forked from alisson-anjos/YuE-exllamav2-UI
-
Notifications
You must be signed in to change notification settings - Fork 8
/
Copy pathinterface.py
1129 lines (979 loc) · 47.3 KB
/
interface.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
import json
import os
import subprocess
import threading
import queue
import signal
import gradio as gr
import time
import tempfile
import glob
import sys
import re
import shutil # Added to copy files
from collections import OrderedDict
import psutil
# TODO
# Choose use SDPA or FlashAttention2
# Choose profiles for differents gpus
# Apply transforms patch
# -------------------------------------------------
# If you are using Conda, set these paths accordingly
CONDA_ACTIVATE_PATH = "/opt/conda/etc/profile.d/conda.sh"
CONDA_ENV_NAME = "pyenv"
BASE_REPO_DIR = ""
BASE_YUE_DIR = "src/yue"
BASE_MODELS_DIR = f"workspace/models"
BASE_OUTPUTS_DIR = "workspace/outputs"
BASE_INPUTS_DIR = "workspace/inputs"
# Default Hugging Face models
DEFAULT_STAGE1_MODEL = f"{BASE_MODELS_DIR}\YuE-s1-7B-anneal-en-cot-exl2-8.0bpw"
DEFAULT_STAGE2_MODEL = f"{BASE_MODELS_DIR}\YuE-s2-1B-general-exl2-8.0bpw"
TOKENIZER_MODEL = f"{BASE_YUE_DIR}/mm_tokenizer_v0.2_hf/tokenizer.model"
sys.path.append(os.path.join(f"xcodec_mini_infer"))
sys.path.append(os.path.join(f"xcodec_mini_infer", "descriptaudiocodec"))
# Output directory
os.makedirs(BASE_OUTPUTS_DIR, exist_ok=True)
os.makedirs(BASE_INPUTS_DIR, exist_ok=True)
with open(f"prompt_egs/lyrics.txt", mode='r') as lyrics_example_file:
lyrics_example = lyrics_example_file.read()
with open(f"prompt_egs/genre.txt", mode='r') as genre_example_file:
genre_example = genre_example_file.read() # not populating gradio
# -------------------------------------------------
# Functions to List and Categorize Models
def get_models(model_dir):
"""
Lists all models in the specified directory and categorizes them as Stage1, Stage2, or both.
"""
if not os.path.isdir(model_dir):
return [], [], []
# List directories only
models = [name for name in os.listdir(model_dir) if os.path.isdir(os.path.join(model_dir, name))]
stage1_models = []
stage2_models = []
both_stage_models = []
for model in models:
lower_name = model.lower()
model_path = os.path.join(model_dir, model)
if 's1' in lower_name:
stage1_models.append(model_path)
if 's2' in lower_name:
stage2_models.append(model_path)
if 's1' not in lower_name and 's2' not in lower_name:
both_stage_models.append(model_path)
return stage1_models, stage2_models, both_stage_models
def get_quantization_type(model_path):
"""
Determines the quantization type based on the model's name.
"""
basename = os.path.basename(model_path).lower()
if '-int4' in basename:
return 'int4'
elif '-nf4' in basename:
return 'nf4'
elif '-int8' in basename:
return 'int8'
else:
return 'bf16'
# Model Directory
stage1_models, stage2_models, both_stage_models = get_models(BASE_MODELS_DIR)
stage1_choices = stage1_models + both_stage_models
stage2_choices = stage2_models + both_stage_models
# Queues for logs and audio paths
log_queue = queue.Queue()
audio_path_queue = queue.Queue()
process_dict = {}
process_lock = threading.Lock()
def load_and_process_genres(json_path):
"""
Loads JSON data, processes genres, timbres, genders, moods, and instruments,
removes duplicates (case insensitive), and returns a sorted list of unique values.
"""
# Load JSON data
with open(json_path, 'r', encoding='utf-8') as f:
data = json.load(f)
# Combine all relevant categories into a single list
categories = ['genre', 'timbre', 'gender', 'mood', 'instrument']
all_items = [item.strip() for category in categories for item in data.get(category, [])]
# Use a set for deduplication (case insensitive)
unique_items = OrderedDict()
for item in all_items:
key = item.lower()
if key not in unique_items and item: # Skip empty strings
unique_items[key] = item
# Sort alphabetically while preserving original capitalization
sorted_items = sorted(unique_items.values(), key=lambda x: x.lower())
return sorted_items
js = """
function createLink() {
let baseUrl = window.location.origin;
baseUrl = baseUrl.replace("7860", "8080");
const tagLink = `${baseUrl}/repo/wav_top_200_tags.json`;
document.getElementById("tags_link").href = tagLink;
}
"""
custom_log_box_css = """
#log_box textarea {
overflow-y: scroll;
max-height: 400px; /* Set a max height for the log box */
white-space: pre-wrap; /* Preserve line breaks and white spaces */
border: 1px solid #ccc;
padding: 10px;
font-family: monospace;
scrollbar-width: thin!important;
}"""
def get_selected_file(file_paths):
"""
Handles file selection and prepares it for download.
"""
# Handle the case when file_paths is a string (single file)
if isinstance(file_paths, str):
if os.path.isdir(file_paths):
return None, "Please select a single file and not a folder."
if not os.path.exists(file_paths):
return None, f"File not found: {file_paths}"
if not file_paths.lower().endswith('.mp3'):
return None, f"File is not in .mp3 format: {file_paths}"
return file_paths, f"File '{os.path.basename(file_paths)}' ready for download."
# Handle the case when file_paths is a list (multiple files)
if isinstance(file_paths, list) and file_paths:
file_path = file_paths[0] # Use the first file
if os.path.isdir(file_path):
return None, "Please select a single file and not a folder."
if not os.path.exists(file_path):
return None, f"File not found: {file_path}"
if not file_path.lower().endswith('.mp3'):
return None, f"File is not in .mp3 format: {file_paths}"
return file_path, f"File '{os.path.basename(file_paths)}' ready for download."
return None, "Invalid or no file selected."
def read_subprocess_output(proc, log_queue, audio_path_queue):
"""Reads subprocess stdout line by line, placing them into log_queue and audio_path_queue."""
for line in iter(proc.stdout.readline, b''):
decoded_line = line.decode("utf-8", errors="replace")
print(f"Subprocess output: {decoded_line}") # Debugging
log_queue.put(decoded_line)
# Detect the line containing "Successfully created '"
if "Created mix:" in decoded_line:
# Extract the audio path using regex
match = re.search(r"Created mix:\s*([^']+\.mp3)", decoded_line)
if match:
audio_path = match.group(1)
print(f"Audio path found: {audio_path}") # Debugging
audio_path_queue.put(audio_path)
proc.stdout.close()
proc.wait()
with process_lock:
if proc.pid in process_dict:
del process_dict[proc.pid]
def stop_generation(pid):
"""Send signals to stop the subprocess if running."""
if pid is None:
return "No process is running."
with process_lock:
proc = process_dict.get(pid)
if not proc:
return "No process found or it has already stopped."
if proc.poll() is not None:
return "Process already finished."
try:
print("tryng to stop")
# Windows process tree termination
main_process = psutil.Process(pid)
children = main_process.children(recursive=True)
# Kill sub processes first
for child in children:
child.kill()
# Kill main process
main_process.kill()
# Wait for cleanup
gone, _ = psutil.wait_procs([main_process] + children, timeout=5)
proc.terminate()
proc.kill()
# Send SIGTERM first
os.killpg(os.getpgid(proc.pid), signal.SIGTERM)
time.sleep(2)
if proc.poll() is None:
# If still running, force kill
os.killpg(os.getpgid(proc.pid), signal.SIGKILL)
proc.kill()
with process_lock:
if pid in process_dict:
del process_dict[pid]
return "Inference stopped successfully."
except Exception as e:
return f"Error stopping process: {str(e)}"
def generate_song(
stage1_model,
# stage1_model_quantization,
stage2_model,
# stage2_model_quantization,
tokenizer_model,
genre_txt_path,
lyrics_txt_path,
run_n_segments,
stage2_batch_size,
output_dir,
cuda_idx,
max_new_tokens,
seed,
use_audio_prompt,
audio_prompt_file,
prompt_start_time,
prompt_end_time,
use_dual_tracks_prompt,
vocal_track_prompt_file,
instrumental_track_prompt_file,
disable_offload_model,
keep_intermediate,
custom_filename,
stage1_cache_size,
stage1_cache_mode,
stage2_cache_size,
stage2_cache_mode,
resume_after_n,
extend_mp3,
extend_mp3_end_time,
extend_current_segment
# use_mmgp,
# mmgp_profile,
# use_sdpa,
# use_torch_compile,
# use_transformers_patch
):
"""Spawns infer.py to generate music, capturing logs in real time."""
os.makedirs(output_dir, exist_ok=True)
# If using an audio prompt, copy the file to DEFAULT_INPUT_DIR
if use_audio_prompt and audio_prompt_file is not None:
# Check if audio_prompt_file is a valid path
if isinstance(audio_prompt_file, str):
audio_filename = os.path.basename(audio_prompt_file)
# Replace all special characters with '_' to avoid issues with the command
audio_filename = re.sub(r"[^a-zA-Z0-9.]", "_", audio_filename)
saved_audio_path = os.path.join(BASE_INPUTS_DIR, audio_filename)
shutil.copy(audio_prompt_file, saved_audio_path)
else:
return "Invalid audio prompt file format.", None
else:
saved_audio_path = ""
if use_dual_tracks_prompt and vocal_track_prompt_file is not None and instrumental_track_prompt_file is not None:
if isinstance(vocal_track_prompt_file, str):
vocal_track_filename = os.path.basename(vocal_track_prompt_file)
vocal_track_filename = re.sub(r"[^a-zA-Z0-9.]", "_", vocal_track_filename)
saved_vocal_track_path = os.path.join(BASE_INPUTS_DIR, vocal_track_filename)
shutil.copy(vocal_track_prompt_file, saved_vocal_track_path)
else:
return "Invalid vocal track prompt file format.", None
if isinstance(instrumental_track_prompt_file, str):
instrumental_track_filename = os.path.basename(instrumental_track_prompt_file)
# Replace all special characters with '_' to avoid issues with the command
instrumental_track_filename = re.sub(r"[^a-zA-Z0-9.]", "_", instrumental_track_filename)
saved_instrumental_track_path = os.path.join(BASE_INPUTS_DIR, instrumental_track_filename)
shutil.copy(instrumental_track_prompt_file, saved_instrumental_track_path)
else:
return "Invalid instrumental track file format.", None
else:
saved_vocal_track_path = ""
saved_instrumental_track_path = ""
if extend_mp3 and vocal_track_prompt_file is not None and instrumental_track_prompt_file is not None:
if isinstance(vocal_track_prompt_file, str):
vocal_track_filename = os.path.basename(vocal_track_prompt_file)
vocal_track_filename = re.sub(r"[^a-zA-Z0-9.]", "_", vocal_track_filename)
saved_vocal_track_path = os.path.join(BASE_INPUTS_DIR, vocal_track_filename)
shutil.copy(vocal_track_prompt_file, saved_vocal_track_path)
else:
return "Invalid vocal track prompt file format.", None
if isinstance(instrumental_track_prompt_file, str):
instrumental_track_filename = os.path.basename(instrumental_track_prompt_file)
# Replace all special characters with '_' to avoid issues with the command
instrumental_track_filename = re.sub(r"[^a-zA-Z0-9.]", "_", instrumental_track_filename)
saved_instrumental_track_path = os.path.join(BASE_INPUTS_DIR, instrumental_track_filename)
shutil.copy(instrumental_track_prompt_file, saved_instrumental_track_path)
else:
return "Invalid instrumental track file format.", None
else:
saved_vocal_track_path = ""
saved_instrumental_track_path = ""
# Build base command with '-u' for unbuffered output
cmd = [
"python", "-u", f"{BASE_YUE_DIR}/infer.py", # Added '-u' here
"--stage1_use_exl2",
"--stage1_model", f'"{stage1_model}"',
"--stage1_cache_size", str(stage1_cache_size),
"--stage1_cache_mode", f"{stage1_cache_mode}",
# "--quantization_stage1", f"{stage1_model_quantization}",
"--stage2_use_exl2",
"--stage2_model", f'"{stage2_model}"',
"--stage2_cache_size", str(stage2_cache_size),
"--stage2_cache_mode", f"'{stage2_cache_mode}'",
# "--quantization_stage2", f"{stage2_model_quantization}",
#"--tokenizer", f"'{tokenizer_model}'",
"--genre_txt", f'"{genre_txt_path}"',
"--lyrics_txt", f'"{lyrics_txt_path}"',
"--run_n_segments", str(run_n_segments),
# "--stage2_batch_size", str(stage2_batch_size),
"--output_dir", f"{output_dir}",
"--cuda_idx", str(cuda_idx),
"--seed", str(seed),
"--max_new_tokens", str(max_new_tokens),
"--basic_model_config", f"xcodec_mini_infer/final_ckpt/config.yaml",
"--resume_path", f"xcodec_mini_infer/final_ckpt/ckpt_00360000.pth",
"--config_path", f"xcodec_mini_infer/decoders/config.yaml",
"--vocal_decoder_path", f"xcodec_mini_infer/decoders/decoder_131000.pth",
"--inst_decoder_path", f"xcodec_mini_infer/decoders/decoder_151000.pth"
]
if custom_filename.strip():
cmd.append(f"--custom_filename '{custom_filename}'")
if use_audio_prompt and saved_audio_path:
cmd += [
"--use_audio_prompt",
"--audio_prompt_path", f'"{saved_audio_path}"',
"--prompt_start_time", str(prompt_start_time),
"--prompt_end_time", str(prompt_end_time)
]
if use_dual_tracks_prompt and saved_vocal_track_path and saved_instrumental_track_path:
cmd += [
"--use_dual_tracks_prompt",
"--vocal_track_prompt_path", f'"{saved_vocal_track_path}"',
"--instrumental_track_prompt_path", f'"{saved_instrumental_track_path}"',
"--prompt_start_time", str(prompt_start_time),
"--prompt_end_time", str(prompt_end_time)
]
# resume uploaded mp3
if extend_mp3 and saved_vocal_track_path and saved_instrumental_track_path:
cmd += [
"--extend_mp3",
"--vocal_track_prompt_path", f'"{saved_vocal_track_path}"',
"--instrumental_track_prompt_path", f'"{saved_instrumental_track_path}"',
"--extend_mp3_end_time", str(extend_mp3_end_time),
"--prompt_start_time", str(prompt_start_time),
"--prompt_end_time", str(prompt_end_time),
]
if extend_current_segment:
cmd += [
"--extend_current_segment"
]
# resume previous generation
if resume_after_n:
cmd += [
"--resume_after_n", str(resume_after_n)
]
if disable_offload_model:
cmd.append("--disable_offload_model")
if keep_intermediate:
cmd.append("--keep_intermediate")
# If using conda, wrap the command
if os.path.isfile(CONDA_ACTIVATE_PATH):
prefix_cmd = (
f"bash -c 'source {CONDA_ACTIVATE_PATH} && "
f"conda activate {CONDA_ENV_NAME} && "
)
suffix_cmd = "'"
final_cmd_str = prefix_cmd + " ".join(cmd) + suffix_cmd
else:
#print(cmd)
final_cmd_str = " ".join(cmd)
print(final_cmd_str)
proc = subprocess.Popen(
cmd,
shell=False,
stdout=subprocess.PIPE,
#stderr=subprocess.STDOUT, # is lagging on windows
#preexec_fn=os.setsid # allows signal handling. not in windows
)
with process_lock:
process_dict[proc.pid] = proc
# Thread to read logs
thread = threading.Thread(target=read_subprocess_output, args=(proc, log_queue, audio_path_queue), daemon=True)
thread.start()
return f"Inference started. Outputs will be saved in {output_dir}...", proc.pid
def update_logs(current_logs):
"""Pull all new lines from log_queue and append to current_logs."""
new_text = ""
while not log_queue.empty():
new_text += log_queue.get()
return current_logs + new_text
def build_gradio_interface():
theme = gr.themes.Base()
with gr.Blocks(title="YuE Exllamav2: Open Full-song Generation Foundation Model", theme=theme, css=custom_log_box_css) as demo:
gr.Markdown("# YuE exllamav2 with mp3 extend\nEnter your Genre and Lyrics, then generate & listen!")
with gr.Row():
with gr.Column():
# Replace Textboxes with Dropdowns for Automatic Model Selection
stage1_model = gr.Dropdown(
label="Stage1 Model",
choices=stage1_choices,
value=DEFAULT_STAGE1_MODEL,
info="Select the checkpoint path for the Stage 1 model. (_cot - for prompt2music, _icl - for lyrics+audio2music. If list is empty - make sure that folders inside /workspace/models/ have 's1' and 's2' in their names)",
interactive=True
)
# stage1_model_quantization = gr.Dropdown(
# choices=["bf16", "int8", "int4", "nf4"],
# label="Select the quantization of the Stage1 model",
# value=get_quantization_type(DEFAULT_STAGE1_MODEL),
# interactive=True
# )
stage2_model = gr.Dropdown(
label="Stage2 Model",
choices=stage2_choices,
value=DEFAULT_STAGE2_MODEL,
info="Select the checkpoint path for the Stage 2 model.",
interactive=True
)
# stage2_model_quantization = gr.Dropdown(
# choices=["bf16", "int8", "int4", "nf4"],
# label="Select the quantization of the Stage2 model",
# value=get_quantization_type(DEFAULT_STAGE2_MODEL),
# interactive=True
# )
# TODO: remove the tokenizer model
tokenizer_model = gr.Textbox(
label="Tokenizer Model",
value=TOKENIZER_MODEL,
info="Path to the model tokenizer.",
visible=False
)
# gr.Markdown("#### Optimizations using MMGP (Memory Management for the GPU Poor) by DeepBeepMeep")
# with gr.Row():
# with gr.Column():
# use_mmgp = gr.Checkbox(
# label="Use MMGP? (Only works with original BF16 model, Quantization will be performed based on the chosen profile.)",
# value=False,
# info="If set, Memory Management for GPU Poor by deepbeepmeep will be used."
# )
# gr.Markdown(f"""
# **MMGP Profile:**
# - Profile 1: The fastest but requires 16 GB of VRAM.
# - Profile 3: A bit slower and the model is quantized to 8 bits but requires 12 GB of VRAM.
# - Profile 4: Very slow as this will incur sequencial offloading.
# """)
# mmgp_profile = gr.Dropdown(
# label="MMGP Profile",
# choices=[1, 3, 4],
# value=1,
# visible=False,
# interactive=True
# )
# def toggle_mmgp_profile(checked):
# return gr.update(visible=checked)
# use_mmgp.change(
# fn=toggle_mmgp_profile,
# inputs=use_mmgp,
# outputs=mmgp_profile
# )
# use_transformers_patch = gr.Checkbox(
# label="Use Transformers Patch (optional)(< 10GB of VRAM)?",
# value=False,
# info="If set, the model will use the transformers patch (this patch overwrites two files from the transformers library, It may take a little longer to start generating after clicking the Generate button as a backup of the transformers will be performed or restored)."
# )
# use_sdpa = gr.Checkbox(
# label="Use SDPA? (Can be used with MMGP Profile 4)",
# value=False,
# info="If set, the model will use SDPA instead of FlashAttention2."
# )
# use_torch_compile = gr.Checkbox(
# label="Torch Compile? (Can be used with MMGP Profile 4)",
# value=False,
# info="If set, the model will be compiled using torch compile."
# )
# Dropdowns for genre and lyrics
genres = load_and_process_genres(f"top_200_tags.json")
genre_select = gr.Dropdown(
label="Select Music Genres",
info="Select genre tags that describe the musical style or characteristics (e.g., instrumental, genre, mood, vocal timbre, vocal gender). This is used as part of the generation prompt.",
choices=genres,
value=genres[367],
interactive=True,
multiselect=True,
max_choices=50
)
genre_textarea = gr.Textbox(
visible=False,
)
genre_select.change(
fn=lambda x: " ".join(x),
inputs=genre_select,
outputs=genre_textarea
)
lyrics_textarea = gr.Textbox(
label="Lyrics Text",
lines=2,
placeholder="Type the lyrics here...",
info="Text containing the lyrics for the music generation. (For mp3 extend: put full lyrics here. Exisiting part should be all inside first segment)",
value=lyrics_example
)
run_n_segments = gr.Number(
label="Number of Segments (verses) to generate",
value=2,
precision=0,
info="Set Number of Segments to the number of lyric sections if you want to generate a full song. Additionally, you can increase Stage2 Batch Size based on your available GPU memory."
)
max_new_tokens = gr.Number(
label="Max tokens per segment",
value=2800,
precision=0,
info="The maximum number of tokens per one segment (verse). 3000 tokens = 30 seconds. 2500-3000 are ok."
)
disable_offload_model = gr.Checkbox(
label="Disable Offload Model?",
value=False,
visible=False,
info="If set, the model will not be offloaded from the GPU to CPU after Stage 1 inference."
)
keep_intermediate = gr.Checkbox(
label="Keep Intermediate Files?",
value=False,
visible=False,
info="If set, intermediate outputs will be saved during processing."
)
with gr.Column():
gr.Markdown(f"""
Extend mp3 (main way).
- Set stage 1 model to s1_en_cot
- Set checkbox "Extend mp3".
- Split your mp3 into vocal.mp3 + instrumental.mp3. To split use: [python-audio-separator](https://huggingface.co/spaces/theneos/audio-separator) or [audiostrip.com](https://www.audiostrip.com/isolate) or [lalal.ai](https://www.lalal.ai/) or [vocalremover.org](https://vocalremover.org/) Upload vocal.mp3 + instrumental.mp3 into 2 file fields.
- Find exact time, when first verse ends in your mp3, e.g. 15, put into "Seconds to take from mp3".
Extend mp3 (alternative way):
- Set stage 1 model to s1_en_icl
- Set checkboxes "Extend mp3" and "Use dual tracks audio prompt"
- Upload vocal.mp3 + instrumental.mp3
- Put end of verse 1 into "Seconds from mp3"; start of verse 2 into "Audio prompt End Time"
""")
extend_mp3 = gr.Checkbox(
label="Extend mp3?",
value=False,
info="If set, the model will extend uploaded mp3"
)
extend_mp3_end_time = gr.Number(
label="Seconds to take from mp3",
value=20,
info="10-30s are the best (Set right after verse 1 end)"
)
extend_current_segment = gr.Checkbox(
label="Extend current 0 segment (verse)?",
value=False,
info="If set - no new [verse] tag is added."
)
vocal_track_prompt_file = gr.File(
label="Vocal Track File",
file_types=["audio"],
visible=False,
file_count="single", # Ensure that only one file is uploaded
)
instrumental_track_prompt_file = gr.File(
label="Instrumental Track File",
file_types=["audio"],
visible=False,
file_count="single", # Ensure that only one file is uploaded
)
prompt_start_time = gr.Number(
label="Audio prompt Start Time (s)",
value=0,
visible=False,
info="The start time in seconds to extract the audio prompt from the given audio file."
)
prompt_end_time = gr.Number(
label="Audio prompt End Time (s)",
value=30,
visible=False,
info="The end time in seconds to extract the audio prompt from the given audio file."
)
def toggle_extend_mp3(checked):
# If the checkbox is checked, set elements to visible, otherwise hide them
visibility = gr.update(visible=checked)
return [visibility, visibility, visibility, visibility]
extend_mp3.change(
fn=toggle_extend_mp3,
inputs=extend_mp3,
outputs=[
vocal_track_prompt_file,
instrumental_track_prompt_file,
extend_mp3_end_time,
extend_current_segment,
]
)
extend_current_segment.change(
fn=toggle_extend_mp3,
inputs=extend_mp3,
outputs=[
vocal_track_prompt_file,
instrumental_track_prompt_file,
extend_mp3_end_time,
extend_current_segment,
]
)
use_dual_tracks_prompt = gr.Checkbox(
label="Use Dual Tracks Audio Prompt? (for remix)",
value=False,
info="Optional. If set, the model will use an dual tracks files as a prompt during generation (can also work with extend_mp3)."
)
use_audio_prompt = gr.Checkbox(
label="Use Audio Prompt? (both vocal and instrumental, for remix only)",
value=False,
info="Optional. If set, the model will use an audio file as a prompt during generation (can also work with extend_mp3)."
)
audio_prompt_file = gr.File(
label="Upload Audio Prompt",
file_types=["audio"],
visible=False,
file_count="single", # Ensure that only one file is uploaded
)
def toggle_audio_prompt(checked):
return [
gr.update(visible=checked),
gr.update(visible=checked),
gr.update(visible=checked)
]
use_audio_prompt.change(
fn=toggle_audio_prompt,
inputs=use_audio_prompt,
outputs=[audio_prompt_file, prompt_start_time, prompt_end_time]
)
def toggle_use_dual_tracks_prompt(checked):
# If the checkbox is checked, set elements to visible, otherwise hide them
visibility = gr.update(visible=checked)
return [visibility, visibility, visibility, visibility]
use_dual_tracks_prompt.change(
fn=toggle_use_dual_tracks_prompt,
inputs=use_dual_tracks_prompt,
outputs=[
vocal_track_prompt_file,
instrumental_track_prompt_file,
prompt_start_time,
prompt_end_time,
]
)
seed = gr.Number(
label="Seed",
value=42,
precision=0,
visible=False,
info="Seed for random number generation (currently not working for exllama)."
)
resume_after_n = gr.Number(
label="Continue previous generation? (put id of a segment)",
value=-1,
info="If set, the model will resume previous generation after selected verse (-1: don't resume; 0: resume after first verse). Don't change other settings when using it."
)
custom_filename = gr.Textbox(
label="Custom Save Filename (Optional)(Do not add .mp3 in the name)",
value="",
visible=False,
info="Custom filename for the generated output."
)
stage1_cache_size = gr.Number(
label="Stage1 Cache Size",
value=16384,
precision=0,
info="The cache size used in Stage 1 inference. (set 5000 for 8GB, 8192 for 12GB, 32768 for 24GB VRAM)"
)
stage1_cache_mode = gr.Dropdown(
label="Stage1 Cache Mode",
choices=["FP16", "Q8", "Q6", "Q4"],
value="FP16",
visible=False,
interactive=True,
info="The cache mode used in Stage 1 inference (FP16, Q8, Q6, Q4). Quantized k/v cache will save VRAM at the cost of some speed and precision."
)
stage2_cache_size = gr.Number(
label="Stage2 Cache Size",
value=16384,
precision=0,
info="The cache size used in Stage 2 inference (8192 for 6GB, 16384 for 8GB, 100000 for 24GB)."
)
stage2_cache_mode = gr.Dropdown(
label="Stage2 Cache Mode",
choices=["FP16", "Q8", "Q6", "Q4"],
value="FP16",
interactive=True,
visible=False,
info="The cache mode used in Stage 2 inference (FP16, Q8, Q6, Q4). Quantized k/v cache will save VRAM at the cost of some speed and precision."
)
stage2_batch_size = gr.Number(
label="Stage2 Batch Size",
value=4,
precision=0,
info="The batch size used in Stage 2 inference.",
visible=False
)
output_dir = gr.Textbox(
label="Output Directory",
value=BASE_OUTPUTS_DIR,
visible=False,
info="The directory where generated outputs will be saved."
)
cuda_idx = gr.Number(
label="CUDA Index (0:main)",
value=0,
visible=True,
precision=0
)
generate_button = gr.Button("Generate Music")
stop_button = gr.Button("Stop", visible=False)
gr.Markdown(f"""
**Tips:**
1. `genres` should include details like instruments, genre, mood, vocal timbre, and vocal gender.
2. The length of `lyrics` segments and the `--max_new_tokens` value should be matched. For example, if `--max_new_tokens` is set to 3000, the maximum duration for a segment is around 30 seconds. Ensure your lyrics fit this time frame.
3. If using audio prompt，the duration around 30s will be fine.
**Notice:**
1. A suitable [Genre] tag consists of five components: genre, instrument, mood, gender, and timbre. All five should be included if possible, separated by spaces. The values of timbre should include "vocal" (e.g., "bright vocal").
2. The order of the tags is flexible. For example, a stable genre control string might look like: "[Genre] inspiring female uplifting pop airy vocal electronic bright vocal vocal."
3. Additionally, we have introduced the "Mandarin" and "Cantonese" tags to distinguish between Mandarin and Cantonese, as their lyrics often share similarities.
""")
gr.Markdown(f"""
If you want to use music in-context-learning (provide a reference song), enable `Use Audio Prompt?` and provide `Audio File`, `Prompt Start Time (s)`, and `Prompt Start Time (s)` to specify the audio segment.
Note:
- ICL requires a different ckpt, e.g. `m-a-p/YuE-s1-7B-anneal-en-icl`.
- Music ICL generally requires a 30s audio segment. The model will write new songs with similar style of the provided audio, and may improve musicality.
- Dual-track ICL works better in general, requiring both vocal and instrumental tracks.
- For single-track ICL, you can provide a mix, vocal, or instrumental track.
""")
# with gr.Row():
# total_steps = gr.State(0)
# steps_per_epoch = gr.State(0)
# current_epoch_display = gr.Textbox(label="Epoch Progress", interactive=False, value="Epoch: N/A")
# current_step_display = gr.Textbox(label="Step Progress", interactive=False, value="Step: N/A")
log_box = gr.Textbox(
label="Logs",
value="",
lines=20,
max_lines=30,
interactive=False,
elem_id="log_box"
)
# workaround for the issue of the file explorer not updating
def update_file_explorer():
return gr.FileExplorer(root_dir=BASE_REPO_DIR)
def update_file_explorer_2():
return gr.FileExplorer(root_dir=BASE_OUTPUTS_DIR)
explorer = gr.FileExplorer(root_dir=BASE_OUTPUTS_DIR, interactive=True, label="File Explorer", file_count="single", elem_id="file_explorer", glob="**/*.mp3", every=1)
update_button = gr.Button("Refresh File Explorer")
update_button.click(update_file_explorer, outputs=explorer).then(update_file_explorer_2, outputs=explorer)
with gr.Column():
gr.Markdown("### Select a single file from the file explorer for download.")
audio_status = gr.Textbox(label="File Status", interactive=False)
# Section to show audio and allow download
audio_player = gr.Audio(
label="Generated Audio",
type="filepath",
value=None,
interactive=False
)
# Event: When a file is selected in the explorer
explorer.change(
fn=get_selected_file,
inputs=[explorer],
outputs=[audio_player, audio_status],
)
# Hidden states
generation_pid = gr.State(None)
current_audio_path = gr.State(None)
# Adding Callbacks to Update Quantization Based on Selected Model
# stage1_model.change(
# fn=lambda model_path: get_quantization_type(model_path),
# inputs=stage1_model,
# outputs=stage1_model_quantization
# )
# stage2_model.change(
# fn=lambda model_path: get_quantization_type(model_path),
# inputs=stage2_model,
# outputs=stage2_model_quantization
# )
def on_generate_click(
stage1_model,
# stage1_model_quantization,
stage2_model,
# stage2_model_quantization,
tokenizer_model,
genre_text,
lyrics_text,
run_n_segments,
stage2_batch_size,
output_dir,
cuda_idx,
max_new_tokens,
seed,
use_audio_prompt,
audio_prompt_file,
prompt_start_time,
prompt_end_time,
use_dual_tracks_prompt,
vocal_track_prompt_file,
instrumental_track_prompt_file,
disable_offload_model,
keep_intermediate,
custom_filename,
stage1_cache_size,
stage1_cache_mode,
stage2_cache_size,
stage2_cache_mode,
resume_after_n,
extend_mp3,
extend_mp3_end_time,
extend_current_segment,
# use_mmgp,
# mmgp_profile,
# use_sdpa,
# use_torch_compile,
# use_transformers_patch
):
"""Triggered when user clicks 'Generate Music'."""
# Check if a process is already running
with process_lock:
if process_dict:
return (
"Another process is running. Please stop it before starting a new one.",
None,
gr.update(visible=True),
gr.update(visible=False),
)
# Writes genre_text and lyrics_text to temporary .txt files
def write_temp_file(content, suffix=".txt"):
fd, path = tempfile.mkstemp(suffix=suffix)
with os.fdopen(fd, "w", encoding="utf-8") as f:
f.write(content)
return path
genre_tmp_path = write_temp_file(genre_text, ".txt")
lyrics_tmp_path = write_temp_file(lyrics_text, ".txt")
msg, pid = generate_song(
stage1_model,
# stage1_model_quantization,
stage2_model,
# stage2_model_quantization,
tokenizer_model,
genre_tmp_path,
lyrics_tmp_path,
run_n_segments,
stage2_batch_size,
output_dir,
cuda_idx,
max_new_tokens,
seed,
use_audio_prompt,
audio_prompt_file,
prompt_start_time,
prompt_end_time,
use_dual_tracks_prompt,
vocal_track_prompt_file,
instrumental_track_prompt_file,
disable_offload_model,
keep_intermediate,
custom_filename,
stage1_cache_size,
stage1_cache_mode,
stage2_cache_size,
stage2_cache_mode,
resume_after_n,
extend_mp3,
extend_mp3_end_time,
extend_current_segment,