mirror of
https://github.com/trustgraph-ai/trustgraph.git
synced 2026-07-03 15:01:00 +02:00
74cc8a4685
git-subtree-dir: ai-context/trustgraph-templates git-subtree-split: 42a5fd1b678f32be378062e30451e2052ccb95dd
62 lines
1.9 KiB
Jsonnet
62 lines
1.9 KiB
Jsonnet
local base = import "base/base.jsonnet";
|
|
local images = import "values/images.jsonnet";
|
|
local url = import "values/url.jsonnet";
|
|
local prompts = import "prompts/mixtral.jsonnet";
|
|
|
|
{
|
|
|
|
with:: function(key, value)
|
|
self + {
|
|
["tgi-rag-" + key]:: value,
|
|
},
|
|
|
|
"tgi-rag-max-output-tokens":: 1024,
|
|
"tgi-rag-temperature":: 0.0,
|
|
|
|
"text-completion-rag" +: {
|
|
|
|
create:: function(engine)
|
|
|
|
local envSecrets = engine.envSecrets("tgi-credentials")
|
|
.with_env_var("TGI_BASE_URL", "tgi-url");
|
|
|
|
local containerRag =
|
|
engine.container("text-completion-rag")
|
|
.with_image(images.trustgraph_flow)
|
|
.with_command([
|
|
"text-completion-tgi",
|
|
"-p",
|
|
url.pulsar,
|
|
"--id",
|
|
"text-completion-rag",
|
|
"--concurrency",
|
|
std.toString($["text-completion-rag-concurrency"]),
|
|
"-x",
|
|
std.toString($["tgi-rag-max-output-tokens"]),
|
|
"-t",
|
|
"%0.3f" % $["tgi-rag-temperature"],
|
|
"--log-level",
|
|
$["log-level"],
|
|
])
|
|
.with_env_var_secrets(envSecrets)
|
|
.with_limits("0.5", "128M")
|
|
.with_reservations("0.1", "128M");
|
|
|
|
local containerSetRag = engine.containers(
|
|
"text-completion-rag", [ containerRag ]
|
|
);
|
|
|
|
local serviceRag =
|
|
engine.internalService(containerSetRag)
|
|
.with_port(8000, 8000, "metrics");
|
|
|
|
engine.resources([
|
|
envSecrets,
|
|
containerSetRag,
|
|
serviceRag,
|
|
])
|
|
|
|
},
|
|
|
|
} + prompts
|
|
|