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
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
|
{
stdenv,
lib,
buildPythonPackage,
fetchFromGitHub,
fetchpatch,
pythonAtLeast,
# buildInputs
llvmPackages,
# build-system
setuptools,
# dependencies
huggingface-hub,
numpy,
packaging,
psutil,
pyyaml,
safetensors,
torch,
# tests
addBinToPathHook,
evaluate,
parameterized,
pytestCheckHook,
transformers,
config,
cudatoolkit,
writableTmpDirAsHomeHook,
}:
buildPythonPackage rec {
pname = "accelerate";
version = "1.5.2";
pyproject = true;
src = fetchFromGitHub {
owner = "huggingface";
repo = "accelerate";
tag = "v${version}";
hash = "sha256-J4eDm/PcyKK3256l6CAWUj4AWTB6neTKgxbBmul0BPE=";
};
patches = [
# Fix tests on darwin: https://github.com/huggingface/accelerate/pull/3464
(fetchpatch {
url = "https://github.com/huggingface/accelerate/commit/8b31a2fe2c6d0246fff9885fb1f8456fb560abc7.patch";
hash = "sha256-Ek9Ou4Y/H1jt3qanf2g3HowBoTsN/bn4yV9O3ogcXMo=";
})
];
buildInputs = [ llvmPackages.openmp ];
build-system = [ setuptools ];
dependencies = [
huggingface-hub
numpy
packaging
psutil
pyyaml
safetensors
torch
];
nativeCheckInputs = [
addBinToPathHook
evaluate
parameterized
pytestCheckHook
transformers
writableTmpDirAsHomeHook
];
preCheck = lib.optionalString config.cudaSupport ''
export TRITON_PTXAS_PATH="${lib.getExe' cudatoolkit "ptxas"}"
'';
pytestFlagsArray = [ "tests" ];
disabledTests =
[
# try to download data:
"FeatureExamplesTests"
"test_infer_auto_device_map_on_t0pp"
# require socket communication
"test_explicit_dtypes"
"test_gated"
"test_invalid_model_name"
"test_invalid_model_name_transformers"
"test_no_metadata"
"test_no_split_modules"
"test_remote_code"
"test_transformers_model"
# nondeterministic, tests GC behaviour by thresholding global ram usage
"test_free_memory_dereferences_prepared_components"
# set the environment variable, CC, which conflicts with standard environment
"test_patch_environment_key_exists"
]
++ lib.optionals ((pythonAtLeast "3.13") || (torch.rocmSupport or false)) [
# RuntimeError: Dynamo is not supported on Python 3.13+
# OR torch.compile tests broken on torch 2.5 + rocm
"test_can_unwrap_distributed_compiled_model_keep_torch_compile"
"test_can_unwrap_distributed_compiled_model_remove_torch_compile"
"test_convert_to_fp32"
"test_send_to_device_compiles"
]
++ lib.optionals (stdenv.hostPlatform.isLinux && stdenv.hostPlatform.isAarch64) [
# usual aarch64-linux RuntimeError: DataLoader worker (pid(s) <...>) exited unexpectedly
"CheckpointTest"
# TypeError: unsupported operand type(s) for /: 'NoneType' and 'int' (it seems cpuinfo doesn't work here)
"test_mpi_multicpu_config_cmd"
]
++ lib.optionals (!config.cudaSupport) [
# requires ptxas from cudatoolkit, which is unfree
"test_dynamo_extract_model"
]
++ lib.optionals stdenv.hostPlatform.isDarwin [
# RuntimeError: 'accelerate-launch /nix/store/a7vhm7b74a7bmxc35j26s9iy1zfaqjs...
"test_accelerate_test"
"test_init_trackers"
"test_init_trackers"
"test_log"
"test_log_with_tensor"
# After enabling MPS in pytorch, these tests started failing
"test_accelerated_optimizer_step_was_skipped"
"test_auto_wrap_policy"
"test_autocast_kwargs"
"test_automatic_loading"
"test_backward_prefetch"
"test_can_resume_training"
"test_can_resume_training_checkpoints_relative_path"
"test_can_resume_training_with_folder"
"test_can_unwrap_model_fp16"
"test_checkpoint_deletion"
"test_cpu_offload"
"test_cpu_ram_efficient_loading"
"test_grad_scaler_kwargs"
"test_invalid_registration"
"test_map_location"
"test_mixed_precision"
"test_mixed_precision_buffer_autocast_override"
"test_project_dir"
"test_project_dir_with_config"
"test_sharding_strategy"
"test_state_dict_type"
"test_with_save_limit"
"test_with_scheduler"
]
++ lib.optionals (stdenv.hostPlatform.isDarwin && stdenv.hostPlatform.isx86_64) [
# RuntimeError: torch_shm_manager: execl failed: Permission denied
"CheckpointTest"
];
disabledTestPaths = lib.optionals (!(stdenv.hostPlatform.isLinux && stdenv.hostPlatform.isx86_64)) [
# numerous instances of torch.multiprocessing.spawn.ProcessRaisedException:
"tests/test_cpu.py"
"tests/test_grad_sync.py"
"tests/test_metrics.py"
"tests/test_scheduler.py"
];
pythonImportsCheck = [ "accelerate" ];
__darwinAllowLocalNetworking = true;
meta = {
homepage = "https://huggingface.co/docs/accelerate";
description = "Simple way to train and use PyTorch models with multi-GPU, TPU, mixed-precision";
changelog = "https://github.com/huggingface/accelerate/releases/tag/v${version}";
license = lib.licenses.asl20;
maintainers = with lib.maintainers; [ bcdarwin ];
mainProgram = "accelerate";
};
}
|
