jandres/GenAIExamples
1
0
Fork 0
Code Issues Pull Requests Actions 5 Packages Projects Releases Wiki Activity

updated chatqna helmcharts image name

Browse Source
This commit is contained in:
Zhenzhong Xu
2024-10-21 06:54:27 +03:00
parent 3dd5475773
commit a70775d3d6
3 changed files with 16 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
ChatQnA/benchmark/performance/helm_charts/customize.yaml
View File

@@ -6,8 +6,6 @@ HUGGINGFACEHUB_API_TOKEN: ${HF_TOKEN}
podSpecs: podSpecs:
- name: chatqna-backend-server-deploy - name: chatqna-backend-server-deploy
spec: spec:
image_name: opea/chatqna
image_tag: latest
replicas: 2 replicas: 2
resources: resources:
limits: limits:
@@ -19,8 +17,6 @@ podSpecs:
- name: embedding-dependency-deploy - name: embedding-dependency-deploy
spec: spec:
image_name: ghcr.io/huggingface/text-embeddings-inference
image_tag: cpu-1.5
replicas: 1 replicas: 1
resources: resources:
limits: limits:
@@ -32,8 +28,6 @@ podSpecs:
- name: reranking-dependency-deploy - name: reranking-dependency-deploy
spec: spec:
image_name: opea/tei-gaudi
image_tag: latest
replicas: 1 replicas: 1
resources: resources:
limits: limits:
@@ -41,8 +35,6 @@ podSpecs:
- name: llm-dependency-deploy - name: llm-dependency-deploy
spec: spec:
image_name: ghcr.io/huggingface/tgi-gaudi
image_tag: 2.0.4
replicas: 7 replicas: 7
resources: resources:
limits: limits:
@@ -50,20 +42,14 @@ podSpecs:
- name: dataprep-deploy - name: dataprep-deploy
spec: spec:
image_name: opea/dataprep-redis
image_tag: latest
replicas: 1 replicas: 1
- name: vector-db - name: vector-db
spec: spec:
image_name: redis/redis-stack
image_tag: 7.2.0-v9
replicas: 1 replicas: 1
- name: retriever-deploy - name: retriever-deploy
spec: spec:
image_name: opea/retriever-redis
image_tag: latest
replicas: 2 replicas: 2
resources: resources:
requests: requests:

4
ChatQnA/benchmark/performance/helm_charts/templates/deployment.yaml
View File

@@ -11,7 +11,7 @@ metadata:
name: {{ $deployment.name }} name: {{ $deployment.name }}
namespace: default namespace: default
spec: spec:
replicas: {{ $podSpec.spec.replicas }} replicas: {{ with $podSpec.spec.replicas }}{{ . }}{{ else }}{{ $deployment.spec.replicas }}{{ end }}
selector: selector:
matchLabels: matchLabels:
app: {{ $deployment.name }} app: {{ $deployment.name }}
@@ -46,7 +46,7 @@ spec:
{{- end }} {{- end }}
{{- end }} {{- end }}
image: {{ $podSpec.spec.image_name }}:{{ $podSpec.spec.image_tag }} image: {{ $deployment.spec.image }}
imagePullPolicy: IfNotPresent imagePullPolicy: IfNotPresent
name: {{ $podSpec.name }} name: {{ $podSpec.name }}

14
ChatQnA/benchmark/performance/helm_charts/values.yaml
View File

@@ -12,28 +12,38 @@ config:
deployments: deployments:
- name: chatqna-backend-server-deploy - name: chatqna-backend-server-deploy
image: opea/chatqna:latest
replicas: 1
spec: spec:
ports: ports:
- containerPort: 8888 - containerPort: 8888
- name: dataprep-deploy - name: dataprep-deploy
spec: spec:
image: opea/dataprep-redis:latest
replicas: 1
ports: ports:
- containerPort: 6007 - containerPort: 6007
- name: vector-db - name: vector-db
spec: spec:
image: redis/redis-stack:7.2.0-v9
replicas: 1
ports: ports:
- containerPort: 6379 - containerPort: 6379
- containerPort: 8001 - containerPort: 8001
- name: retriever-deploy - name: retriever-deploy
spec: spec:
image: opea/retriever-redis:latest
replicas: 1
ports: ports:
- containerPort: 7000 - containerPort: 7000
- name: embedding-dependency-deploy - name: embedding-dependency-deploy
spec: spec:
image: ghcr.io/huggingface/text-embeddings-inference:cpu-1.5
replicas: 1
ports: ports:
- containerPort: 80 - containerPort: 80
args: args:
@@ -57,6 +67,8 @@ deployments:
- name: reranking-dependency-deploy - name: reranking-dependency-deploy
spec: spec:
image: opea/tei-gaudi:latest
replicas: 1
args: args:
- name: "--model-id" - name: "--model-id"
- value: $(RERANK_MODEL_ID) - value: $(RERANK_MODEL_ID)
@@ -91,6 +103,8 @@ deployments:
- name: llm-dependency-deploy - name: llm-dependency-deploy
spec: spec:
image: ghcr.io/huggingface/tgi-gaudi:2.0.4
replicas: 1
ports: ports:
- containerPort: 80 - containerPort: 80
resources: resources: