blob: 7cebc6fff9387ae0b1dce0eec1348e06cda9e00e (
plain) (
blame)
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
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
|
{ self
, lib
, testers
, fetchzip
, fetchurl
, writers
, symlinkJoin
, jq
}:
let
common-config = { config, ... }: {
imports = [ ./module.nix ];
services.local-ai = {
enable = true;
package = self;
threads = config.virtualisation.cores;
};
};
inherit (self.lib) genModels;
in
{
version = testers.testVersion {
package = self;
version = "v" + self.version;
command = "local-ai --help";
};
health = testers.runNixOSTest ({ config, ... }: {
name = self.name + "-health";
nodes.machine = common-config;
testScript =
let
port = "8080";
in
''
machine.wait_for_open_port(${port})
machine.succeed("curl -f http://localhost:${port}/readyz")
'';
});
# https://localai.io/features/embeddings/#bert-embeddings
bert =
let
model = "embedding";
model-configs.${model} = {
# Note: q4_0 and q4_1 models can not be loaded
parameters.model = fetchurl {
url = "https://huggingface.co/skeskinen/ggml/resolve/main/all-MiniLM-L6-v2/ggml-model-f16.bin";
sha256 = "9c195b2453a4fef60a4f6be3a88a39211366214df6498a4fe4885c9e22314f50";
};
backend = "bert-embeddings";
embeddings = true;
};
models = genModels model-configs;
requests.request = {
inherit model;
input = "Your text string goes here";
};
in
testers.runNixOSTest {
name = self.name + "-bert";
nodes.machine = {
imports = [ common-config ];
virtualisation.cores = 2;
virtualisation.memorySize = 2048;
services.local-ai.models = models;
};
passthru = { inherit models requests; };
testScript =
let
port = "8080";
in
''
machine.wait_for_open_port(${port})
machine.succeed("curl -f http://localhost:${port}/readyz")
machine.succeed("curl -f http://localhost:${port}/v1/models --output models.json")
machine.succeed("${jq}/bin/jq --exit-status 'debug | .data[].id == \"${model}\"' models.json")
machine.succeed("curl -f http://localhost:${port}/embeddings --json @${writers.writeJSON "request.json" requests.request} --output embeddings.json")
machine.succeed("${jq}/bin/jq --exit-status 'debug | .model == \"${model}\"' embeddings.json")
'';
};
} // lib.optionalAttrs (!self.features.with_cublas && !self.features.with_clblas) {
# https://localai.io/docs/getting-started/manual/
llama =
let
model = "gpt-3.5-turbo";
# https://localai.io/advanced/#full-config-model-file-reference
model-configs.${model} = rec {
context_size = 8192;
parameters = {
# https://huggingface.co/lmstudio-community/Meta-Llama-3-8B-Instruct-GGUF
# https://ai.meta.com/blog/meta-llama-3/
model = fetchurl {
url = "https://huggingface.co/lmstudio-community/Meta-Llama-3-8B-Instruct-GGUF/resolve/main/Meta-Llama-3-8B-Instruct-Q4_K_M.gguf";
sha256 = "ab9e4eec7e80892fd78f74d9a15d0299f1e22121cea44efd68a7a02a3fe9a1da";
};
# defaults from:
# https://deepinfra.com/meta-llama/Meta-Llama-3-8B-Instruct
temperature = 0.7;
top_p = 0.9;
top_k = 0;
# following parameter leads to outputs like: !!!!!!!!!!!!!!!!!!!
#repeat_penalty = 1;
presence_penalty = 0;
frequency_penalty = 0;
max_tokens = 100;
};
stopwords = [ "<|eot_id|>" ];
template = {
# Templates implement following specifications
# https://github.com/meta-llama/llama3/tree/main?tab=readme-ov-file#instruction-tuned-models
# ... and are insprired by:
# https://github.com/mudler/LocalAI/blob/master/embedded/models/llama3-instruct.yaml
#
# The rules for template evaluateion are defined here:
# https://pkg.go.dev/text/template
chat_message = ''
<|start_header_id|>{{.RoleName}}<|end_header_id|>
{{.Content}}${builtins.head stopwords}'';
chat = "<|begin_of_text|>{{.Input}}<|start_header_id|>assistant<|end_header_id|>";
};
};
models = genModels model-configs;
requests = {
# https://localai.io/features/text-generation/#chat-completions
chat-completions = {
inherit model;
messages = [{ role = "user"; content = "1 + 2 = ?"; }];
};
# https://localai.io/features/text-generation/#edit-completions
edit-completions = {
inherit model;
instruction = "rephrase";
input = "Black cat jumped out of the window";
max_tokens = 50;
};
# https://localai.io/features/text-generation/#completions
completions = {
inherit model;
prompt = "A long time ago in a galaxy far, far away";
};
};
in
testers.runNixOSTest {
name = self.name + "-llama";
nodes.machine = {
imports = [ common-config ];
virtualisation.cores = 4;
virtualisation.memorySize = 8192;
services.local-ai.models = models;
};
passthru = { inherit models requests; };
testScript =
let
port = "8080";
in
''
machine.wait_for_open_port(${port})
machine.succeed("curl -f http://localhost:${port}/readyz")
machine.succeed("curl -f http://localhost:${port}/v1/models --output models.json")
machine.succeed("${jq}/bin/jq --exit-status 'debug | .data[].id == \"${model}\"' models.json")
machine.succeed("curl -f http://localhost:${port}/v1/chat/completions --json @${writers.writeJSON "request-chat-completions.json" requests.chat-completions} --output chat-completions.json")
machine.succeed("${jq}/bin/jq --exit-status 'debug | .object == \"chat.completion\"' chat-completions.json")
machine.succeed("${jq}/bin/jq --exit-status 'debug | .choices | first.message.content | tonumber == 3' chat-completions.json")
machine.succeed("curl -f http://localhost:${port}/v1/edits --json @${writers.writeJSON "request-edit-completions.json" requests.edit-completions} --output edit-completions.json")
machine.succeed("${jq}/bin/jq --exit-status 'debug | .object == \"edit\"' edit-completions.json")
machine.succeed("${jq}/bin/jq --exit-status '.usage.completion_tokens | debug == ${toString requests.edit-completions.max_tokens}' edit-completions.json")
machine.succeed("curl -f http://localhost:${port}/v1/completions --json @${writers.writeJSON "request-completions.json" requests.completions} --output completions.json")
machine.succeed("${jq}/bin/jq --exit-status 'debug | .object ==\"text_completion\"' completions.json")
machine.succeed("${jq}/bin/jq --exit-status '.usage.completion_tokens | debug == ${toString model-configs.${model}.parameters.max_tokens}' completions.json")
'';
};
} // lib.optionalAttrs (self.features.with_tts && !self.features.with_cublas && !self.features.with_clblas) {
# https://localai.io/features/text-to-audio/#piper
tts =
let
model-stt = "whisper-en";
model-configs.${model-stt} = {
backend = "whisper";
parameters.model = fetchurl {
url = "https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-tiny.en-q5_1.bin";
hash = "sha256-x3xXZvHO8JtrfUfyG1Rsvd1BV4hrO11tT3CekeZsfCs=";
};
};
model-tts = "piper-en";
model-configs.${model-tts} = {
backend = "piper";
parameters.model = "en-us-danny-low.onnx";
};
models =
let
models = genModels model-configs;
in
symlinkJoin {
inherit (models) name;
paths = [
models
(fetchzip {
url = "https://github.com/rhasspy/piper/releases/download/v0.0.2/voice-en-us-danny-low.tar.gz";
hash = "sha256-5wf+6H5HeQY0qgdqnAG1vSqtjIFM9lXH53OgouuPm0M=";
stripRoot = false;
})
];
};
requests.request = {
model = model-tts;
input = "Hello, how are you?";
};
in
testers.runNixOSTest {
name = self.name + "-tts";
nodes.machine = {
imports = [ common-config ];
virtualisation.cores = 2;
services.local-ai.models = models;
};
passthru = { inherit models requests; };
testScript =
let
port = "8080";
in
''
machine.wait_for_open_port(${port})
machine.succeed("curl -f http://localhost:${port}/readyz")
machine.succeed("curl -f http://localhost:${port}/v1/models --output models.json")
machine.succeed("${jq}/bin/jq --exit-status 'debug' models.json")
machine.succeed("curl -f http://localhost:${port}/tts --json @${writers.writeJSON "request.json" requests.request} --output out.wav")
machine.succeed("curl -f http://localhost:${port}/v1/audio/transcriptions --header 'Content-Type: multipart/form-data' --form file=@out.wav --form model=${model-stt} --output transcription.json")
machine.succeed("${jq}/bin/jq --exit-status 'debug | .segments | first.text == \"${requests.request.input}\"' transcription.json")
'';
};
}
|
