Spaces:
Runtime error
Runtime error
torch seems slow, bring back onnx
Browse files- app.py +61 -45
- requirements.txt +1 -3
app.py
CHANGED
@@ -1,54 +1,79 @@
|
|
1 |
import argparse
|
2 |
import glob
|
3 |
-
import
|
4 |
-
import os
|
5 |
import time
|
|
|
6 |
|
7 |
import gradio as gr
|
8 |
import numpy as np
|
9 |
-
import
|
10 |
-
|
11 |
-
import torch.nn.functional as F
|
12 |
import tqdm
|
|
|
|
|
13 |
|
14 |
import MIDI
|
15 |
-
from midi_model import MIDIModel
|
16 |
-
from midi_tokenizer import MIDITokenizer
|
17 |
from midi_synthesizer import synthesis
|
18 |
-
from
|
19 |
|
20 |
MAX_SEED = np.iinfo(np.int32).max
|
21 |
in_space = os.getenv("SYSTEM") == "spaces"
|
22 |
|
23 |
|
24 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
def generate(model, prompt=None, max_len=512, temp=1.0, top_p=0.98, top_k=20,
|
26 |
-
disable_patch_change=False, disable_control_change=False, disable_channels=None,
|
27 |
if disable_channels is not None:
|
28 |
disable_channels = [tokenizer.parameter_ids["channel"][c] for c in disable_channels]
|
29 |
else:
|
30 |
disable_channels = []
|
|
|
|
|
31 |
max_token_seq = tokenizer.max_token_seq
|
32 |
if prompt is None:
|
33 |
-
input_tensor =
|
34 |
input_tensor[0, 0] = tokenizer.bos_id # bos
|
35 |
else:
|
36 |
prompt = prompt[:, :max_token_seq]
|
37 |
if prompt.shape[-1] < max_token_seq:
|
38 |
prompt = np.pad(prompt, ((0, 0), (0, max_token_seq - prompt.shape[-1])),
|
39 |
mode="constant", constant_values=tokenizer.pad_id)
|
40 |
-
input_tensor =
|
41 |
-
input_tensor = input_tensor
|
42 |
cur_len = input_tensor.shape[1]
|
43 |
bar = tqdm.tqdm(desc="generating", total=max_len - cur_len, disable=in_space)
|
44 |
-
with bar
|
45 |
while cur_len < max_len:
|
46 |
end = False
|
47 |
-
hidden = model.
|
48 |
-
next_token_seq =
|
49 |
event_name = ""
|
50 |
for i in range(max_token_seq):
|
51 |
-
mask =
|
52 |
if i == 0:
|
53 |
mask_ids = list(tokenizer.event_ids.values()) + [tokenizer.eos_id]
|
54 |
if disable_patch_change:
|
@@ -62,9 +87,9 @@ def generate(model, prompt=None, max_len=512, temp=1.0, top_p=0.98, top_k=20,
|
|
62 |
if param_name == "channel":
|
63 |
mask_ids = [i for i in mask_ids if i not in disable_channels]
|
64 |
mask[mask_ids] = 1
|
65 |
-
logits = model.
|
66 |
-
scores =
|
67 |
-
sample =
|
68 |
if i == 0:
|
69 |
next_token_seq = sample
|
70 |
eid = sample.item()
|
@@ -73,17 +98,17 @@ def generate(model, prompt=None, max_len=512, temp=1.0, top_p=0.98, top_k=20,
|
|
73 |
break
|
74 |
event_name = tokenizer.id_events[eid]
|
75 |
else:
|
76 |
-
next_token_seq =
|
77 |
if len(tokenizer.events[event_name]) == i:
|
78 |
break
|
79 |
if next_token_seq.shape[1] < max_token_seq:
|
80 |
-
next_token_seq =
|
81 |
-
|
82 |
-
next_token_seq = next_token_seq
|
83 |
-
input_tensor =
|
84 |
cur_len += 1
|
85 |
bar.update(1)
|
86 |
-
yield next_token_seq.reshape(-1)
|
87 |
if end:
|
88 |
break
|
89 |
|
@@ -104,7 +129,7 @@ def run(model_name, tab, instruments, drum_kit, bpm, mid, midi_events, midi_opt,
|
|
104 |
max_len = gen_events
|
105 |
if seed_rand:
|
106 |
seed = np.random.randint(0, MAX_SEED)
|
107 |
-
generator =
|
108 |
disable_patch_change = False
|
109 |
disable_channels = None
|
110 |
if tab == 0:
|
@@ -135,16 +160,14 @@ def run(model_name, tab, instruments, drum_kit, bpm, mid, midi_events, midi_opt,
|
|
135 |
for token_seq in mid:
|
136 |
mid_seq.append(token_seq.tolist())
|
137 |
max_len += len(mid)
|
138 |
-
|
139 |
events = [tokenizer.tokens2event(tokens) for tokens in mid_seq]
|
140 |
init_msgs = [create_msg("visualizer_clear", None), create_msg("visualizer_append", events)]
|
141 |
t = time.time() + 1
|
142 |
yield mid_seq, None, None, seed, send_msgs(init_msgs)
|
143 |
model = models[model_name]
|
144 |
-
amp = device == "cuda"
|
145 |
midi_generator = generate(model, mid, max_len=max_len, temp=temp, top_p=top_p, top_k=top_k,
|
146 |
disable_patch_change=disable_patch_change, disable_control_change=not allow_cc,
|
147 |
-
disable_channels=disable_channels,
|
148 |
events = []
|
149 |
for i, token_seq in enumerate(midi_generator):
|
150 |
token_seq = token_seq.tolist()
|
@@ -222,21 +245,15 @@ if __name__ == "__main__":
|
|
222 |
"j-pop finetune model": ["skytnt/midi-model-ft", "jpop/"],
|
223 |
"touhou finetune model": ["skytnt/midi-model-ft", "touhou/"],
|
224 |
}
|
225 |
-
device = "cuda" if torch.cuda.is_available() else "cpu"
|
226 |
-
if device=="cuda": # flash attn
|
227 |
-
torch.backends.cuda.enable_mem_efficient_sdp(True)
|
228 |
-
torch.backends.cuda.enable_flash_sdp(True)
|
229 |
models = {}
|
230 |
tokenizer = MIDITokenizer()
|
|
|
231 |
for name, (repo_id, path) in models_info.items():
|
232 |
-
|
233 |
-
|
234 |
-
|
235 |
-
|
236 |
-
|
237 |
-
model.load_state_dict(state_dict, strict=False)
|
238 |
-
model.eval()
|
239 |
-
models[name] = model
|
240 |
|
241 |
load_javascript()
|
242 |
app = gr.Blocks()
|
@@ -248,8 +265,7 @@ if __name__ == "__main__":
|
|
248 |
"[Open In Colab]"
|
249 |
"(https://colab.research.google.com/github/SkyTNT/midi-model/blob/main/demo.ipynb)"
|
250 |
" for faster running and longer generation\n\n"
|
251 |
-
"**Update v1.2**: Optimise the tokenizer and dataset
|
252 |
-
f"Device: {device}"
|
253 |
)
|
254 |
js_msg = gr.Textbox(elem_id="msg_receiver", visible=False)
|
255 |
js_msg.change(None, [js_msg], [], js="""
|
@@ -319,4 +335,4 @@ if __name__ == "__main__":
|
|
319 |
[output_midi_seq, output_midi, output_audio, input_seed, js_msg],
|
320 |
concurrency_limit=3)
|
321 |
stop_btn.click(cancel_run, [output_midi_seq], [output_midi, output_audio, js_msg], cancels=run_event, queue=False)
|
322 |
-
app.launch(server_port=opt.port, share=opt.share, inbrowser=True)
|
|
|
1 |
import argparse
|
2 |
import glob
|
3 |
+
import os.path
|
|
|
4 |
import time
|
5 |
+
import uuid
|
6 |
|
7 |
import gradio as gr
|
8 |
import numpy as np
|
9 |
+
import onnxruntime as rt
|
|
|
|
|
10 |
import tqdm
|
11 |
+
import json
|
12 |
+
from huggingface_hub import hf_hub_download
|
13 |
|
14 |
import MIDI
|
|
|
|
|
15 |
from midi_synthesizer import synthesis
|
16 |
+
from midi_tokenizer import MIDITokenizer
|
17 |
|
18 |
MAX_SEED = np.iinfo(np.int32).max
|
19 |
in_space = os.getenv("SYSTEM") == "spaces"
|
20 |
|
21 |
|
22 |
+
def softmax(x, axis):
|
23 |
+
x_max = np.amax(x, axis=axis, keepdims=True)
|
24 |
+
exp_x_shifted = np.exp(x - x_max)
|
25 |
+
return exp_x_shifted / np.sum(exp_x_shifted, axis=axis, keepdims=True)
|
26 |
+
|
27 |
+
|
28 |
+
def sample_top_p_k(probs, p, k, generator=None):
|
29 |
+
if generator is None:
|
30 |
+
generator = np.random
|
31 |
+
probs_idx = np.argsort(-probs, axis=-1)
|
32 |
+
probs_sort = np.take_along_axis(probs, probs_idx, -1)
|
33 |
+
probs_sum = np.cumsum(probs_sort, axis=-1)
|
34 |
+
mask = probs_sum - probs_sort > p
|
35 |
+
probs_sort[mask] = 0.0
|
36 |
+
mask = np.zeros(probs_sort.shape[-1])
|
37 |
+
mask[:k] = 1
|
38 |
+
probs_sort = probs_sort * mask
|
39 |
+
probs_sort /= np.sum(probs_sort, axis=-1, keepdims=True)
|
40 |
+
shape = probs_sort.shape
|
41 |
+
probs_sort_flat = probs_sort.reshape(-1, shape[-1])
|
42 |
+
probs_idx_flat = probs_idx.reshape(-1, shape[-1])
|
43 |
+
next_token = np.stack([generator.choice(idxs, p=pvals) for pvals, idxs in zip(probs_sort_flat, probs_idx_flat)])
|
44 |
+
next_token = next_token.reshape(*shape[:-1])
|
45 |
+
return next_token
|
46 |
+
|
47 |
+
|
48 |
def generate(model, prompt=None, max_len=512, temp=1.0, top_p=0.98, top_k=20,
|
49 |
+
disable_patch_change=False, disable_control_change=False, disable_channels=None, generator=None):
|
50 |
if disable_channels is not None:
|
51 |
disable_channels = [tokenizer.parameter_ids["channel"][c] for c in disable_channels]
|
52 |
else:
|
53 |
disable_channels = []
|
54 |
+
if generator is None:
|
55 |
+
generator = np.random
|
56 |
max_token_seq = tokenizer.max_token_seq
|
57 |
if prompt is None:
|
58 |
+
input_tensor = np.full((1, max_token_seq), tokenizer.pad_id, dtype=np.int64)
|
59 |
input_tensor[0, 0] = tokenizer.bos_id # bos
|
60 |
else:
|
61 |
prompt = prompt[:, :max_token_seq]
|
62 |
if prompt.shape[-1] < max_token_seq:
|
63 |
prompt = np.pad(prompt, ((0, 0), (0, max_token_seq - prompt.shape[-1])),
|
64 |
mode="constant", constant_values=tokenizer.pad_id)
|
65 |
+
input_tensor = prompt
|
66 |
+
input_tensor = input_tensor[None, :, :]
|
67 |
cur_len = input_tensor.shape[1]
|
68 |
bar = tqdm.tqdm(desc="generating", total=max_len - cur_len, disable=in_space)
|
69 |
+
with bar:
|
70 |
while cur_len < max_len:
|
71 |
end = False
|
72 |
+
hidden = model[0].run(None, {'x': input_tensor})[0][:, -1]
|
73 |
+
next_token_seq = np.empty((1, 0), dtype=np.int64)
|
74 |
event_name = ""
|
75 |
for i in range(max_token_seq):
|
76 |
+
mask = np.zeros(tokenizer.vocab_size, dtype=np.int64)
|
77 |
if i == 0:
|
78 |
mask_ids = list(tokenizer.event_ids.values()) + [tokenizer.eos_id]
|
79 |
if disable_patch_change:
|
|
|
87 |
if param_name == "channel":
|
88 |
mask_ids = [i for i in mask_ids if i not in disable_channels]
|
89 |
mask[mask_ids] = 1
|
90 |
+
logits = model[1].run(None, {'x': next_token_seq, "hidden": hidden})[0][:, -1:]
|
91 |
+
scores = softmax(logits / temp, -1) * mask
|
92 |
+
sample = sample_top_p_k(scores, top_p, top_k, generator)
|
93 |
if i == 0:
|
94 |
next_token_seq = sample
|
95 |
eid = sample.item()
|
|
|
98 |
break
|
99 |
event_name = tokenizer.id_events[eid]
|
100 |
else:
|
101 |
+
next_token_seq = np.concatenate([next_token_seq, sample], axis=1)
|
102 |
if len(tokenizer.events[event_name]) == i:
|
103 |
break
|
104 |
if next_token_seq.shape[1] < max_token_seq:
|
105 |
+
next_token_seq = np.pad(next_token_seq, ((0, 0), (0, max_token_seq - next_token_seq.shape[-1])),
|
106 |
+
mode="constant", constant_values=tokenizer.pad_id)
|
107 |
+
next_token_seq = next_token_seq[None, :, :]
|
108 |
+
input_tensor = np.concatenate([input_tensor, next_token_seq], axis=1)
|
109 |
cur_len += 1
|
110 |
bar.update(1)
|
111 |
+
yield next_token_seq.reshape(-1)
|
112 |
if end:
|
113 |
break
|
114 |
|
|
|
129 |
max_len = gen_events
|
130 |
if seed_rand:
|
131 |
seed = np.random.randint(0, MAX_SEED)
|
132 |
+
generator = np.random.RandomState(seed)
|
133 |
disable_patch_change = False
|
134 |
disable_channels = None
|
135 |
if tab == 0:
|
|
|
160 |
for token_seq in mid:
|
161 |
mid_seq.append(token_seq.tolist())
|
162 |
max_len += len(mid)
|
|
|
163 |
events = [tokenizer.tokens2event(tokens) for tokens in mid_seq]
|
164 |
init_msgs = [create_msg("visualizer_clear", None), create_msg("visualizer_append", events)]
|
165 |
t = time.time() + 1
|
166 |
yield mid_seq, None, None, seed, send_msgs(init_msgs)
|
167 |
model = models[model_name]
|
|
|
168 |
midi_generator = generate(model, mid, max_len=max_len, temp=temp, top_p=top_p, top_k=top_k,
|
169 |
disable_patch_change=disable_patch_change, disable_control_change=not allow_cc,
|
170 |
+
disable_channels=disable_channels, generator=generator)
|
171 |
events = []
|
172 |
for i, token_seq in enumerate(midi_generator):
|
173 |
token_seq = token_seq.tolist()
|
|
|
245 |
"j-pop finetune model": ["skytnt/midi-model-ft", "jpop/"],
|
246 |
"touhou finetune model": ["skytnt/midi-model-ft", "touhou/"],
|
247 |
}
|
|
|
|
|
|
|
|
|
248 |
models = {}
|
249 |
tokenizer = MIDITokenizer()
|
250 |
+
providers = ['CUDAExecutionProvider', 'CPUExecutionProvider']
|
251 |
for name, (repo_id, path) in models_info.items():
|
252 |
+
model_base_path = hf_hub_download_retry(repo_id=repo_id, filename=f"{path}onnx/model_base.onnx")
|
253 |
+
model_token_path = hf_hub_download_retry(repo_id=repo_id, filename=f"{path}onnx/model_token.onnx")
|
254 |
+
model_base = rt.InferenceSession(model_base_path, providers=providers)
|
255 |
+
model_token = rt.InferenceSession(model_token_path, providers=providers)
|
256 |
+
models[name] = [model_base, model_token]
|
|
|
|
|
|
|
257 |
|
258 |
load_javascript()
|
259 |
app = gr.Blocks()
|
|
|
265 |
"[Open In Colab]"
|
266 |
"(https://colab.research.google.com/github/SkyTNT/midi-model/blob/main/demo.ipynb)"
|
267 |
" for faster running and longer generation\n\n"
|
268 |
+
"**Update v1.2**: Optimise the tokenizer and dataset"
|
|
|
269 |
)
|
270 |
js_msg = gr.Textbox(elem_id="msg_receiver", visible=False)
|
271 |
js_msg.change(None, [js_msg], [], js="""
|
|
|
335 |
[output_midi_seq, output_midi, output_audio, input_seed, js_msg],
|
336 |
concurrency_limit=3)
|
337 |
stop_btn.click(cancel_run, [output_midi_seq], [output_midi, output_audio, js_msg], cancels=run_event, queue=False)
|
338 |
+
app.launch(server_port=opt.port, share=opt.share, inbrowser=True)
|
requirements.txt
CHANGED
@@ -1,8 +1,6 @@
|
|
1 |
-
--extra-index-url https://download.pytorch.org/whl/cu124
|
2 |
Pillow
|
3 |
numpy
|
4 |
-
|
5 |
-
transformers>=4.36
|
6 |
gradio==4.43.0
|
7 |
pyfluidsynth
|
8 |
tqdm
|
|
|
|
|
1 |
Pillow
|
2 |
numpy
|
3 |
+
onnxruntime-gpu
|
|
|
4 |
gradio==4.43.0
|
5 |
pyfluidsynth
|
6 |
tqdm
|