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
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
|
{ lib
, fetchFromGitHub
, buildPythonPackage
, substituteAll
, cudaSupport ? false
# runtime
, ffmpeg-headless
# propagates
, numpy
, torch
, torchWithCuda
, tqdm
, more-itertools
, transformers
, numba
, openai-triton
, scipy
, tiktoken
# tests
, pytestCheckHook
}:
buildPythonPackage rec {
pname = "whisper";
version = "20230918";
format = "setuptools";
src = fetchFromGitHub {
owner = "openai";
repo = pname;
rev = "refs/tags/v${version}";
hash = "sha256-wBAanFVEIIzTcoX40P9eI26UdEu0SC/xuife/zi2Xho=";
};
patches = [
(substituteAll {
src = ./ffmpeg-path.patch;
ffmpeg = ffmpeg-headless;
})
];
propagatedBuildInputs = [
numpy
tqdm
more-itertools
transformers
numba
scipy
tiktoken
] ++ lib.optionals (!cudaSupport) [
torch
] ++ lib.optionals (cudaSupport) [
openai-triton
torchWithCuda
];
postPatch = ''
substituteInPlace requirements.txt \
--replace "tiktoken==0.3.3" "tiktoken>=0.3.3"
''
# openai-triton is only needed for CUDA support.
# triton needs CUDA to be build.
# -> by making it optional, we can build whisper without unfree packages enabled
+ lib.optionalString (!cudaSupport) ''
sed -i '/if sys.platform.startswith("linux") and platform.machine() == "x86_64":/{N;d}' setup.py
'';
preCheck = ''
export HOME=$TMPDIR
'';
nativeCheckInputs = [
pytestCheckHook
];
disabledTests = [
# requires network access to download models
"test_transcribe"
# requires NVIDIA drivers
"test_dtw_cuda_equivalence"
"test_median_filter_equivalence"
];
meta = with lib; {
changelog = "https://github.com/openai/whisper/blob/v$[version}/CHANGELOG.md";
description = "General-purpose speech recognition model";
homepage = "https://github.com/openai/whisper";
license = licenses.mit;
maintainers = with maintainers; [ hexa MayNiklas ];
};
}
|