Spaces:
Running
on
Zero
Running
on
Zero
torch
Browse files- app.py +58 -92
- midi_model.py +56 -16
- requirements.txt +3 -1
app.py
CHANGED
@@ -1,79 +1,53 @@
|
|
1 |
import argparse
|
2 |
import glob
|
3 |
import json
|
4 |
-
import os
|
5 |
import time
|
6 |
|
7 |
import gradio as gr
|
8 |
import numpy as np
|
9 |
-
import
|
|
|
10 |
import tqdm
|
11 |
from huggingface_hub import hf_hub_download
|
12 |
|
13 |
import MIDI
|
|
|
14 |
from midi_synthesizer import MidiSynthesizer
|
15 |
-
from midi_tokenizer import MIDITokenizer
|
16 |
|
17 |
MAX_SEED = np.iinfo(np.int32).max
|
18 |
in_space = os.getenv("SYSTEM") == "spaces"
|
19 |
|
20 |
|
21 |
-
|
22 |
-
|
23 |
-
exp_x_shifted = np.exp(x - x_max)
|
24 |
-
return exp_x_shifted / np.sum(exp_x_shifted, axis=axis, keepdims=True)
|
25 |
-
|
26 |
-
|
27 |
-
def sample_top_p_k(probs, p, k, generator=None):
|
28 |
-
if generator is None:
|
29 |
-
generator = np.random
|
30 |
-
probs_idx = np.argsort(-probs, axis=-1)
|
31 |
-
probs_sort = np.take_along_axis(probs, probs_idx, -1)
|
32 |
-
probs_sum = np.cumsum(probs_sort, axis=-1)
|
33 |
-
mask = probs_sum - probs_sort > p
|
34 |
-
probs_sort[mask] = 0.0
|
35 |
-
mask = np.zeros(probs_sort.shape[-1])
|
36 |
-
mask[:k] = 1
|
37 |
-
probs_sort = probs_sort * mask
|
38 |
-
probs_sort /= np.sum(probs_sort, axis=-1, keepdims=True)
|
39 |
-
shape = probs_sort.shape
|
40 |
-
probs_sort_flat = probs_sort.reshape(-1, shape[-1])
|
41 |
-
probs_idx_flat = probs_idx.reshape(-1, shape[-1])
|
42 |
-
next_token = np.stack([generator.choice(idxs, p=pvals) for pvals, idxs in zip(probs_sort_flat, probs_idx_flat)])
|
43 |
-
next_token = next_token.reshape(*shape[:-1])
|
44 |
-
return next_token
|
45 |
-
|
46 |
-
|
47 |
-
def generate(model, prompt=None, max_len=512, temp=1.0, top_p=0.98, top_k=20,
|
48 |
disable_patch_change=False, disable_control_change=False, disable_channels=None, generator=None):
|
49 |
-
tokenizer = model
|
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 =
|
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
|
67 |
cur_len = input_tensor.shape[1]
|
68 |
-
bar = tqdm.tqdm(desc="generating", total=max_len - cur_len
|
69 |
with bar:
|
70 |
while cur_len < max_len:
|
71 |
end = False
|
72 |
-
hidden = model
|
73 |
-
next_token_seq =
|
74 |
event_name = ""
|
75 |
for i in range(max_token_seq):
|
76 |
-
mask =
|
77 |
if i == 0:
|
78 |
mask_ids = list(tokenizer.event_ids.values()) + [tokenizer.eos_id]
|
79 |
if disable_patch_change:
|
@@ -87,9 +61,9 @@ def generate(model, prompt=None, max_len=512, temp=1.0, top_p=0.98, top_k=20,
|
|
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
|
91 |
-
scores = softmax(logits / temp,
|
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()
|
@@ -98,17 +72,17 @@ def generate(model, prompt=None, max_len=512, temp=1.0, top_p=0.98, top_k=20,
|
|
98 |
break
|
99 |
event_name = tokenizer.id_events[eid]
|
100 |
else:
|
101 |
-
next_token_seq =
|
102 |
if len(tokenizer.events[event_name]) == i:
|
103 |
break
|
104 |
if next_token_seq.shape[1] < max_token_seq:
|
105 |
-
next_token_seq =
|
106 |
-
|
107 |
-
next_token_seq = next_token_seq
|
108 |
-
input_tensor =
|
109 |
cur_len += 1
|
110 |
bar.update(1)
|
111 |
-
yield next_token_seq.reshape(-1)
|
112 |
if end:
|
113 |
break
|
114 |
|
@@ -125,7 +99,7 @@ def run(model_name, tab, mid_seq, continuation_state, instruments, drum_kit, bpm
|
|
125 |
reduce_cc_st, remap_track_channel, add_default_instr, remove_empty_channels, seed, seed_rand,
|
126 |
gen_events, temp, top_p, top_k, allow_cc):
|
127 |
model = models[model_name]
|
128 |
-
tokenizer = model
|
129 |
bpm = int(bpm)
|
130 |
if time_sig == "auto":
|
131 |
time_sig = None
|
@@ -147,7 +121,7 @@ def run(model_name, tab, mid_seq, continuation_state, instruments, drum_kit, bpm
|
|
147 |
max_len = gen_events
|
148 |
if seed_rand:
|
149 |
seed = np.random.randint(0, MAX_SEED)
|
150 |
-
generator =
|
151 |
disable_patch_change = False
|
152 |
disable_channels = None
|
153 |
if tab == 0:
|
@@ -203,22 +177,24 @@ def run(model_name, tab, mid_seq, continuation_state, instruments, drum_kit, bpm
|
|
203 |
init_msgs += [create_msg("visualizer_clear", tokenizer.version),
|
204 |
create_msg("visualizer_append", events)]
|
205 |
yield mid_seq, continuation_state, None, None, seed, send_msgs(init_msgs)
|
206 |
-
|
207 |
-
|
208 |
-
|
209 |
-
|
210 |
-
|
211 |
-
|
212 |
-
|
213 |
-
|
214 |
-
|
215 |
-
|
216 |
-
|
217 |
-
|
218 |
-
|
219 |
-
|
220 |
-
|
221 |
-
|
|
|
|
|
222 |
|
223 |
events = [tokenizer.tokens2event(tokens) for tokens in mid_seq]
|
224 |
mid = tokenizer.detokenize(mid_seq)
|
@@ -235,7 +211,7 @@ def run(model_name, tab, mid_seq, continuation_state, instruments, drum_kit, bpm
|
|
235 |
def cancel_run(model_name, mid_seq):
|
236 |
if mid_seq is None:
|
237 |
return None, None, []
|
238 |
-
tokenizer = models[model_name]
|
239 |
events = [tokenizer.tokens2event(tokens) for tokens in mid_seq]
|
240 |
mid = tokenizer.detokenize(mid_seq)
|
241 |
audio = synthesizer.synthesis(MIDI.score2opus(mid))
|
@@ -248,11 +224,12 @@ def cancel_run(model_name, mid_seq):
|
|
248 |
return "output.mid", (44100, audio), send_msgs(end_msgs)
|
249 |
|
250 |
|
251 |
-
def undo_continuation(mid_seq, continuation_state):
|
252 |
if mid_seq is None or len(continuation_state) < 2:
|
253 |
return mid_seq, continuation_state, send_msgs([])
|
254 |
mid_seq = mid_seq[:continuation_state[-1]]
|
255 |
continuation_state = continuation_state[:-1]
|
|
|
256 |
events = [tokenizer.tokens2event(tokens) for tokens in mid_seq]
|
257 |
end_msgs = [create_msg("visualizer_clear", tokenizer.version),
|
258 |
create_msg("visualizer_append", events),
|
@@ -293,21 +270,6 @@ def hf_hub_download_retry(repo_id, filename):
|
|
293 |
raise err
|
294 |
|
295 |
|
296 |
-
def get_tokenizer(config_name):
|
297 |
-
tv, size = config_name.split("-")
|
298 |
-
tv = tv[1:]
|
299 |
-
if tv[-1] == "o":
|
300 |
-
o = True
|
301 |
-
tv = tv[:-1]
|
302 |
-
else:
|
303 |
-
o = False
|
304 |
-
if tv not in ["v1", "v2"]:
|
305 |
-
raise ValueError(f"Unknown tokenizer version {tv}")
|
306 |
-
tokenizer = MIDITokenizer(tv)
|
307 |
-
tokenizer.set_optimise_midi(o)
|
308 |
-
return tokenizer
|
309 |
-
|
310 |
-
|
311 |
number2drum_kits = {-1: "None", 0: "Standard", 8: "Room", 16: "Power", 24: "Electric", 25: "TR-808", 32: "Jazz",
|
312 |
40: "Blush", 48: "Orchestra"}
|
313 |
patch2number = {v: k for k, v in MIDI.Number2patch.items()}
|
@@ -319,6 +281,7 @@ if __name__ == "__main__":
|
|
319 |
parser = argparse.ArgumentParser()
|
320 |
parser.add_argument("--share", action="store_true", default=False, help="share gradio app")
|
321 |
parser.add_argument("--port", type=int, default=7860, help="gradio server port")
|
|
|
322 |
parser.add_argument("--max-gen", type=int, default=1024, help="max")
|
323 |
opt = parser.parse_args()
|
324 |
soundfont_path = hf_hub_download_retry(repo_id="skytnt/midi-model", filename="soundfont.sf2")
|
@@ -331,14 +294,17 @@ if __name__ == "__main__":
|
|
331 |
"touhou finetune model (tv1-medium) by skytnt": ["skytnt/midi-model-ft", "touhou/", "tv1-medium"],
|
332 |
}
|
333 |
models = {}
|
334 |
-
|
|
|
|
|
335 |
for name, (repo_id, path, config) in models_info.items():
|
336 |
-
|
337 |
-
|
338 |
-
|
339 |
-
|
340 |
-
|
341 |
-
|
|
|
342 |
|
343 |
load_javascript()
|
344 |
app = gr.Blocks()
|
@@ -447,6 +413,6 @@ if __name__ == "__main__":
|
|
447 |
stop_btn.click(cancel_run, [input_model, output_midi_seq],
|
448 |
[output_midi, output_audio, js_msg],
|
449 |
cancels=run_event, queue=False)
|
450 |
-
undo_btn.click(undo_continuation, [output_midi_seq, output_continuation_state],
|
451 |
[output_midi_seq, output_continuation_state, js_msg], queue=False)
|
452 |
app.launch(server_port=opt.port, share=opt.share, inbrowser=True)
|
|
|
1 |
import argparse
|
2 |
import glob
|
3 |
import json
|
4 |
+
import os
|
5 |
import time
|
6 |
|
7 |
import gradio as gr
|
8 |
import numpy as np
|
9 |
+
import torch
|
10 |
+
import torch.nn.functional as F
|
11 |
import tqdm
|
12 |
from huggingface_hub import hf_hub_download
|
13 |
|
14 |
import MIDI
|
15 |
+
from midi_model import MIDIModel, MIDIModelConfig
|
16 |
from midi_synthesizer import MidiSynthesizer
|
|
|
17 |
|
18 |
MAX_SEED = np.iinfo(np.int32).max
|
19 |
in_space = os.getenv("SYSTEM") == "spaces"
|
20 |
|
21 |
|
22 |
+
@torch.inference_mode()
|
23 |
+
def generate(model: MIDIModel, prompt=None, max_len=512, temp=1.0, top_p=0.98, top_k=20,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
disable_patch_change=False, disable_control_change=False, disable_channels=None, generator=None):
|
25 |
+
tokenizer = model.tokenizer
|
26 |
if disable_channels is not None:
|
27 |
disable_channels = [tokenizer.parameter_ids["channel"][c] for c in disable_channels]
|
28 |
else:
|
29 |
disable_channels = []
|
|
|
|
|
30 |
max_token_seq = tokenizer.max_token_seq
|
31 |
if prompt is None:
|
32 |
+
input_tensor = torch.full((1, max_token_seq), tokenizer.pad_id, dtype=torch.long, device=model.device)
|
33 |
input_tensor[0, 0] = tokenizer.bos_id # bos
|
34 |
else:
|
35 |
prompt = prompt[:, :max_token_seq]
|
36 |
if prompt.shape[-1] < max_token_seq:
|
37 |
prompt = np.pad(prompt, ((0, 0), (0, max_token_seq - prompt.shape[-1])),
|
38 |
mode="constant", constant_values=tokenizer.pad_id)
|
39 |
+
input_tensor = torch.from_numpy(prompt).to(dtype=torch.long, device=model.device)
|
40 |
+
input_tensor = input_tensor.unsqueeze(0)
|
41 |
cur_len = input_tensor.shape[1]
|
42 |
+
bar = tqdm.tqdm(desc="generating", total=max_len - cur_len)
|
43 |
with bar:
|
44 |
while cur_len < max_len:
|
45 |
end = False
|
46 |
+
hidden = model.forward(input_tensor)[0, -1].unsqueeze(0)
|
47 |
+
next_token_seq = None
|
48 |
event_name = ""
|
49 |
for i in range(max_token_seq):
|
50 |
+
mask = torch.zeros(tokenizer.vocab_size, dtype=torch.int64, device=model.device)
|
51 |
if i == 0:
|
52 |
mask_ids = list(tokenizer.event_ids.values()) + [tokenizer.eos_id]
|
53 |
if disable_patch_change:
|
|
|
61 |
if param_name == "channel":
|
62 |
mask_ids = [i for i in mask_ids if i not in disable_channels]
|
63 |
mask[mask_ids] = 1
|
64 |
+
logits = model.forward_token(hidden, next_token_seq)[:, -1:]
|
65 |
+
scores = torch.softmax(logits / temp, dim=-1) * mask
|
66 |
+
sample = model.sample_top_p_k(scores, top_p, top_k, generator=generator)
|
67 |
if i == 0:
|
68 |
next_token_seq = sample
|
69 |
eid = sample.item()
|
|
|
72 |
break
|
73 |
event_name = tokenizer.id_events[eid]
|
74 |
else:
|
75 |
+
next_token_seq = torch.cat([next_token_seq, sample], dim=1)
|
76 |
if len(tokenizer.events[event_name]) == i:
|
77 |
break
|
78 |
if next_token_seq.shape[1] < max_token_seq:
|
79 |
+
next_token_seq = F.pad(next_token_seq, (0, max_token_seq - next_token_seq.shape[1]),
|
80 |
+
"constant", value=tokenizer.pad_id)
|
81 |
+
next_token_seq = next_token_seq.unsqueeze(1)
|
82 |
+
input_tensor = torch.cat([input_tensor, next_token_seq], dim=1)
|
83 |
cur_len += 1
|
84 |
bar.update(1)
|
85 |
+
yield next_token_seq.reshape(-1).cpu().numpy()
|
86 |
if end:
|
87 |
break
|
88 |
|
|
|
99 |
reduce_cc_st, remap_track_channel, add_default_instr, remove_empty_channels, seed, seed_rand,
|
100 |
gen_events, temp, top_p, top_k, allow_cc):
|
101 |
model = models[model_name]
|
102 |
+
tokenizer = model.tokenizer
|
103 |
bpm = int(bpm)
|
104 |
if time_sig == "auto":
|
105 |
time_sig = None
|
|
|
121 |
max_len = gen_events
|
122 |
if seed_rand:
|
123 |
seed = np.random.randint(0, MAX_SEED)
|
124 |
+
generator = torch.Generator(opt.device).manual_seed(seed)
|
125 |
disable_patch_change = False
|
126 |
disable_channels = None
|
127 |
if tab == 0:
|
|
|
177 |
init_msgs += [create_msg("visualizer_clear", tokenizer.version),
|
178 |
create_msg("visualizer_append", events)]
|
179 |
yield mid_seq, continuation_state, None, None, seed, send_msgs(init_msgs)
|
180 |
+
ctx = torch.amp.autocast(device_type=opt.device, dtype=torch.bfloat16, enabled=opt.device != "cpu")
|
181 |
+
with ctx:
|
182 |
+
midi_generator = generate(model, mid, max_len=max_len, temp=temp, top_p=top_p, top_k=top_k,
|
183 |
+
disable_patch_change=disable_patch_change, disable_control_change=not allow_cc,
|
184 |
+
disable_channels=disable_channels, generator=generator)
|
185 |
+
events = []
|
186 |
+
t = time.time() + 1
|
187 |
+
for i, token_seq in enumerate(midi_generator):
|
188 |
+
token_seq = token_seq.tolist()
|
189 |
+
mid_seq.append(token_seq)
|
190 |
+
events.append(tokenizer.tokens2event(token_seq))
|
191 |
+
ct = time.time()
|
192 |
+
if ct - t > 0.5:
|
193 |
+
yield (mid_seq, continuation_state, None, None, seed,
|
194 |
+
send_msgs([create_msg("visualizer_append", events),
|
195 |
+
create_msg("progress", [i + 1, gen_events])]))
|
196 |
+
t = ct
|
197 |
+
events = []
|
198 |
|
199 |
events = [tokenizer.tokens2event(tokens) for tokens in mid_seq]
|
200 |
mid = tokenizer.detokenize(mid_seq)
|
|
|
211 |
def cancel_run(model_name, mid_seq):
|
212 |
if mid_seq is None:
|
213 |
return None, None, []
|
214 |
+
tokenizer = models[model_name].tokenizer
|
215 |
events = [tokenizer.tokens2event(tokens) for tokens in mid_seq]
|
216 |
mid = tokenizer.detokenize(mid_seq)
|
217 |
audio = synthesizer.synthesis(MIDI.score2opus(mid))
|
|
|
224 |
return "output.mid", (44100, audio), send_msgs(end_msgs)
|
225 |
|
226 |
|
227 |
+
def undo_continuation(model_name, mid_seq, continuation_state):
|
228 |
if mid_seq is None or len(continuation_state) < 2:
|
229 |
return mid_seq, continuation_state, send_msgs([])
|
230 |
mid_seq = mid_seq[:continuation_state[-1]]
|
231 |
continuation_state = continuation_state[:-1]
|
232 |
+
tokenizer = models[model_name].tokenizer
|
233 |
events = [tokenizer.tokens2event(tokens) for tokens in mid_seq]
|
234 |
end_msgs = [create_msg("visualizer_clear", tokenizer.version),
|
235 |
create_msg("visualizer_append", events),
|
|
|
270 |
raise err
|
271 |
|
272 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
273 |
number2drum_kits = {-1: "None", 0: "Standard", 8: "Room", 16: "Power", 24: "Electric", 25: "TR-808", 32: "Jazz",
|
274 |
40: "Blush", 48: "Orchestra"}
|
275 |
patch2number = {v: k for k, v in MIDI.Number2patch.items()}
|
|
|
281 |
parser = argparse.ArgumentParser()
|
282 |
parser.add_argument("--share", action="store_true", default=False, help="share gradio app")
|
283 |
parser.add_argument("--port", type=int, default=7860, help="gradio server port")
|
284 |
+
parser.add_argument("--device", type=str, default="cuda", help="device to run model")
|
285 |
parser.add_argument("--max-gen", type=int, default=1024, help="max")
|
286 |
opt = parser.parse_args()
|
287 |
soundfont_path = hf_hub_download_retry(repo_id="skytnt/midi-model", filename="soundfont.sf2")
|
|
|
294 |
"touhou finetune model (tv1-medium) by skytnt": ["skytnt/midi-model-ft", "touhou/", "tv1-medium"],
|
295 |
}
|
296 |
models = {}
|
297 |
+
if opt.device == "cuda":
|
298 |
+
torch.backends.cuda.enable_mem_efficient_sdp(True)
|
299 |
+
torch.backends.cuda.enable_flash_sdp(True)
|
300 |
for name, (repo_id, path, config) in models_info.items():
|
301 |
+
model_path = hf_hub_download_retry(repo_id=repo_id, filename=f"{path}model.ckpt")
|
302 |
+
model = MIDIModel(config=MIDIModelConfig.from_name(config))
|
303 |
+
ckpt = torch.load(model_path, map_location="cpu")
|
304 |
+
state_dict = ckpt.get("state_dict", ckpt)
|
305 |
+
model.load_state_dict(state_dict, strict=False)
|
306 |
+
model.to(device=opt.device, dtype=torch.bfloat16 if opt.device == "cuda" else torch.float32).eval()
|
307 |
+
models[name] = model
|
308 |
|
309 |
load_javascript()
|
310 |
app = gr.Blocks()
|
|
|
413 |
stop_btn.click(cancel_run, [input_model, output_midi_seq],
|
414 |
[output_midi, output_audio, js_msg],
|
415 |
cancels=run_event, queue=False)
|
416 |
+
undo_btn.click(undo_continuation, [input_model, output_midi_seq, output_continuation_state],
|
417 |
[output_midi_seq, output_continuation_state, js_msg], queue=False)
|
418 |
app.launch(server_port=opt.port, share=opt.share, inbrowser=True)
|
midi_model.py
CHANGED
@@ -1,3 +1,5 @@
|
|
|
|
|
|
1 |
import numpy as np
|
2 |
import torch
|
3 |
import torch.nn as nn
|
@@ -5,23 +7,61 @@ import torch.nn.functional as F
|
|
5 |
import tqdm
|
6 |
from transformers import LlamaModel, LlamaConfig
|
7 |
|
8 |
-
from midi_tokenizer import MIDITokenizer
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
|
10 |
|
11 |
class MIDIModel(nn.Module):
|
12 |
-
def __init__(self,
|
13 |
-
*args, **kwargs):
|
14 |
super(MIDIModel, self).__init__()
|
15 |
-
self.tokenizer = tokenizer
|
16 |
-
self.net = LlamaModel(
|
17 |
-
|
18 |
-
|
19 |
-
pad_token_id=tokenizer.pad_id, max_position_embeddings=4096))
|
20 |
-
self.net_token = LlamaModel(LlamaConfig(vocab_size=tokenizer.vocab_size,
|
21 |
-
hidden_size=n_embd, num_attention_heads=n_head // 4,
|
22 |
-
num_hidden_layers=n_layer // 4, intermediate_size=n_inner // 4,
|
23 |
-
pad_token_id=tokenizer.pad_id, max_position_embeddings=4096))
|
24 |
-
self.lm_head = nn.Linear(n_embd, tokenizer.vocab_size, bias=False)
|
25 |
self.device = "cpu"
|
26 |
|
27 |
def to(self, *args, **kwargs):
|
@@ -71,7 +111,7 @@ class MIDIModel(nn.Module):
|
|
71 |
return next_token
|
72 |
|
73 |
@torch.inference_mode()
|
74 |
-
def generate(self, prompt=None, max_len=512, temp=1.0, top_p=0.98, top_k=20,
|
75 |
tokenizer = self.tokenizer
|
76 |
max_token_seq = tokenizer.max_token_seq
|
77 |
if prompt is None:
|
@@ -86,7 +126,7 @@ class MIDIModel(nn.Module):
|
|
86 |
input_tensor = input_tensor.unsqueeze(0)
|
87 |
cur_len = input_tensor.shape[1]
|
88 |
bar = tqdm.tqdm(desc="generating", total=max_len - cur_len)
|
89 |
-
with bar
|
90 |
while cur_len < max_len:
|
91 |
end = False
|
92 |
hidden = self.forward(input_tensor)[0, -1].unsqueeze(0)
|
@@ -123,4 +163,4 @@ class MIDIModel(nn.Module):
|
|
123 |
bar.update(1)
|
124 |
if end:
|
125 |
break
|
126 |
-
return input_tensor[0].cpu().numpy()
|
|
|
1 |
+
from typing import Union
|
2 |
+
|
3 |
import numpy as np
|
4 |
import torch
|
5 |
import torch.nn as nn
|
|
|
7 |
import tqdm
|
8 |
from transformers import LlamaModel, LlamaConfig
|
9 |
|
10 |
+
from midi_tokenizer import MIDITokenizerV1, MIDITokenizerV2, MIDITokenizer
|
11 |
+
|
12 |
+
config_name_list = ["tv1-medium", "tv2-medium", "tv2o-medium", "tv2-large", "tv2o-large"]
|
13 |
+
|
14 |
+
|
15 |
+
class MIDIModelConfig:
|
16 |
+
def __init__(self, tokenizer: Union[MIDITokenizerV1, MIDITokenizerV2],
|
17 |
+
net_config: LlamaConfig, net_token_config: LlamaConfig):
|
18 |
+
self.tokenizer = tokenizer
|
19 |
+
self.net_config = net_config
|
20 |
+
self.net_token_config = net_token_config
|
21 |
+
self.n_embd = net_token_config.hidden_size
|
22 |
+
|
23 |
+
@staticmethod
|
24 |
+
def get_config(tokenizer_ver="v2", optimise_midi=True, n_layer=12, n_head=16, n_embd=1024, n_inner=4096):
|
25 |
+
tokenizer = MIDITokenizer(tokenizer_ver)
|
26 |
+
tokenizer.set_optimise_midi(optimise_midi)
|
27 |
+
net_config = LlamaConfig(vocab_size=tokenizer.vocab_size,
|
28 |
+
hidden_size=n_embd, num_attention_heads=n_head,
|
29 |
+
num_hidden_layers=n_layer, intermediate_size=n_inner,
|
30 |
+
pad_token_id=tokenizer.pad_id, max_position_embeddings=4096)
|
31 |
+
net_token_config = LlamaConfig(vocab_size=tokenizer.vocab_size,
|
32 |
+
hidden_size=n_embd, num_attention_heads=n_head // 4,
|
33 |
+
num_hidden_layers=n_layer // 4, intermediate_size=n_inner // 4,
|
34 |
+
pad_token_id=tokenizer.pad_id, max_position_embeddings=4096)
|
35 |
+
return MIDIModelConfig(tokenizer, net_config, net_token_config)
|
36 |
+
|
37 |
+
@staticmethod
|
38 |
+
def from_name(name="tv2o-medium"):
|
39 |
+
tv, size = name.split("-")
|
40 |
+
tv = tv[1:]
|
41 |
+
if tv[-1] == "o":
|
42 |
+
o = True
|
43 |
+
tv = tv[:-1]
|
44 |
+
else:
|
45 |
+
o = False
|
46 |
+
if tv not in ["v1", "v2"]:
|
47 |
+
raise ValueError(f"Unknown tokenizer version {tv}")
|
48 |
+
if size == "medium":
|
49 |
+
return MIDIModelConfig.get_config(tokenizer_ver=tv, optimise_midi=o,
|
50 |
+
n_layer=12, n_head=16, n_embd=1024, n_inner=4096)
|
51 |
+
elif size == "large":
|
52 |
+
return MIDIModelConfig.get_config(tokenizer_ver=tv, optimise_midi=o,
|
53 |
+
n_layer=24, n_head=16, n_embd=1024, n_inner=4096)
|
54 |
+
else:
|
55 |
+
raise ValueError(f"Unknown model size {size}")
|
56 |
|
57 |
|
58 |
class MIDIModel(nn.Module):
|
59 |
+
def __init__(self, config: MIDIModelConfig, *args, **kwargs):
|
|
|
60 |
super(MIDIModel, self).__init__()
|
61 |
+
self.tokenizer = config.tokenizer
|
62 |
+
self.net = LlamaModel(config.net_config)
|
63 |
+
self.net_token = LlamaModel(config.net_token_config)
|
64 |
+
self.lm_head = nn.Linear(config.n_embd, self.tokenizer.vocab_size, bias=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
65 |
self.device = "cpu"
|
66 |
|
67 |
def to(self, *args, **kwargs):
|
|
|
111 |
return next_token
|
112 |
|
113 |
@torch.inference_mode()
|
114 |
+
def generate(self, prompt=None, max_len=512, temp=1.0, top_p=0.98, top_k=20, generator=None):
|
115 |
tokenizer = self.tokenizer
|
116 |
max_token_seq = tokenizer.max_token_seq
|
117 |
if prompt is None:
|
|
|
126 |
input_tensor = input_tensor.unsqueeze(0)
|
127 |
cur_len = input_tensor.shape[1]
|
128 |
bar = tqdm.tqdm(desc="generating", total=max_len - cur_len)
|
129 |
+
with bar:
|
130 |
while cur_len < max_len:
|
131 |
end = False
|
132 |
hidden = self.forward(input_tensor)[0, -1].unsqueeze(0)
|
|
|
163 |
bar.update(1)
|
164 |
if end:
|
165 |
break
|
166 |
+
return input_tensor[0].cpu().numpy()
|
requirements.txt
CHANGED
@@ -1,6 +1,8 @@
|
|
|
|
1 |
Pillow
|
2 |
numpy
|
3 |
-
|
|
|
4 |
gradio==4.43.0
|
5 |
pyfluidsynth
|
6 |
tqdm
|
|
|
1 |
+
--extra-index-url https://download.pytorch.org/whl/cu124
|
2 |
Pillow
|
3 |
numpy
|
4 |
+
torch
|
5 |
+
transformers>=4.36
|
6 |
gradio==4.43.0
|
7 |
pyfluidsynth
|
8 |
tqdm
|