tjh.dev
1{ lib
2, buildPythonPackage
3, hatch-fancy-pypi-readme
4, hatch-vcs
5, hatchling
6, pytestCheckHook
7, pythonOlder
8, pythonRelaxDepsHook
9, accelerate
10, bentoml
11, bitsandbytes
12, click
13, datasets
14, docker
15, einops
16, fairscale
17, flax
18, ghapi
19, hypothesis
20, ipython
21, jax
22, jaxlib
23, jupyter
24, jupytext
25, keras
26, nbformat
27, notebook
28, openai
29, openllm-client
30, openllm-core
31, optimum
32, peft
33, pytest-mock
34, pytest-randomly
35, pytest-rerunfailures
36, pytest-xdist
37, ray
38, safetensors
39, scipy
40, sentencepiece
41, soundfile
42, syrupy
43, tabulate
44, tensorflow
45, tiktoken
46, transformers
47, openai-triton
48, xformers
49}:
50
51buildPythonPackage rec {
52 inherit (openllm-core) src version;
53 pname = "openllm";
54 pyproject = true;
55
56 disabled = pythonOlder "3.8";
57
58 sourceRoot = "source/openllm-python";
59
60 nativeBuildInputs = [
61 hatch-fancy-pypi-readme
62 hatch-vcs
63 hatchling
64 pythonRelaxDepsHook
65 ];
66
67 pythonRemoveDeps = [
68 # remove cuda-python as it has an unfree license
69 "cuda-python"
70 ];
71
72 propagatedBuildInputs = [
73 accelerate
74 bentoml
75 bitsandbytes
76 click
77 ghapi
78 openllm-client
79 openllm-core
80 optimum
81 safetensors
82 tabulate
83 transformers
84 ] ++ bentoml.optional-dependencies.io
85 ++ tabulate.optional-dependencies.widechars
86 ++ transformers.optional-dependencies.tokenizers
87 ++ transformers.optional-dependencies.torch;
88
89 passthru.optional-dependencies = {
90 agents = [
91 # diffusers
92 soundfile
93 transformers
94 ] ++ transformers.optional-dependencies.agents;
95 baichuan = [
96 # cpm-kernels
97 sentencepiece
98 ];
99 chatglm = [
100 # cpm-kernels
101 sentencepiece
102 ];
103 falcon = [
104 einops
105 xformers
106 ];
107 fine-tune = [
108 accelerate
109 bitsandbytes
110 datasets
111 peft
112 # trl
113 ];
114 flan-t5 = [
115 flax
116 jax
117 jaxlib
118 keras
119 tensorflow
120 ];
121 ggml = [
122 # ctransformers
123 ];
124 gptq = [
125 # auto-gptq
126 optimum
127 ]; # ++ autogptq.optional-dependencies.triton;
128 grpc = [
129 openllm-client
130 ] ++ openllm-client.optional-dependencies.grpc;
131 llama = [
132 fairscale
133 sentencepiece
134 scipy
135 ];
136 mpt = [
137 einops
138 openai-triton
139 ];
140 openai = [
141 openai
142 tiktoken
143 ] ++ openai.optional-dependencies.embeddings;
144 opt = [
145 flax
146 jax
147 jaxlib
148 keras
149 tensorflow
150 ];
151 playground = [
152 ipython
153 jupyter
154 jupytext
155 nbformat
156 notebook
157 ];
158 starcoder = [
159 bitsandbytes
160 ];
161 vllm = [
162 ray
163 # vllm
164 ];
165 full = with passthru.optional-dependencies; (
166 agents ++ baichuan ++ chatglm ++ falcon ++ fine-tune ++ flan-t5 ++ ggml ++ gptq ++ llama ++ mpt ++ openai ++ opt ++ playground ++ starcoder ++ vllm
167 );
168 all = passthru.optional-dependencies.full;
169 };
170
171 nativeCheckInputs = [
172 docker
173 hypothesis
174 pytest-mock
175 pytest-randomly
176 pytest-rerunfailures
177 pytest-xdist
178 pytestCheckHook
179 syrupy
180 ];
181
182 preCheck = ''
183 export HOME=$TMPDIR
184 # skip GPUs test on CI
185 export GITHUB_ACTIONS=1
186 # disable hypothesis' deadline
187 export CI=1
188 '';
189
190 disabledTests = [
191 # these tests access to huggingface.co
192 "test_opt_125m"
193 "test_opt_125m"
194 "test_flan_t5"
195 "test_flan_t5"
196 ];
197
198 pythonImportsCheck = [ "openllm" ];
199
200 meta = with lib; {
201 description = "Operating LLMs in production";
202 homepage = "https://github.com/bentoml/OpenLLM/tree/main/openllm-python";
203 changelog = "https://github.com/bentoml/OpenLLM/blob/${src.rev}/CHANGELOG.md";
204 license = licenses.asl20;
205 maintainers = with maintainers; [ happysalada natsukium ];
206 };
207}