MusicGenLargeRunPod/Note.java at main · redLiquidator/MusicGenLargeRunPod · GitHub

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
Runpod Mypod : https://console.runpod.io/pods?id=l9lqa2m7v0ctwa

<Mar 21>
Jupyter Lab 클릭 (Port 8888) -> Jupyter 들어가면: 새 Notebook 생성, MusicGen 설치, 바로 음악 생성

1.
  !pip install audiocraft

2.
from audiocraft.models import MusicGen

model = MusicGen.get_pretrained("facebook/musicgen-large")
model.set_generation_params(duration=15)

3.
descriptions = [
    "warm bossa nova, acoustic guitar, soft jazz chords, groovy rhythm, emotional melody, cafe atmosphere"
]

wav = model.generate(descriptions)

import torchaudio
torchaudio.save("output.wav", wav[0].cpu(), 32000)

**********************************************************************
1. 한 번에 3개 뽑기
descriptions = [
    "bossa nova track with a dominant melodic theme, acoustic guitar lead carrying a clear and catchy melody, emotional progression, jazz harmony supporting the melody, soft groove, musical storytelling, non-repetitive structure"
]

model.set_generation_params(duration=20, temperature=1.1)
wav = model.generate(descriptions * 3)

👉 핵심: * 3 → 3개 생성, temperature 1.1 → 멜로디 다양성 증가
**********************************************************************
<Mar22>
<Network Volume>
기본 설치는 보통 pod의 임시 컨테이너 쪽에 들어가서 terminate 하면 같이 날아간다.
그런데 /workspace가 network volume에 연결되어 있으면,

/workspace/venv 안의 패키지
notebook
wav 파일
requirements.txt

저장가능