Dual model templates (#263)

* Dual-mode templates

* Fixed generate-all so that works

* Fix ability to specify Claude model
This commit is contained in:
cybermaggedon 2025-01-11 11:40:42 +00:00 committed by GitHub
parent 1bb1112569
commit cec9e29222
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
24 changed files with 725 additions and 360 deletions

View file

@ -61,59 +61,5 @@ local prompts = import "prompts/mixtral.jsonnet";
},
"text-completion-rag" +: {
create:: function(engine)
local cfgVol = engine.secretVolume(
"vertexai-creds",
"./vertexai",
{
"private.json": importstr "vertexai/private.json",
}
);
local container =
engine.container("text-completion-rag")
.with_image(images.trustgraph)
.with_command([
"text-completion-vertexai",
"-p",
url.pulsar,
"-k",
$["vertexai-private-key"],
"-r",
$["vertexai-region"],
"-x",
std.toString($["vertexai-max-output-tokens"]),
"-t",
"%0.3f" % $["vertexai-temperature"],
"-m",
$["vertexai-model"],
"-i",
"non-persistent://tg/request/text-completion-rag",
"-o",
"non-persistent://tg/response/text-completion-rag",
])
.with_limits("0.5", "256M")
.with_reservations("0.1", "256M")
.with_volume_mount(cfgVol, "/vertexai");
local containerSet = engine.containers(
"text-completion-rag", [ container ]
);
local service =
engine.internalService(containerSet)
.with_port(8000, 8000, "metrics");
engine.resources([
cfgVol,
containerSet,
service,
])
}
} + prompts