|
|
import argparse |
|
|
import glob |
|
|
import os.path |
|
|
|
|
|
import torch |
|
|
import torch.nn.functional as F |
|
|
|
|
|
import gradio as gr |
|
|
import numpy as np |
|
|
import onnxruntime as rt |
|
|
import tqdm |
|
|
import json |
|
|
|
|
|
from midi_synthesizer import synthesis |
|
|
import TMIDIX |
|
|
|
|
|
in_space = os.getenv("SYSTEM") == "spaces" |
|
|
|
|
|
providers = ['CPUExecutionProvider'] |
|
|
|
|
|
|
|
|
|
|
|
def generate( |
|
|
start_tokens, |
|
|
seq_len, |
|
|
max_seq_len = 2048, |
|
|
temperature = 0.9, |
|
|
verbose=True, |
|
|
return_prime=False, |
|
|
): |
|
|
|
|
|
out = torch.LongTensor([start_tokens]) |
|
|
|
|
|
st = len(start_tokens) |
|
|
|
|
|
if verbose: |
|
|
print("Generating sequence of max length:", seq_len) |
|
|
|
|
|
for s in range(seq_len): |
|
|
x = out[:, -max_seq_len:] |
|
|
|
|
|
torch_in = x.tolist()[0] |
|
|
|
|
|
logits = torch.FloatTensor(session.run(None, {'input': [torch_in]})[0])[:, -1] |
|
|
|
|
|
filtered_logits = logits |
|
|
|
|
|
probs = F.softmax(filtered_logits / temperature, dim=-1) |
|
|
|
|
|
sample = torch.multinomial(probs, 1) |
|
|
|
|
|
out = torch.cat((out, sample), dim=-1) |
|
|
|
|
|
if verbose: |
|
|
if s % 32 == 0: |
|
|
print(s, '/', seq_len) |
|
|
|
|
|
if return_prime: |
|
|
return out[:, :] |
|
|
|
|
|
else: |
|
|
return out[:, st:] |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def load_javascript(dir="javascript"): |
|
|
scripts_list = glob.glob(f"{dir}/*.js") |
|
|
javascript = "" |
|
|
for path in scripts_list: |
|
|
with open(path, "r", encoding="utf8") as jsfile: |
|
|
javascript += f"\n<!-- {path} --><script>{jsfile.read()}</script>" |
|
|
template_response_ori = gr.routes.templates.TemplateResponse |
|
|
|
|
|
def template_response(*args, **kwargs): |
|
|
res = template_response_ori(*args, **kwargs) |
|
|
res.body = res.body.replace( |
|
|
b'</head>', f'{javascript}</head>'.encode("utf8")) |
|
|
res.init_headers() |
|
|
return res |
|
|
|
|
|
gr.routes.templates.TemplateResponse = template_response |
|
|
|
|
|
|
|
|
class JSMsgReceiver(gr.HTML): |
|
|
|
|
|
def __init__(self, **kwargs): |
|
|
super().__init__(elem_id="msg_receiver", visible=False, **kwargs) |
|
|
|
|
|
def postprocess(self, y): |
|
|
if y: |
|
|
y = f"<p>{json.dumps(y)}</p>" |
|
|
return super().postprocess(y) |
|
|
|
|
|
def get_block_name(self) -> str: |
|
|
return "html" |
|
|
|
|
|
|
|
|
|
|
|
if __name__ == "__main__": |
|
|
|
|
|
parser = argparse.ArgumentParser() |
|
|
parser.add_argument("--share", action="store_true", default=False, help="share gradio app") |
|
|
parser.add_argument("--port", type=int, default=7860, help="gradio server port") |
|
|
parser.add_argument("--max-gen", type=int, default=1024, help="max") |
|
|
opt = parser.parse_args() |
|
|
|
|
|
|
|
|
providers = ['CPUExecutionProvider'] |
|
|
|
|
|
|
|
|
|
|
|
app = gr.Blocks() |
|
|
with app: |
|
|
gr.Markdown("<h1 style='text-align: center; margin-bottom: 1rem'>Allegro Music Transformer</h1>") |
|
|
gr.Markdown("\n\n" |
|
|
"Full-attention multi-instrumental music transformer featuring asymmetrical encoding with octo-velocity, and chords counters tokens, optimized for speed and performance\n\n" |
|
|
"Check out [Allegro Music Transformer](https://github.com/asigalov61/Allegro-Music-Transformer) on GitHub!\n\n" |
|
|
"[Open In Colab]" |
|
|
"(https://colab.research.google.com/github/asigalov61/Allegro-Music-Transformer/blob/main/Allegro_Music_Transformer_Composer.ipynb)" |
|
|
" for faster execution and endless generation" |
|
|
) |
|
|
|
|
|
js_msg = JSMsgReceiver() |
|
|
tab_select = gr.Variable(value=0) |
|
|
|
|
|
app.queue(2).launch(server_port=opt.port, share=opt.share, inbrowser=True) |
