mrfakename commited on
Commit
f97f1bb
1 Parent(s): f117179

Rename app.py to musiclib.py

Browse files
Files changed (2) hide show
  1. app.py +0 -0
  2. musiclib.py +67 -0
app.py DELETED
File without changes
musiclib.py ADDED
@@ -0,0 +1,67 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # apache 2.0 license, modified by mrfakename, from https://github.com/BlinkDL/ChatRWKV/tree/main/music
2
+
3
+ import os, sys
4
+ import numpy as np
5
+ from cached_path import cached_path
6
+ np.set_printoptions(precision=4, suppress=True, linewidth=200)
7
+
8
+ os.environ['RWKV_JIT_ON'] = '1' #### set these before import RWKV
9
+ os.environ["RWKV_CUDA_ON"] = '0'
10
+ os.environ["RWKV_RESCALE_LAYER"] = '999' # must set this for RWKV-music models and "pip install rwkv --upgrade" to v0.8.12+
11
+
12
+ from rwkv.model import RWKV
13
+ from rwkv.utils import PIPELINE
14
+
15
+ MODEL_FILE = str(cached_path('hf://BlinkDL/rwkv-4-music/RWKV-4-MIDI-120M-v1-20230714-ctx4096.pth'))
16
+
17
+ ABC_MODE = ('-ABC-' in MODEL_FILE)
18
+ MIDI_MODE = ('-MIDI-' in MODEL_FILE)
19
+
20
+ model = RWKV(model=MODEL_FILE, strategy='mps fp32')
21
+ pipeline = PIPELINE(model, "tokenizer-midi.json")
22
+
23
+ tokenizer = pipeline
24
+ EOS_ID = 0
25
+ TOKEN_SEP = ' '
26
+
27
+ def musicgen(ccc='<pad>', piano_only=False):
28
+ # ccc = '<pad>'
29
+ ccc_output = '<start>'
30
+ # ccc = "v:5b:3 v:5b:2 t125 t125 t125 t106 pi:43:5 t24 pi:4a:7 t15 pi:4f:7 t17 pi:56:7 t18 pi:54:7 t125 t49 pi:51:7 t117 pi:4d:7 t125 t125 t111 pi:37:7 t14 pi:3e:6 t15 pi:43:6 t12 pi:4a:7 t17 pi:48:7 t125 t60 pi:45:7 t121 pi:41:7 t125 t117 s:46:5 s:52:5 f:46:5 f:52:5 t121 s:45:5 s:46:0 s:51:5 s:52:0 f:45:5 f:46:0 f:51:5 f:52:0 t121 s:41:5 s:45:0 s:4d:5 s:51:0 f:41:5 f:45:0 f:4d:5 f:51:0 t102 pi:37:0 pi:3e:0 pi:41:0 pi:43:0 pi:45:0 pi:48:0 pi:4a:0 pi:4d:0 pi:4f:0 pi:51:0 pi:54:0 pi:56:0 t19 s:3e:5 s:41:0 s:4a:5 s:4d:0 f:3e:5 f:41:0 f:4a:5 f:4d:0 t121 v:3a:5 t121 v:39:7 t15 v:3a:0 t106 v:35:8 t10 v:39:0 t111 v:30:8 v:35:0 t125 t117 v:32:8 t10 v:30:0 t125 t125 t103 v:5b:0 v:5b:0 t9 pi:4a:7"
31
+ # ccc = '<pad> ' + ccc
32
+ # ccc_output = '<start> pi:4a:7'
33
+ output = ''
34
+
35
+ output += (ccc_output)
36
+
37
+ occurrence = {}
38
+ state = None
39
+ for i in range(4096): # only trained with ctx4096 (will be longer soon)
40
+ if i == 0:
41
+ out, state = model.forward(tokenizer.encode(ccc), state)
42
+ else:
43
+ out, state = model.forward([token], state)
44
+
45
+ if MIDI_MODE: # seems only required for MIDI mode
46
+ for n in occurrence:
47
+ out[n] -= (0 + occurrence[n] * 0.5)
48
+
49
+ out[0] += (i - 2000) / 500 # not too short, not too long
50
+ out[127] -= 1 # avoid "t125"
51
+ if piano_only:
52
+ out[128:12416] -= 1e10
53
+ out[13952:20096] -= 1e10
54
+ token = pipeline.sample_logits(out, temperature=1.0, top_k=8, top_p=0.8)
55
+ if token == EOS_ID: break
56
+
57
+ if MIDI_MODE: # seems only required for MIDI mode
58
+ for n in occurrence: occurrence[n] *= 0.997 #### decay repetition penalty
59
+ if token >= 128 or token == 127:
60
+ occurrence[token] = 1 + (occurrence[token] if token in occurrence else 0)
61
+ else:
62
+ occurrence[token] = 0.3 + (occurrence[token] if token in occurrence else 0)
63
+
64
+ output += (TOKEN_SEP + tokenizer.decode([token]))
65
+
66
+ output += (' <end>')
67
+ return output