |
if max_val > 0:
|
163 |
-
audio_data = audio_data / max_val * 0.95
|
164 |
|
165 |
-
audio_data = (audio_data * 32767).astype(np.int16)
|
166 |
|
167 |
print(f"[GENERATION] Final audio shape: {audio_data.shape}")
|
168 |
print(f"[GENERATION] Audio range: [{np.min(audio_data)}, {np.max(audio_data)}]")
|
@@ -180,6 +226,7 @@ def generate_music(text_prompt, duration=10, guidance_scale=3.0):
|
|
180 |
print(f"[GENERATION] Audio saved to: {temp_path}")
|
181 |
print(f"[GENERATION] File size: {file_size} bytes")
|
182 |
|
|
|
183 |
print(f"[GENERATION] Returning numpy tuple: ({sample_rate}, audio_array)")
|
184 |
return (sample_rate, audio_data)
|
185 |
else:
|
@@ -192,9 +239,150 @@ def generate_music(text_prompt, duration=10, guidance_scale=3.0):
|
|
192 |
return None
|
193 |
|
194 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
195 |
with gr.Blocks(title="MusicGen Large - Music Generation") as demo:
|
196 |
gr.Markdown("# π΅ MusicGen Large Music Generator")
|
197 |
-
gr.Markdown("Generate music from text descriptions using Facebook's MusicGen Large model
|
198 |
|
199 |
with gr.Row():
|
200 |
with gr.Column():
|
@@ -204,7 +392,7 @@ with gr.Blocks(title="MusicGen Large - Music Generation") as demo:
|
|
204 |
lines=3,
|
205 |
value="A groovy funk bassline with a tight drum beat"
|
206 |
)
|
207 |
-
|
208 |
with gr.Row():
|
209 |
duration = gr.Slider(
|
210 |
minimum=5,
|
@@ -222,13 +410,18 @@ with gr.Blocks(title="MusicGen Large - Music Generation") as demo:
|
|
222 |
info="Higher values follow prompt more closely"
|
223 |
)
|
224 |
|
225 |
-
generate_btn = gr.Button("π΅ Generate Music", variant="primary", size="lg")
|
226 |
|
227 |
with gr.Column():
|
228 |
-
|
229 |
-
|
230 |
-
|
231 |
-
|
|
|
|
|
|
|
|
|
|
|
232 |
|
233 |
with gr.Accordion("Tips", open=False):
|
234 |
gr.Markdown("""
|
@@ -238,10 +431,13 @@ with gr.Blocks(title="MusicGen Large - Music Generation") as demo:
|
|
238 |
- Duration is limited to 30 seconds for faster generation
|
239 |
""")
|
240 |
|
|
|
|
|
|
|
241 |
generate_btn.click(
|
242 |
-
fn=
|
243 |
inputs=[text_input, duration, guidance_scale],
|
244 |
-
outputs=
|
245 |
show_progress=True
|
246 |
)
|
247 |
|
|
|
6 |
import os
|
7 |
import tempfile
|
8 |
import soundfile as sf
|
9 |
+
import time
|
10 |
|
11 |
os.environ['ELASTIC_LOG_LEVEL'] = 'DEBUG'
|
12 |
from transformers import AutoProcessor, pipeline
|
13 |
from elastic_models.transformers import MusicgenForConditionalGeneration
|
14 |
|
15 |
|
16 |
+
MODEL_CONFIG = {
|
17 |
+
'cost_per_hour': 1.8, # $1.8 per hour
|
18 |
+
}
|
19 |
+
|
20 |
+
original_time_cache = {}
|
21 |
+
|
22 |
+
|
23 |
def set_seed(seed: int = 42):
|
24 |
random.seed(seed)
|
25 |
np.random.seed(seed)
|
|
|
31 |
|
32 |
|
33 |
def cleanup_gpu():
|
|
|
34 |
if torch.cuda.is_available():
|
35 |
torch.cuda.empty_cache()
|
36 |
torch.cuda.synchronize()
|
|
|
38 |
|
39 |
|
40 |
def cleanup_temp_files():
|
|
|
41 |
import glob
|
42 |
import time
|
43 |
temp_dir = tempfile.gettempdir()
|
|
|
53 |
|
54 |
_generator = None
|
55 |
_processor = None
|
56 |
+
_original_generator = None
|
57 |
+
_original_processor = None
|
58 |
|
59 |
|
60 |
def load_model():
|
|
|
96 |
return _generator, _processor
|
97 |
|
98 |
|
99 |
+
def load_original_model():
|
100 |
+
global _original_generator, _original_processor
|
101 |
+
|
102 |
+
if _original_generator is None:
|
103 |
+
print("[ORIGINAL MODEL] Starting original model initialization...")
|
104 |
+
cleanup_gpu()
|
105 |
+
|
106 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
107 |
+
print(f"[ORIGINAL MODEL] Using device: {device}")
|
108 |
+
|
109 |
+
print("[ORIGINAL MODEL] Loading processor...")
|
110 |
+
_original_processor = AutoProcessor.from_pretrained(
|
111 |
+
"facebook/musicgen-large"
|
112 |
+
)
|
113 |
+
from transformers import MusicgenForConditionalGeneration as HFMusicgenForConditionalGeneration
|
114 |
+
|
115 |
+
print("[ORIGINAL MODEL] Loading original model...")
|
116 |
+
model = HFMusicgenForConditionalGeneration.from_pretrained(
|
117 |
+
"facebook/musicgen-large",
|
118 |
+
torch_dtype=torch.float16,
|
119 |
+
device=device,
|
120 |
+
)
|
121 |
+
|
122 |
+
model.eval()
|
123 |
+
|
124 |
+
print("[ORIGINAL MODEL] Creating pipeline...")
|
125 |
+
_original_generator = pipeline(
|
126 |
+
task="text-to-audio",
|
127 |
+
model=model,
|
128 |
+
tokenizer=_original_processor.tokenizer,
|
129 |
+
device=device,
|
130 |
+
)
|
131 |
+
|
132 |
+
print("[ORIGINAL MODEL] Original model initialization completed successfully")
|
133 |
+
|
134 |
+
return _original_generator, _original_processor
|
135 |
+
|
136 |
+
|
137 |
def calculate_max_tokens(duration_seconds):
|
138 |
token_rate = 50
|
139 |
max_new_tokens = int(duration_seconds * token_rate)
|
|
|
206 |
|
207 |
max_val = np.max(np.abs(audio_data))
|
208 |
if max_val > 0:
|
209 |
+
audio_data = audio_data / max_val * 0.95
|
210 |
|
211 |
+
audio_data = (audio_data * 32767).astype(np.int16)
|
212 |
|
213 |
print(f"[GENERATION] Final audio shape: {audio_data.shape}")
|
214 |
print(f"[GENERATION] Audio range: [{np.min(audio_data)}, {np.max(audio_data)}]")
|
|
|
226 |
print(f"[GENERATION] Audio saved to: {temp_path}")
|
227 |
print(f"[GENERATION] File size: {file_size} bytes")
|
228 |
|
229 |
+
# Try returning numpy format instead
|
230 |
print(f"[GENERATION] Returning numpy tuple: ({sample_rate}, audio_array)")
|
231 |
return (sample_rate, audio_data)
|
232 |
else:
|
|
|
239 |
return None
|
240 |
|
241 |
|
242 |
+
def calculate_generation_cost(generation_time_seconds, mode='S'):
|
243 |
+
hours = generation_time_seconds / 3600
|
244 |
+
cost_per_hour = MODEL_CONFIG['cost_per_hour']
|
245 |
+
return hours * cost_per_hour
|
246 |
+
|
247 |
+
|
248 |
+
def calculate_cost_savings(compressed_time, original_time):
|
249 |
+
compressed_cost = calculate_generation_cost(compressed_time, 'S')
|
250 |
+
original_cost = calculate_generation_cost(original_time, 'original')
|
251 |
+
savings = original_cost - compressed_cost
|
252 |
+
savings_percent = (savings / original_cost * 100) if original_cost > 0 else 0
|
253 |
+
return {
|
254 |
+
'compressed_cost': compressed_cost,
|
255 |
+
'original_cost': original_cost,
|
256 |
+
'savings': savings,
|
257 |
+
'savings_percent': savings_percent
|
258 |
+
}
|
259 |
+
|
260 |
+
|
261 |
+
def get_cache_key(prompt, duration, guidance_scale):
|
262 |
+
return f"{hash(prompt)}_{duration}_{guidance_scale}"
|
263 |
+
|
264 |
+
|
265 |
+
def generate_music_batch(text_prompt, duration=10, guidance_scale=3.0, model_mode="compressed"):
|
266 |
+
try:
|
267 |
+
cache_key = get_cache_key(text_prompt, duration, guidance_scale)
|
268 |
+
|
269 |
+
generator, processor = load_model()
|
270 |
+
model_name = "Compressed (S)"
|
271 |
+
|
272 |
+
print(f"[GENERATION] Starting batch generation using {model_name} model...")
|
273 |
+
print(f"[GENERATION] Prompt: '{text_prompt}'")
|
274 |
+
print(f"[GENERATION] Duration: {duration}s")
|
275 |
+
print(f"[GENERATION] Guidance scale: {guidance_scale}")
|
276 |
+
|
277 |
+
cleanup_gpu()
|
278 |
+
set_seed(42)
|
279 |
+
print(f"[GENERATION] Using seed: 42")
|
280 |
+
|
281 |
+
max_new_tokens = calculate_max_tokens(duration)
|
282 |
+
|
283 |
+
generation_params = {
|
284 |
+
'do_sample': True,
|
285 |
+
'guidance_scale': guidance_scale,
|
286 |
+
'max_new_tokens': max_new_tokens,
|
287 |
+
'min_new_tokens': max_new_tokens,
|
288 |
+
'cache_implementation': 'paged',
|
289 |
+
}
|
290 |
+
|
291 |
+
prompts = [text_prompt] * 4
|
292 |
+
start_time = time.time()
|
293 |
+
outputs = generator(
|
294 |
+
prompts,
|
295 |
+
batch_size=4,
|
296 |
+
generate_kwargs=generation_params
|
297 |
+
)
|
298 |
+
generation_time = time.time() - start_time
|
299 |
+
|
300 |
+
print(f"[GENERATION] Batch generation completed in {generation_time:.2f}s")
|
301 |
+
|
302 |
+
audio_variants = []
|
303 |
+
sample_rate = outputs[0]['sampling_rate']
|
304 |
+
|
305 |
+
for i, output in enumerate(outputs):
|
306 |
+
audio_data = output['audio']
|
307 |
+
|
308 |
+
print(f"[GENERATION] Processing variant {i+1} audio shape: {audio_data.shape}")
|
309 |
+
|
310 |
+
if hasattr(audio_data, 'cpu'):
|
311 |
+
audio_data = audio_data.cpu().numpy()
|
312 |
+
|
313 |
+
if len(audio_data.shape) == 3:
|
314 |
+
audio_data = audio_data[0]
|
315 |
+
|
316 |
+
if len(audio_data.shape) == 2:
|
317 |
+
if audio_data.shape[0] < audio_data.shape[1]:
|
318 |
+
audio_data = audio_data.T
|
319 |
+
if audio_data.shape[1] > 1:
|
320 |
+
audio_data = audio_data[:, 0]
|
321 |
+
else:
|
322 |
+
audio_data = audio_data.flatten()
|
323 |
+
|
324 |
+
audio_data = audio_data.flatten()
|
325 |
+
|
326 |
+
max_val = np.max(np.abs(audio_data))
|
327 |
+
if max_val > 0:
|
328 |
+
audio_data = audio_data / max_val * 0.95
|
329 |
+
|
330 |
+
audio_data = (audio_data * 32767).astype(np.int16)
|
331 |
+
audio_variants.append((sample_rate, audio_data))
|
332 |
+
|
333 |
+
print(f"[GENERATION] Variant {i+1} final shape: {audio_data.shape}")
|
334 |
+
|
335 |
+
comparison_message = ""
|
336 |
+
|
337 |
+
if cache_key in original_time_cache:
|
338 |
+
original_time = original_time_cache[cache_key]
|
339 |
+
cost_info = calculate_cost_savings(generation_time, original_time)
|
340 |
+
|
341 |
+
comparison_message = f"π° Cost Savings: ${cost_info['savings']:.4f} ({cost_info['savings_percent']:.1f}%) - Compressed: ${cost_info['compressed_cost']:.4f} vs Original: ${cost_info['original_cost']:.4f}"
|
342 |
+
print(f"[COST] Savings: ${cost_info['savings']:.4f} ({cost_info['savings_percent']:.1f}%)")
|
343 |
+
else:
|
344 |
+
try:
|
345 |
+
print(f"[TIMING] Measuring original model speed for comparison...")
|
346 |
+
original_generator, original_processor = load_original_model()
|
347 |
+
|
348 |
+
original_start = time.time()
|
349 |
+
original_outputs = original_generator(
|
350 |
+
prompts,
|
351 |
+
batch_size=4,
|
352 |
+
generate_kwargs=generation_params
|
353 |
+
)
|
354 |
+
original_time = time.time() - original_start
|
355 |
+
|
356 |
+
original_time_cache[cache_key] = original_time
|
357 |
+
|
358 |
+
cost_info = calculate_cost_savings(generation_time, original_time)
|
359 |
+
comparison_message = f"π° Cost Savings: ${cost_info['savings']:.4f} ({cost_info['savings_percent']:.1f}%) - Compressed: ${cost_info['compressed_cost']:.4f} vs Original: ${cost_info['original_cost']:.4f}"
|
360 |
+
print(f"[COST] First comparison - Savings: ${cost_info['savings']:.4f} ({cost_info['savings_percent']:.1f}%)")
|
361 |
+
print(f"[TIMING] Original: {original_time:.2f}s, Compressed: {generation_time:.2f}s")
|
362 |
+
|
363 |
+
del original_generator, original_processor
|
364 |
+
cleanup_gpu()
|
365 |
+
print(f"[CLEANUP] Original model cleaned up after timing measurement")
|
366 |
+
|
367 |
+
except Exception as e:
|
368 |
+
print(f"[WARNING] Could not measure original timing: {e}")
|
369 |
+
compressed_cost = calculate_generation_cost(generation_time, 'S')
|
370 |
+
comparison_message = f"πΈ Compressed Cost: ${compressed_cost:.4f} (could not compare with original)"
|
371 |
+
|
372 |
+
generation_info = f"β
Generated 4 variants in {generation_time:.2f}s\n{comparison_message}"
|
373 |
+
|
374 |
+
return audio_variants[0], audio_variants[1], audio_variants[2], audio_variants[3], generation_info
|
375 |
+
|
376 |
+
except Exception as e:
|
377 |
+
print(f"[ERROR] Batch generation failed: {str(e)}")
|
378 |
+
cleanup_gpu()
|
379 |
+
error_msg = f"β Generation failed: {str(e)}"
|
380 |
+
return None, None, None, None, error_msg
|
381 |
+
|
382 |
+
|
383 |
with gr.Blocks(title="MusicGen Large - Music Generation") as demo:
|
384 |
gr.Markdown("# π΅ MusicGen Large Music Generator")
|
385 |
+
gr.Markdown("Generate music from text descriptions using Facebook's MusicGen Large model accelerated by TheStage for 2.3x faster performance")
|
386 |
|
387 |
with gr.Row():
|
388 |
with gr.Column():
|
|
|
392 |
lines=3,
|
393 |
value="A groovy funk bassline with a tight drum beat"
|
394 |
)
|
395 |
+
|
396 |
with gr.Row():
|
397 |
duration = gr.Slider(
|
398 |
minimum=5,
|
|
|
410 |
info="Higher values follow prompt more closely"
|
411 |
)
|
412 |
|
413 |
+
generate_btn = gr.Button("π΅ Generate 4 Music Variants", variant="primary", size="lg")
|
414 |
|
415 |
with gr.Column():
|
416 |
+
generation_info = gr.Markdown("Ready to generate music variants with cost comparison vs original model")
|
417 |
+
|
418 |
+
with gr.Row():
|
419 |
+
audio_output1 = gr.Audio(label="Variant 1", type="numpy")
|
420 |
+
audio_output2 = gr.Audio(label="Variant 2", type="numpy")
|
421 |
+
|
422 |
+
with gr.Row():
|
423 |
+
audio_output3 = gr.Audio(label="Variant 3", type="numpy")
|
424 |
+
audio_output4 = gr.Audio(label="Variant 4", type="numpy")
|
425 |
|
426 |
with gr.Accordion("Tips", open=False):
|
427 |
gr.Markdown("""
|
|
|
431 |
- Duration is limited to 30 seconds for faster generation
|
432 |
""")
|
433 |
|
434 |
+
def generate_simple(text_prompt, duration, guidance_scale):
|
435 |
+
return generate_music_batch(text_prompt, duration, guidance_scale, "compressed")
|
436 |
+
|
437 |
generate_btn.click(
|
438 |
+
fn=generate_simple,
|
439 |
inputs=[text_input, duration, guidance_scale],
|
440 |
+
outputs=[audio_output1, audio_output2, audio_output3, audio_output4, generation_info],
|
441 |
show_progress=True
|
442 |
)
|
443 |
|
requirements.txt
CHANGED
@@ -4,6 +4,7 @@
|
|
4 |
|
5 |
torch
|
6 |
thestage
|
7 |
-
elastic_models[nvidia]
|
8 |
scipy
|
9 |
transformers
|
|
|
|
4 |
|
5 |
torch
|
6 |
thestage
|
7 |
+
# elastic_models[nvidia]
|
8 |
scipy
|
9 |
transformers
|
10 |
+
soundfile
|