mirror of
https://github.com/trustgraph-ai/trustgraph.git
synced 2026-06-30 00:49:38 +02:00
Feature/environment var creds (#116)
- Change templates to interpolate environment variables in docker compose - Change templates to invoke secrets for environment variable credentials in K8s configuration - Update LLMs to pull in credentials from environment variables if not specified
This commit is contained in:
cybermaggedon
GitHub
parent
43756d872b
commit
86288339cf
20 changed files with 327 additions and 271 deletions
|
|
@ -5,7 +5,6 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
|
||||
{
|
||||
|
||||
"azure-openai-token":: "${AZURE_OPENAI_TOKEN}",
|
||||
"azure-openai-model":: "GPT-3.5-Turbo",
|
||||
"azure-openai-max-output-tokens":: 4192,
|
||||
"azure-openai-temperature":: 0.0,
|
||||
|
|
@ -14,6 +13,9 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
|
||||
create:: function(engine)
|
||||
|
||||
local envSecrets = engine.envSecrets("azure-openai-credentials")
|
||||
.with_env_var("AZURE_TOKEN", "azure-token");
|
||||
|
||||
local container =
|
||||
engine.container("text-completion")
|
||||
.with_image(images.trustgraph)
|
||||
|
|
@ -21,8 +23,6 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
"text-completion-azure-openai",
|
||||
"-p",
|
||||
url.pulsar,
|
||||
"-k",
|
||||
$["azure-openai-token"],
|
||||
"-m",
|
||||
$["azure-openai-model"],
|
||||
"-x",
|
||||
|
|
@ -30,39 +30,17 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
"-t",
|
||||
std.toString($["azure-openai-temperature"]),
|
||||
])
|
||||
.with_env_var_secrets(envSecrets)
|
||||
.with_limits("0.5", "128M")
|
||||
.with_reservations("0.1", "128M");
|
||||
|
||||
local containerSet = engine.containers(
|
||||
"text-completion", [ container ]
|
||||
);
|
||||
|
||||
local service =
|
||||
engine.internalService(containerSet)
|
||||
.with_port(8000, 8000, "metrics");
|
||||
|
||||
engine.resources([
|
||||
containerSet,
|
||||
service,
|
||||
])
|
||||
|
||||
},
|
||||
|
||||
"text-completion-rag" +: {
|
||||
|
||||
create:: function(engine)
|
||||
|
||||
local container =
|
||||
local containerRag =
|
||||
engine.container("text-completion-rag")
|
||||
.with_image(images.trustgraph)
|
||||
.with_command([
|
||||
"text-completion-azure",
|
||||
"-p",
|
||||
url.pulsar,
|
||||
"-k",
|
||||
$["azure-openai-token"],
|
||||
"-e",
|
||||
$["azure-openai-model"],
|
||||
"-x",
|
||||
std.toString($["azure-openai-max-output-tokens"]),
|
||||
"-t",
|
||||
|
|
@ -72,24 +50,35 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
"-o",
|
||||
"non-persistent://tg/response/text-completion-rag-response",
|
||||
])
|
||||
.with_env_var_secrets(envSecrets)
|
||||
.with_limits("0.5", "128M")
|
||||
.with_reservations("0.1", "128M");
|
||||
|
||||
local containerSet = engine.containers(
|
||||
"text-completion-rag", [ container ]
|
||||
"text-completion", [ container ]
|
||||
);
|
||||
|
||||
local containerSetRag = engine.containers(
|
||||
"text-completion-rag", [ containerRag ]
|
||||
);
|
||||
|
||||
local service =
|
||||
engine.internalService(containerSet)
|
||||
.with_port(8000, 8000, "metrics");
|
||||
|
||||
local serviceRag =
|
||||
engine.internalService(containerSetRag)
|
||||
.with_port(8000, 8000, "metrics");
|
||||
|
||||
engine.resources([
|
||||
envSecrets,
|
||||
containerSet,
|
||||
containerSetRag,
|
||||
service,
|
||||
serviceRag,
|
||||
])
|
||||
|
||||
|
||||
}
|
||||
},
|
||||
|
||||
} + prompts
|
||||
|
||||
|
|
|
|||
|
|
@ -5,8 +5,6 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
|
||||
{
|
||||
|
||||
"azure-token":: "${AZURE_TOKEN}",
|
||||
"azure-endpoint":: "${AZURE_ENDPOINT}",
|
||||
"azure-max-output-tokens":: 4096,
|
||||
"azure-temperature":: 0.0,
|
||||
|
||||
|
|
@ -14,6 +12,10 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
|
||||
create:: function(engine)
|
||||
|
||||
local envSecrets = engine.envSecrets("azure-credentials")
|
||||
.with_env_var("AZURE_TOKEN", "azure-token")
|
||||
.with_env_var("AZURE_ENDPOINT", "azure-endpoint");
|
||||
|
||||
local container =
|
||||
engine.container("text-completion")
|
||||
.with_image(images.trustgraph)
|
||||
|
|
@ -21,48 +23,22 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
"text-completion-azure",
|
||||
"-p",
|
||||
url.pulsar,
|
||||
"-k",
|
||||
$["azure-token"],
|
||||
"-e",
|
||||
$["azure-endpoint"],
|
||||
"-x",
|
||||
std.toString($["azure-max-output-tokens"]),
|
||||
"-t",
|
||||
std.toString($["azure-temperature"]),
|
||||
])
|
||||
.with_env_var_secrets(envSecrets)
|
||||
.with_limits("0.5", "128M")
|
||||
.with_reservations("0.1", "128M");
|
||||
|
||||
local containerSet = engine.containers(
|
||||
"text-completion", [ container ]
|
||||
);
|
||||
|
||||
local service =
|
||||
engine.internalService(containerSet)
|
||||
.with_port(8000, 8000, "metrics");
|
||||
|
||||
engine.resources([
|
||||
containerSet,
|
||||
service,
|
||||
])
|
||||
|
||||
},
|
||||
|
||||
"text-completion-rag" +: {
|
||||
|
||||
create:: function(engine)
|
||||
|
||||
local container =
|
||||
local containerRag =
|
||||
engine.container("text-completion-rag")
|
||||
.with_image(images.trustgraph)
|
||||
.with_command([
|
||||
"text-completion-azure",
|
||||
"-p",
|
||||
url.pulsar,
|
||||
"-k",
|
||||
$["azure-token"],
|
||||
"-e",
|
||||
$["azure-endpoint"],
|
||||
"-x",
|
||||
std.toString($["azure-max-output-tokens"]),
|
||||
"-t",
|
||||
|
|
@ -72,22 +48,33 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
"-o",
|
||||
"non-persistent://tg/response/text-completion-rag-response",
|
||||
])
|
||||
.with_env_var_secrets(envSecrets)
|
||||
.with_limits("0.5", "128M")
|
||||
.with_reservations("0.1", "128M");
|
||||
|
||||
local containerSet = engine.containers(
|
||||
"text-completion-rag", [ container ]
|
||||
"text-completion", [ container ]
|
||||
);
|
||||
|
||||
local containerSetRag = engine.containers(
|
||||
"text-completion-rag", [ containerRag ]
|
||||
);
|
||||
|
||||
local service =
|
||||
engine.internalService(containerSet)
|
||||
.with_port(8000, 8000, "metrics");
|
||||
|
||||
engine.resources([
|
||||
containerSet,
|
||||
service,
|
||||
])
|
||||
local serviceRag =
|
||||
engine.internalService(containerSetRag)
|
||||
.with_port(8000, 8000, "metrics");
|
||||
|
||||
engine.resources([
|
||||
envSecrets,
|
||||
containerSet,
|
||||
containerSetRag,
|
||||
service,
|
||||
serviceRag,
|
||||
])
|
||||
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -6,9 +6,6 @@ local chunker = import "chunker-recursive.jsonnet";
|
|||
|
||||
{
|
||||
|
||||
"aws-id-key":: "${AWS_ID_KEY}",
|
||||
"aws-secret-key":: "${AWS_SECRET_KEY}",
|
||||
"aws-region":: "us-west-2",
|
||||
"bedrock-max-output-tokens":: 4096,
|
||||
"bedrock-temperature":: 0.0,
|
||||
"bedrock-model":: "mistral.mixtral-8x7b-instruct-v0:1",
|
||||
|
|
@ -17,6 +14,11 @@ local chunker = import "chunker-recursive.jsonnet";
|
|||
|
||||
create:: function(engine)
|
||||
|
||||
local envSecrets = engine.envSecrets("bedrock-credentials")
|
||||
.with_env_var("AWS_ID_KEY", "aws-id-key")
|
||||
.with_env_var("AWS_SECRET_KEY", "aws-secret-key")
|
||||
.with_env_var("AWS_REGION", "aws-region");
|
||||
|
||||
local container =
|
||||
engine.container("text-completion")
|
||||
.with_image(images.trustgraph)
|
||||
|
|
@ -24,12 +26,6 @@ local chunker = import "chunker-recursive.jsonnet";
|
|||
"text-completion-bedrock",
|
||||
"-p",
|
||||
url.pulsar,
|
||||
"-z",
|
||||
$["aws-id-key"],
|
||||
"-k",
|
||||
$["aws-secret-key"],
|
||||
"-r",
|
||||
$["aws-region"],
|
||||
"-x",
|
||||
std.toString($["bedrock-max-output-tokens"]),
|
||||
"-t",
|
||||
|
|
@ -37,41 +33,17 @@ local chunker = import "chunker-recursive.jsonnet";
|
|||
"-m",
|
||||
$["bedrock-model"],
|
||||
])
|
||||
.with_env_var_secrets(envSecrets)
|
||||
.with_limits("0.5", "128M")
|
||||
.with_reservations("0.1", "128M");
|
||||
|
||||
local containerSet = engine.containers(
|
||||
"text-completion", [ container ]
|
||||
);
|
||||
|
||||
local service =
|
||||
engine.internalService(containerSet)
|
||||
.with_port(8000, 8000, "metrics");
|
||||
|
||||
engine.resources([
|
||||
containerSet,
|
||||
service,
|
||||
])
|
||||
|
||||
},
|
||||
|
||||
"text-completion-rag" +: {
|
||||
|
||||
create:: function(engine)
|
||||
|
||||
local container =
|
||||
local containerRag =
|
||||
engine.container("text-completion-rag")
|
||||
.with_image(images.trustgraph)
|
||||
.with_command([
|
||||
"text-completion-bedrock",
|
||||
"-p",
|
||||
url.pulsar,
|
||||
"-z",
|
||||
$["aws-id-key"],
|
||||
"-k",
|
||||
$["aws-secret-key"],
|
||||
"-r",
|
||||
$["aws-region"],
|
||||
"-x",
|
||||
std.toString($["bedrock-max-output-tokens"]),
|
||||
"-t",
|
||||
|
|
@ -83,24 +55,35 @@ local chunker = import "chunker-recursive.jsonnet";
|
|||
"-o",
|
||||
"non-persistent://tg/response/text-completion-rag-response",
|
||||
])
|
||||
.with_env_var_secrets(envSecrets)
|
||||
.with_limits("0.5", "128M")
|
||||
.with_reservations("0.1", "128M");
|
||||
|
||||
local containerSet = engine.containers(
|
||||
"text-completion-rag", [ container ]
|
||||
"text-completion", [ container ]
|
||||
);
|
||||
|
||||
local containerSetRag = engine.containers(
|
||||
"text-completion-rag", [ containerRag ]
|
||||
);
|
||||
|
||||
local service =
|
||||
engine.internalService(containerSet)
|
||||
.with_port(8000, 8000, "metrics");
|
||||
|
||||
local serviceRag =
|
||||
engine.internalService(containerSetRag)
|
||||
.with_port(8000, 8000, "metrics");
|
||||
|
||||
engine.resources([
|
||||
envSecrets,
|
||||
containerSet,
|
||||
containerSetRag,
|
||||
service,
|
||||
serviceRag,
|
||||
])
|
||||
|
||||
|
||||
}
|
||||
},
|
||||
|
||||
} + prompts + chunker
|
||||
|
||||
|
|
|
|||
|
|
@ -5,7 +5,6 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
|
||||
{
|
||||
|
||||
"claude-key":: "${CLAUDE_KEY}",
|
||||
"claude-max-output-tokens":: 4096,
|
||||
"claude-temperature":: 0.0,
|
||||
|
||||
|
|
@ -13,6 +12,9 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
|
||||
create:: function(engine)
|
||||
|
||||
local envSecrets = engine.envSecrets("claude-credentials")
|
||||
.with_env_var("CLAUDE_KEY_TOKEN", "claude-key");
|
||||
|
||||
local container =
|
||||
engine.container("text-completion")
|
||||
.with_image(images.trustgraph)
|
||||
|
|
@ -20,44 +22,22 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
"text-completion-claude",
|
||||
"-p",
|
||||
url.pulsar,
|
||||
"-k",
|
||||
$["claude-key"],
|
||||
"-x",
|
||||
std.toString($["claude-max-output-tokens"]),
|
||||
"-t",
|
||||
std.toString($["claude-temperature"]),
|
||||
])
|
||||
.with_env_var_secrets(envSecrets)
|
||||
.with_limits("0.5", "128M")
|
||||
.with_reservations("0.1", "128M");
|
||||
|
||||
local containerSet = engine.containers(
|
||||
"text-completion", [ container ]
|
||||
);
|
||||
|
||||
local service =
|
||||
engine.internalService(containerSet)
|
||||
.with_port(8000, 8000, "metrics");
|
||||
|
||||
engine.resources([
|
||||
containerSet,
|
||||
service,
|
||||
])
|
||||
|
||||
},
|
||||
|
||||
"text-completion-rag" +: {
|
||||
|
||||
create:: function(engine)
|
||||
|
||||
local container =
|
||||
local containerRag =
|
||||
engine.container("text-completion-rag")
|
||||
.with_image(images.trustgraph)
|
||||
.with_command([
|
||||
"text-completion-claude",
|
||||
"-p",
|
||||
url.pulsar,
|
||||
"-k",
|
||||
$["claude-key"],
|
||||
"-x",
|
||||
std.toString($["claude-max-output-tokens"]),
|
||||
"-t",
|
||||
|
|
@ -67,24 +47,35 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
"-o",
|
||||
"non-persistent://tg/response/text-completion-rag-response",
|
||||
])
|
||||
.with_env_var_secrets(envSecrets)
|
||||
.with_limits("0.5", "128M")
|
||||
.with_reservations("0.1", "128M");
|
||||
|
||||
local containerSet = engine.containers(
|
||||
"text-completion-rag", [ container ]
|
||||
"text-completion", [ container ]
|
||||
);
|
||||
|
||||
local containerSetRag = engine.containers(
|
||||
"text-completion-rag", [ containerRag ]
|
||||
);
|
||||
|
||||
local service =
|
||||
engine.internalService(containerSet)
|
||||
.with_port(8000, 8000, "metrics");
|
||||
|
||||
local serviceRag =
|
||||
engine.internalService(containerSetRag)
|
||||
.with_port(8000, 8000, "metrics");
|
||||
|
||||
engine.resources([
|
||||
envSecrets,
|
||||
containerSet,
|
||||
containerSetRag,
|
||||
service,
|
||||
serviceRag,
|
||||
])
|
||||
|
||||
|
||||
}
|
||||
},
|
||||
|
||||
} + prompts
|
||||
|
||||
|
|
|
|||
|
|
@ -9,13 +9,15 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
"chunk-size":: 150,
|
||||
"chunk-overlap":: 10,
|
||||
|
||||
"cohere-key":: "${COHERE_KEY}",
|
||||
"cohere-temperature":: 0.0,
|
||||
|
||||
"text-completion" +: {
|
||||
|
||||
create:: function(engine)
|
||||
|
||||
local envSecrets = engine.envSecrets("cohere-credentials")
|
||||
.with_env_var("COHERE_KEY", "cohere-key");
|
||||
|
||||
local container =
|
||||
engine.container("text-completion")
|
||||
.with_image(images.trustgraph)
|
||||
|
|
@ -23,42 +25,19 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
"text-completion-cohere",
|
||||
"-p",
|
||||
url.pulsar,
|
||||
"-k",
|
||||
$["cohere-key"],
|
||||
"-t",
|
||||
std.toString($["cohere-temperature"]),
|
||||
])
|
||||
.with_limits("0.5", "128M")
|
||||
.with_reservations("0.1", "128M");
|
||||
|
||||
local containerSet = engine.containers(
|
||||
"text-completion", [ container ]
|
||||
);
|
||||
|
||||
local service =
|
||||
engine.internalService(containerSet)
|
||||
.with_port(8000, 8000, "metrics");
|
||||
|
||||
engine.resources([
|
||||
containerSet,
|
||||
service,
|
||||
])
|
||||
|
||||
},
|
||||
|
||||
"text-completion-rag" +: {
|
||||
|
||||
create:: function(engine)
|
||||
|
||||
local container =
|
||||
local containerRag =
|
||||
engine.container("text-completion-rag")
|
||||
.with_image(images.trustgraph)
|
||||
.with_command([
|
||||
"text-completion-cohere",
|
||||
"-p",
|
||||
url.pulsar,
|
||||
"-k",
|
||||
$["cohere-key"],
|
||||
"-t",
|
||||
std.toString($["cohere-temperature"]),
|
||||
"-i",
|
||||
|
|
@ -70,20 +49,30 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
.with_reservations("0.1", "128M");
|
||||
|
||||
local containerSet = engine.containers(
|
||||
"text-completion-rag", [ container ]
|
||||
"text-completion", [ container ]
|
||||
);
|
||||
|
||||
local containerSetRag = engine.containers(
|
||||
"text-completion-rag", [ containerRag ]
|
||||
);
|
||||
|
||||
local service =
|
||||
engine.internalService(containerSet)
|
||||
.with_port(8000, 8000, "metrics");
|
||||
|
||||
local serviceRag =
|
||||
engine.internalService(containerSetRag)
|
||||
.with_port(8000, 8000, "metrics");
|
||||
|
||||
engine.resources([
|
||||
envSecrets,
|
||||
containerSet,
|
||||
containerSetRag,
|
||||
service,
|
||||
serviceRag,
|
||||
])
|
||||
|
||||
|
||||
}
|
||||
},
|
||||
|
||||
} + prompts
|
||||
|
||||
|
|
|
|||
|
|
@ -6,12 +6,14 @@ local prompts = import "prompts/slm.jsonnet";
|
|||
{
|
||||
|
||||
"llamafile-model":: "LLaMA_CPP",
|
||||
"llamafile-url":: "${LLAMAFILE_URL}",
|
||||
|
||||
"text-completion" +: {
|
||||
|
||||
create:: function(engine)
|
||||
|
||||
local envSecrets = engine.envSecrets("llamafile-credentials")
|
||||
.with_env_var("LLAMAFILE_URL", "llamafile-url");
|
||||
|
||||
local container =
|
||||
engine.container("text-completion")
|
||||
.with_image(images.trustgraph)
|
||||
|
|
@ -21,27 +23,12 @@ local prompts = import "prompts/slm.jsonnet";
|
|||
url.pulsar,
|
||||
"-m",
|
||||
$["llamafile-model"],
|
||||
"-r",
|
||||
$["llamafile-url"],
|
||||
])
|
||||
.with_env_var_secrets(envSecrets)
|
||||
.with_limits("0.5", "128M")
|
||||
.with_reservations("0.1", "128M");
|
||||
|
||||
local containerSet = engine.containers(
|
||||
"text-completion", [ container ]
|
||||
);
|
||||
|
||||
engine.resources([
|
||||
containerSet,
|
||||
])
|
||||
|
||||
},
|
||||
|
||||
"text-completion-rag" +: {
|
||||
|
||||
create:: function(engine)
|
||||
|
||||
local container =
|
||||
local containerRag =
|
||||
engine.container("text-completion-rag")
|
||||
.with_image(images.trustgraph)
|
||||
.with_command([
|
||||
|
|
@ -50,26 +37,40 @@ local prompts = import "prompts/slm.jsonnet";
|
|||
url.pulsar,
|
||||
"-m",
|
||||
$["llamafile-model"],
|
||||
"-r",
|
||||
$["llamafile-url"],
|
||||
"-i",
|
||||
"non-persistent://tg/request/text-completion-rag",
|
||||
"-o",
|
||||
"non-persistent://tg/response/text-completion-rag-response",
|
||||
])
|
||||
.with_env_var_secrets(envSecrets)
|
||||
.with_limits("0.5", "128M")
|
||||
.with_reservations("0.1", "128M");
|
||||
|
||||
local containerSet = engine.containers(
|
||||
"text-completion-rag", [ container ]
|
||||
"text-completion", [ container ]
|
||||
);
|
||||
|
||||
local containerSetRag = engine.containers(
|
||||
"text-completion-rag", [ containerRag ]
|
||||
);
|
||||
|
||||
local service =
|
||||
engine.internalService(containerSet)
|
||||
.with_port(8080, 8080, "metrics");
|
||||
|
||||
local serviceRag =
|
||||
engine.internalService(containerSetRag)
|
||||
.with_port(8080, 8080, "metrics");
|
||||
|
||||
engine.resources([
|
||||
envSecrets,
|
||||
containerSet,
|
||||
containerSetRag,
|
||||
service,
|
||||
serviceRag,
|
||||
])
|
||||
|
||||
|
||||
}
|
||||
},
|
||||
|
||||
} + prompts
|
||||
|
||||
|
|
|
|||
|
|
@ -6,12 +6,14 @@ local prompts = import "prompts/slm.jsonnet";
|
|||
{
|
||||
|
||||
"ollama-model":: "gemma2:9b",
|
||||
"ollama-url":: "${OLLAMA_HOST}",
|
||||
|
||||
"text-completion" +: {
|
||||
|
||||
create:: function(engine)
|
||||
|
||||
local envSecrets = engine.envSecrets("ollama-credentials")
|
||||
.with_env_var("OLLAMA_HOST", "ollama-host");
|
||||
|
||||
local container =
|
||||
engine.container("text-completion")
|
||||
.with_image(images.trustgraph)
|
||||
|
|
@ -21,32 +23,12 @@ local prompts = import "prompts/slm.jsonnet";
|
|||
url.pulsar,
|
||||
"-m",
|
||||
$["ollama-model"],
|
||||
"-r",
|
||||
$["ollama-url"],
|
||||
])
|
||||
.with_env_var_secrets(envSecrets)
|
||||
.with_limits("0.5", "128M")
|
||||
.with_reservations("0.1", "128M");
|
||||
|
||||
local containerSet = engine.containers(
|
||||
"text-completion", [ container ]
|
||||
);
|
||||
|
||||
local service =
|
||||
engine.internalService(containerSet)
|
||||
.with_port(8080, 8080, "metrics");
|
||||
|
||||
engine.resources([
|
||||
containerSet,
|
||||
service,
|
||||
])
|
||||
|
||||
},
|
||||
|
||||
"text-completion-rag" +: {
|
||||
|
||||
create:: function(engine)
|
||||
|
||||
local container =
|
||||
local containerRag =
|
||||
engine.container("text-completion-rag")
|
||||
.with_image(images.trustgraph)
|
||||
.with_command([
|
||||
|
|
@ -55,31 +37,40 @@ local prompts = import "prompts/slm.jsonnet";
|
|||
url.pulsar,
|
||||
"-m",
|
||||
$["ollama-model"],
|
||||
"-r",
|
||||
$["ollama-url"],
|
||||
"-i",
|
||||
"non-persistent://tg/request/text-completion-rag",
|
||||
"-o",
|
||||
"non-persistent://tg/response/text-completion-rag-response",
|
||||
])
|
||||
.with_env_var_secrets(envSecrets)
|
||||
.with_limits("0.5", "128M")
|
||||
.with_reservations("0.1", "128M");
|
||||
|
||||
local containerSet = engine.containers(
|
||||
"text-completion-rag", [ container ]
|
||||
"text-completion", [ container ]
|
||||
);
|
||||
|
||||
local containerSetRag = engine.containers(
|
||||
"text-completion-rag", [ containerRag ]
|
||||
);
|
||||
|
||||
local service =
|
||||
engine.internalService(containerSet)
|
||||
.with_port(8080, 8080, "metrics");
|
||||
|
||||
local serviceRag =
|
||||
engine.internalService(containerSetRag)
|
||||
.with_port(8080, 8080, "metrics");
|
||||
|
||||
engine.resources([
|
||||
envSecrets,
|
||||
containerSet,
|
||||
containerSetRag,
|
||||
service,
|
||||
serviceRag,
|
||||
])
|
||||
|
||||
|
||||
}
|
||||
},
|
||||
|
||||
} + prompts
|
||||
|
||||
|
|
|
|||
|
|
@ -5,7 +5,6 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
|
||||
{
|
||||
|
||||
"openai-key":: "${OPENAI_KEY}",
|
||||
"openai-max-output-tokens":: 4096,
|
||||
"openai-temperature":: 0.0,
|
||||
"openai-model":: "GPT-3.5-Turbo",
|
||||
|
|
@ -14,6 +13,9 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
|
||||
create:: function(engine)
|
||||
|
||||
local envSecrets = engine.envSecrets("openai-credentials")
|
||||
.with_env_var("OPENAI_TOKEN", "openai-token");
|
||||
|
||||
local container =
|
||||
engine.container("text-completion")
|
||||
.with_image(images.trustgraph)
|
||||
|
|
@ -21,8 +23,6 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
"text-completion-openai",
|
||||
"-p",
|
||||
url.pulsar,
|
||||
"-k",
|
||||
$["openai-key"],
|
||||
"-x",
|
||||
std.toString($["openai-max-output-tokens"]),
|
||||
"-t",
|
||||
|
|
@ -30,37 +30,17 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
"-m",
|
||||
$["openai-model"],
|
||||
])
|
||||
.with_env_var_secrets(envSecrets)
|
||||
.with_limits("0.5", "128M")
|
||||
.with_reservations("0.1", "128M");
|
||||
|
||||
local containerSet = engine.containers(
|
||||
"text-completion", [ container ]
|
||||
);
|
||||
|
||||
local service =
|
||||
engine.internalService(containerSet)
|
||||
.with_port(8080, 8080, "metrics");
|
||||
|
||||
engine.resources([
|
||||
containerSet,
|
||||
service,
|
||||
])
|
||||
|
||||
},
|
||||
|
||||
"text-completion-rag" +: {
|
||||
|
||||
create:: function(engine)
|
||||
|
||||
local container =
|
||||
local containerRag =
|
||||
engine.container("text-completion-rag")
|
||||
.with_image(images.trustgraph)
|
||||
.with_command([
|
||||
"text-completion-openai",
|
||||
"-p",
|
||||
url.pulsar,
|
||||
"-k",
|
||||
$["openai-key"],
|
||||
"-x",
|
||||
std.toString($["openai-max-output-tokens"]),
|
||||
"-t",
|
||||
|
|
@ -72,24 +52,35 @@ local prompts = import "prompts/mixtral.jsonnet";
|
|||
"-o",
|
||||
"non-persistent://tg/response/text-completion-rag-response",
|
||||
])
|
||||
.with_env_var_secrets(envSecrets)
|
||||
.with_limits("0.5", "128M")
|
||||
.with_reservations("0.1", "128M");
|
||||
|
||||
local containerSet = engine.containers(
|
||||
"text-completion-rag", [ container ]
|
||||
"text-completion", [ container ]
|
||||
);
|
||||
|
||||
local containerSetRag = engine.containers(
|
||||
"text-completion-rag", [ containerRag ]
|
||||
);
|
||||
|
||||
local service =
|
||||
engine.internalService(containerSet)
|
||||
.with_port(8080, 8080, "metrics");
|
||||
|
||||
local serviceRag =
|
||||
engine.internalService(containerSetRag)
|
||||
.with_port(8080, 8080, "metrics");
|
||||
|
||||
engine.resources([
|
||||
envSecrets,
|
||||
containerSet,
|
||||
containerSetRag,
|
||||
service,
|
||||
serviceRag,
|
||||
])
|
||||
|
||||
|
||||
}
|
||||
},
|
||||
|
||||
} + prompts
|
||||
|
||||
|
|
|
|||
|
|
@ -18,12 +18,15 @@
|
|||
reservations: {},
|
||||
ports: [],
|
||||
volumes: [],
|
||||
environment: {},
|
||||
|
||||
with_image:: function(x) self + { image: x },
|
||||
|
||||
with_command:: function(x) self + { command: x },
|
||||
|
||||
with_environment:: function(x) self + { environment: x },
|
||||
with_environment:: function(x) self + {
|
||||
environment: super.environment + x,
|
||||
},
|
||||
|
||||
with_limits:: function(c, m) self + { limits: { cpus: c, memory: m } },
|
||||
|
||||
|
|
@ -45,6 +48,16 @@
|
|||
]
|
||||
},
|
||||
|
||||
with_env_var_secrets::
|
||||
function(vars)
|
||||
std.foldl(
|
||||
function(obj, x) obj.with_environment(
|
||||
{ [x]: "${" + x + "}" }
|
||||
),
|
||||
vars.variables,
|
||||
self
|
||||
),
|
||||
|
||||
add:: function() {
|
||||
services +: {
|
||||
[container.name]: {
|
||||
|
|
@ -62,7 +75,7 @@
|
|||
{ command: container.command }
|
||||
else {}) +
|
||||
|
||||
(if std.objectHas(container, "environment") then
|
||||
(if ! std.isEmpty(container.environment) then
|
||||
{ environment: container.environment }
|
||||
else {}) +
|
||||
|
||||
|
|
@ -170,6 +183,27 @@
|
|||
|
||||
},
|
||||
|
||||
envSecrets:: function(name)
|
||||
{
|
||||
|
||||
local volume = self,
|
||||
|
||||
name: name,
|
||||
|
||||
volid:: name,
|
||||
|
||||
variables:: [],
|
||||
|
||||
with_env_var::
|
||||
function(name, key) self + {
|
||||
variables: super.variables + [name],
|
||||
},
|
||||
|
||||
add:: function() {
|
||||
}
|
||||
|
||||
},
|
||||
|
||||
containers:: function(name, containers)
|
||||
{
|
||||
|
||||
|
|
|
|||
|
|
@ -10,12 +10,20 @@
|
|||
reservations: {},
|
||||
ports: [],
|
||||
volumes: [],
|
||||
environment: [],
|
||||
|
||||
with_image:: function(x) self + { image: x },
|
||||
|
||||
with_command:: function(x) self + { command: x },
|
||||
|
||||
with_environment:: function(x) self + { environment: x },
|
||||
with_environment:: function(x) self + {
|
||||
environment: super.environment + [
|
||||
{
|
||||
name: v.key, value: v.value
|
||||
}
|
||||
for v in std.objectKeysValues(x)
|
||||
],
|
||||
},
|
||||
|
||||
with_limits:: function(c, m) self + { limits: { cpu: c, memory: m } },
|
||||
|
||||
|
|
@ -37,6 +45,24 @@
|
|||
]
|
||||
},
|
||||
|
||||
with_env_var_secrets::
|
||||
function(vars)
|
||||
std.foldl(
|
||||
function(obj, x) obj + {
|
||||
environment: super.environment + [{
|
||||
name: x,
|
||||
valueFrom: {
|
||||
secretKeyRef: {
|
||||
name: vars.name,
|
||||
key: vars.keyMap[x],
|
||||
}
|
||||
}
|
||||
}]
|
||||
},
|
||||
vars.variables,
|
||||
self
|
||||
),
|
||||
|
||||
add:: function() [
|
||||
|
||||
{
|
||||
|
|
@ -97,16 +123,11 @@
|
|||
(if std.objectHas(container, "command") then
|
||||
{ command: container.command }
|
||||
else {}) +
|
||||
(if std.objectHas(container, "environment") then
|
||||
{ env: [ {
|
||||
name: e.key, value: e.value
|
||||
}
|
||||
for e in
|
||||
std.objectKeysValues(
|
||||
container.environment
|
||||
)
|
||||
]
|
||||
}
|
||||
|
||||
(if ! std.isEmpty(container.environment) then
|
||||
{
|
||||
env: container.environment,
|
||||
}
|
||||
else {}) +
|
||||
|
||||
(if std.length(container.volumes) > 0 then
|
||||
|
|
@ -283,6 +304,34 @@
|
|||
|
||||
},
|
||||
|
||||
envSecrets:: function(name)
|
||||
{
|
||||
|
||||
local volume = self,
|
||||
|
||||
name: name,
|
||||
|
||||
variables: [],
|
||||
keyMap: {},
|
||||
|
||||
with_size:: function(size) self + { size: size },
|
||||
|
||||
add:: function() [
|
||||
],
|
||||
|
||||
volRef:: function() {
|
||||
name: volume.name,
|
||||
secret: { secretName: volume.name },
|
||||
},
|
||||
|
||||
with_env_var::
|
||||
function(name, key) self + {
|
||||
variables: super.variables + [name],
|
||||
keyMap: super.keyMap + { [name]: key },
|
||||
},
|
||||
|
||||
},
|
||||
|
||||
containers:: function(name, containers)
|
||||
{
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue