Skip to content

Commit ce5885d

Browse files
ashwinbleseb
ashwinb
authored and
leseb
committed
fix(k8s): unwedge run.yaml to add files
1 parent 5c9cf28 commit ce5885d

File tree

4 files changed

+77
-142
lines changed

4 files changed

+77
-142
lines changed

‎docs/source/distributions/k8s-benchmark/stack-configmap.yaml‎

Lines changed: 9 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -6,6 +6,7 @@ data:
66
apis:
77
- agents
88
- inference
9+
- files
910
- safety
1011
- telemetry
1112
- tool_runtime
@@ -19,13 +20,6 @@ data:
1920
max_tokens: ${env.VLLM_MAX_TOKENS:=4096}
2021
api_token: ${env.VLLM_API_TOKEN:=fake}
2122
tls_verify: ${env.VLLM_TLS_VERIFY:=true}
22-
- provider_id: vllm-safety
23-
provider_type: remote::vllm
24-
config:
25-
url: ${env.VLLM_SAFETY_URL:=http://localhost:8000/v1}
26-
max_tokens: ${env.VLLM_MAX_TOKENS:=4096}
27-
api_token: ${env.VLLM_API_TOKEN:=fake}
28-
tls_verify: ${env.VLLM_TLS_VERIFY:=true}
2923
- provider_id: sentence-transformers
3024
provider_type: inline::sentence-transformers
3125
config: {}
@@ -41,6 +35,14 @@ data:
4135
db: ${env.POSTGRES_DB:=llamastack}
4236
user: ${env.POSTGRES_USER:=llamastack}
4337
password: ${env.POSTGRES_PASSWORD:=llamastack}
38+
files:
39+
- provider_id: meta-reference-files
40+
provider_type: inline::localfs
41+
config:
42+
storage_dir: ${env.FILES_STORAGE_DIR:=~/.llama/distributions/starter/files}
43+
metadata_store:
44+
type: sqlite
45+
db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/files_metadata.db
4446
safety:
4547
- provider_id: llama-guard
4648
provider_type: inline::llama-guard
@@ -111,9 +113,6 @@ data:
111113
- model_id: ${env.INFERENCE_MODEL}
112114
provider_id: vllm-inference
113115
model_type: llm
114-
- model_id: ${env.SAFETY_MODEL}
115-
provider_id: vllm-safety
116-
model_type: llm
117116
shields:
118117
- shield_id: ${env.SAFETY_MODEL:=meta-llama/Llama-Guard-3-1B}
119118
vector_dbs: []

‎docs/source/distributions/k8s-benchmark/stack_run_config.yaml‎

Lines changed: 9 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -3,6 +3,7 @@ image_name: kubernetes-benchmark-demo
33
apis:
44
- agents
55
- inference
6+
- files
67
- safety
78
- telemetry
89
- tool_runtime
@@ -31,6 +32,14 @@ providers:
3132
db: ${env.POSTGRES_DB:=llamastack}
3233
user: ${env.POSTGRES_USER:=llamastack}
3334
password: ${env.POSTGRES_PASSWORD:=llamastack}
35+
files:
36+
- provider_id: meta-reference-files
37+
provider_type: inline::localfs
38+
config:
39+
storage_dir: ${env.FILES_STORAGE_DIR:=~/.llama/distributions/starter/files}
40+
metadata_store:
41+
type: sqlite
42+
db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/files_metadata.db
3443
safety:
3544
- provider_id: llama-guard
3645
provider_type: inline::llama-guard

‎docs/source/distributions/k8s/stack-configmap.yaml‎

Lines changed: 50 additions & 132 deletions
Original file line numberDiff line numberDiff line change
@@ -1,137 +1,55 @@
11
apiVersion: v1
22
data:
3-
stack_run_config.yaml: |
4-
version: '2'
5-
image_name: kubernetes-demo
6-
apis:
7-
- agents
8-
- inference
9-
- safety
10-
- telemetry
11-
- tool_runtime
12-
- vector_io
13-
providers:
14-
inference:
15-
- provider_id: vllm-inference
16-
provider_type: remote::vllm
17-
config:
18-
url: ${env.VLLM_URL:=http://localhost:8000/v1}
19-
max_tokens: ${env.VLLM_MAX_TOKENS:=4096}
20-
api_token: ${env.VLLM_API_TOKEN:=fake}
21-
tls_verify: ${env.VLLM_TLS_VERIFY:=true}
22-
- provider_id: vllm-safety
23-
provider_type: remote::vllm
24-
config:
25-
url: ${env.VLLM_SAFETY_URL:=http://localhost:8000/v1}
26-
max_tokens: ${env.VLLM_MAX_TOKENS:=4096}
27-
api_token: ${env.VLLM_API_TOKEN:=fake}
28-
tls_verify: ${env.VLLM_TLS_VERIFY:=true}
29-
- provider_id: sentence-transformers
30-
provider_type: inline::sentence-transformers
31-
config: {}
32-
vector_io:
33-
- provider_id: ${env.ENABLE_CHROMADB:+chromadb}
34-
provider_type: remote::chromadb
35-
config:
36-
url: ${env.CHROMADB_URL:=}
37-
kvstore:
38-
type: postgres
39-
host: ${env.POSTGRES_HOST:=localhost}
40-
port: ${env.POSTGRES_PORT:=5432}
41-
db: ${env.POSTGRES_DB:=llamastack}
42-
user: ${env.POSTGRES_USER:=llamastack}
43-
password: ${env.POSTGRES_PASSWORD:=llamastack}
44-
safety:
45-
- provider_id: llama-guard
46-
provider_type: inline::llama-guard
47-
config:
48-
excluded_categories: []
49-
agents:
50-
- provider_id: meta-reference
51-
provider_type: inline::meta-reference
52-
config:
53-
persistence_store:
54-
type: postgres
55-
host: ${env.POSTGRES_HOST:=localhost}
56-
port: ${env.POSTGRES_PORT:=5432}
57-
db: ${env.POSTGRES_DB:=llamastack}
58-
user: ${env.POSTGRES_USER:=llamastack}
59-
password: ${env.POSTGRES_PASSWORD:=llamastack}
60-
responses_store:
61-
type: postgres
62-
host: ${env.POSTGRES_HOST:=localhost}
63-
port: ${env.POSTGRES_PORT:=5432}
64-
db: ${env.POSTGRES_DB:=llamastack}
65-
user: ${env.POSTGRES_USER:=llamastack}
66-
password: ${env.POSTGRES_PASSWORD:=llamastack}
67-
telemetry:
68-
- provider_id: meta-reference
69-
provider_type: inline::meta-reference
70-
config:
71-
service_name: "${env.OTEL_SERVICE_NAME:=\u200B}"
72-
sinks: ${env.TELEMETRY_SINKS:=console}
73-
tool_runtime:
74-
- provider_id: brave-search
75-
provider_type: remote::brave-search
76-
config:
77-
api_key: ${env.BRAVE_SEARCH_API_KEY:+}
78-
max_results: 3
79-
- provider_id: tavily-search
80-
provider_type: remote::tavily-search
81-
config:
82-
api_key: ${env.TAVILY_SEARCH_API_KEY:+}
83-
max_results: 3
84-
- provider_id: rag-runtime
85-
provider_type: inline::rag-runtime
86-
config: {}
87-
- provider_id: model-context-protocol
88-
provider_type: remote::model-context-protocol
89-
config: {}
90-
metadata_store:
91-
type: postgres
92-
host: ${env.POSTGRES_HOST:=localhost}
93-
port: ${env.POSTGRES_PORT:=5432}
94-
db: ${env.POSTGRES_DB:=llamastack}
95-
user: ${env.POSTGRES_USER:=llamastack}
96-
password: ${env.POSTGRES_PASSWORD:=llamastack}
97-
table_name: llamastack_kvstore
98-
inference_store:
99-
type: postgres
100-
host: ${env.POSTGRES_HOST:=localhost}
101-
port: ${env.POSTGRES_PORT:=5432}
102-
db: ${env.POSTGRES_DB:=llamastack}
103-
user: ${env.POSTGRES_USER:=llamastack}
104-
password: ${env.POSTGRES_PASSWORD:=llamastack}
105-
models:
106-
- metadata:
107-
embedding_dimension: 384
108-
model_id: all-MiniLM-L6-v2
109-
provider_id: sentence-transformers
110-
model_type: embedding
111-
- metadata: {}
112-
model_id: ${env.INFERENCE_MODEL}
113-
provider_id: vllm-inference
114-
model_type: llm
115-
- metadata: {}
116-
model_id: ${env.SAFETY_MODEL:=meta-llama/Llama-Guard-3-1B}
117-
provider_id: vllm-safety
118-
model_type: llm
119-
shields:
120-
- shield_id: ${env.SAFETY_MODEL:=meta-llama/Llama-Guard-3-1B}
121-
vector_dbs: []
122-
datasets: []
123-
scoring_fns: []
124-
benchmarks: []
125-
tool_groups:
126-
- toolgroup_id: builtin::websearch
127-
provider_id: tavily-search
128-
- toolgroup_id: builtin::rag
129-
provider_id: rag-runtime
130-
server:
131-
port: 8321
132-
auth:
133-
provider_config:
134-
type: github_token
3+
stack_run_config.yaml: "version: '2'\nimage_name: kubernetes-demo\napis:\n- agents\n-
4+
inference\n- files\n- safety\n- telemetry\n- tool_runtime\n- vector_io\nproviders:\n
5+
\ inference:\n - provider_id: vllm-inference\n provider_type: remote::vllm\n
6+
\ config:\n url: ${env.VLLM_URL:=http://localhost:8000/v1}\n max_tokens:
7+
${env.VLLM_MAX_TOKENS:=4096}\n api_token: ${env.VLLM_API_TOKEN:=fake}\n tls_verify:
8+
${env.VLLM_TLS_VERIFY:=true}\n - provider_id: vllm-safety\n provider_type:
9+
remote::vllm\n config:\n url: ${env.VLLM_SAFETY_URL:=http://localhost:8000/v1}\n
10+
\ max_tokens: ${env.VLLM_MAX_TOKENS:=4096}\n api_token: ${env.VLLM_API_TOKEN:=fake}\n
11+
\ tls_verify: ${env.VLLM_TLS_VERIFY:=true}\n - provider_id: sentence-transformers\n
12+
\ provider_type: inline::sentence-transformers\n config: {}\n vector_io:\n
13+
\ - provider_id: ${env.ENABLE_CHROMADB:+chromadb}\n provider_type: remote::chromadb\n
14+
\ config:\n url: ${env.CHROMADB_URL:=}\n kvstore:\n type: postgres\n
15+
\ host: ${env.POSTGRES_HOST:=localhost}\n port: ${env.POSTGRES_PORT:=5432}\n
16+
\ db: ${env.POSTGRES_DB:=llamastack}\n user: ${env.POSTGRES_USER:=llamastack}\n
17+
\ password: ${env.POSTGRES_PASSWORD:=llamastack}\n files:\n - provider_id:
18+
meta-reference-files\n provider_type: inline::localfs\n config:\n storage_dir:
19+
${env.FILES_STORAGE_DIR:=~/.llama/distributions/starter/files}\n metadata_store:\n
20+
\ type: sqlite\n db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/files_metadata.db
21+
\ \n safety:\n - provider_id: llama-guard\n provider_type: inline::llama-guard\n
22+
\ config:\n excluded_categories: []\n agents:\n - provider_id: meta-reference\n
23+
\ provider_type: inline::meta-reference\n config:\n persistence_store:\n
24+
\ type: postgres\n host: ${env.POSTGRES_HOST:=localhost}\n port:
25+
${env.POSTGRES_PORT:=5432}\n db: ${env.POSTGRES_DB:=llamastack}\n user:
26+
${env.POSTGRES_USER:=llamastack}\n password: ${env.POSTGRES_PASSWORD:=llamastack}\n
27+
\ responses_store:\n type: postgres\n host: ${env.POSTGRES_HOST:=localhost}\n
28+
\ port: ${env.POSTGRES_PORT:=5432}\n db: ${env.POSTGRES_DB:=llamastack}\n
29+
\ user: ${env.POSTGRES_USER:=llamastack}\n password: ${env.POSTGRES_PASSWORD:=llamastack}\n
30+
\ telemetry:\n - provider_id: meta-reference\n provider_type: inline::meta-reference\n
31+
\ config:\n service_name: \"${env.OTEL_SERVICE_NAME:=\\u200B}\"\n sinks:
32+
${env.TELEMETRY_SINKS:=console}\n tool_runtime:\n - provider_id: brave-search\n
33+
\ provider_type: remote::brave-search\n config:\n api_key: ${env.BRAVE_SEARCH_API_KEY:+}\n
34+
\ max_results: 3\n - provider_id: tavily-search\n provider_type: remote::tavily-search\n
35+
\ config:\n api_key: ${env.TAVILY_SEARCH_API_KEY:+}\n max_results:
36+
3\n - provider_id: rag-runtime\n provider_type: inline::rag-runtime\n config:
37+
{}\n - provider_id: model-context-protocol\n provider_type: remote::model-context-protocol\n
38+
\ config: {}\nmetadata_store:\n type: postgres\n host: ${env.POSTGRES_HOST:=localhost}\n
39+
\ port: ${env.POSTGRES_PORT:=5432}\n db: ${env.POSTGRES_DB:=llamastack}\n user:
40+
${env.POSTGRES_USER:=llamastack}\n password: ${env.POSTGRES_PASSWORD:=llamastack}\n
41+
\ table_name: llamastack_kvstore\ninference_store:\n type: postgres\n host:
42+
${env.POSTGRES_HOST:=localhost}\n port: ${env.POSTGRES_PORT:=5432}\n db: ${env.POSTGRES_DB:=llamastack}\n
43+
\ user: ${env.POSTGRES_USER:=llamastack}\n password: ${env.POSTGRES_PASSWORD:=llamastack}\nmodels:\n-
44+
metadata:\n embedding_dimension: 384\n model_id: all-MiniLM-L6-v2\n provider_id:
45+
sentence-transformers\n model_type: embedding\n- metadata: {}\n model_id: ${env.INFERENCE_MODEL}\n
46+
\ provider_id: vllm-inference\n model_type: llm\n- metadata: {}\n model_id:
47+
${env.SAFETY_MODEL:=meta-llama/Llama-Guard-3-1B}\n provider_id: vllm-safety\n
48+
\ model_type: llm\nshields:\n- shield_id: ${env.SAFETY_MODEL:=meta-llama/Llama-Guard-3-1B}\nvector_dbs:
49+
[]\ndatasets: []\nscoring_fns: []\nbenchmarks: []\ntool_groups:\n- toolgroup_id:
50+
builtin::websearch\n provider_id: tavily-search\n- toolgroup_id: builtin::rag\n
51+
\ provider_id: rag-runtime\nserver:\n port: 8321\n auth:\n provider_config:\n
52+
\ type: github_token\n"
13553
kind: ConfigMap
13654
metadata:
13755
creationTimestamp: null

‎docs/source/distributions/k8s/stack_run_config.yaml‎

Lines changed: 9 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -3,6 +3,7 @@ image_name: kubernetes-demo
33
apis:
44
- agents
55
- inference
6+
- files
67
- safety
78
- telemetry
89
- tool_runtime
@@ -38,6 +39,14 @@ providers:
3839
db: ${env.POSTGRES_DB:=llamastack}
3940
user: ${env.POSTGRES_USER:=llamastack}
4041
password: ${env.POSTGRES_PASSWORD:=llamastack}
42+
files:
43+
- provider_id: meta-reference-files
44+
provider_type: inline::localfs
45+
config:
46+
storage_dir: ${env.FILES_STORAGE_DIR:=~/.llama/distributions/starter/files}
47+
metadata_store:
48+
type: sqlite
49+
db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/files_metadata.db
4150
safety:
4251
- provider_id: llama-guard
4352
provider_type: inline::llama-guard

0 commit comments

Comments
 (0)