From 3255392dff2eded24e7f8d190d464e69e4184681 Mon Sep 17 00:00:00 2001 From: Ying Chun Guo Date: Wed, 29 May 2024 18:53:21 +0800 Subject: [PATCH] improve ChatQnA manifests (#213) Signed-off-by: Yingchun Guo --- ChatQnA/kubernetes/manifests/chaqna-xeon-backend-server.yaml | 1 - ChatQnA/kubernetes/manifests/embedding.yaml | 1 - ChatQnA/kubernetes/manifests/llm.yaml | 1 - ChatQnA/kubernetes/manifests/qna_configmap_gaudi.yaml | 3 +-- ChatQnA/kubernetes/manifests/qna_configmap_xeon.yaml | 3 +-- ChatQnA/kubernetes/manifests/reranking.yaml | 1 - ChatQnA/kubernetes/manifests/retriever.yaml | 1 - ChatQnA/kubernetes/manifests/tei_embedding_gaudi_service.yaml | 1 - ChatQnA/kubernetes/manifests/tei_embedding_service.yaml | 1 - ChatQnA/kubernetes/manifests/tei_reranking_service.yaml | 1 - ChatQnA/kubernetes/manifests/tgi_gaudi_service.yaml | 1 - ChatQnA/kubernetes/manifests/tgi_service.yaml | 1 - 12 files changed, 2 insertions(+), 14 deletions(-) diff --git a/ChatQnA/kubernetes/manifests/chaqna-xeon-backend-server.yaml b/ChatQnA/kubernetes/manifests/chaqna-xeon-backend-server.yaml index 447382815..ae0aaf4fb 100644 --- a/ChatQnA/kubernetes/manifests/chaqna-xeon-backend-server.yaml +++ b/ChatQnA/kubernetes/manifests/chaqna-xeon-backend-server.yaml @@ -6,7 +6,6 @@ apiVersion: apps/v1 kind: Deployment metadata: name: chaqna-xeon-backend-server-deploy - namespace: default spec: replicas: 1 selector: diff --git a/ChatQnA/kubernetes/manifests/embedding.yaml b/ChatQnA/kubernetes/manifests/embedding.yaml index 78e81c685..31d5ffa5c 100644 --- a/ChatQnA/kubernetes/manifests/embedding.yaml +++ b/ChatQnA/kubernetes/manifests/embedding.yaml @@ -6,7 +6,6 @@ apiVersion: apps/v1 kind: Deployment metadata: name: embedding-deploy - namespace: default spec: replicas: 1 selector: diff --git a/ChatQnA/kubernetes/manifests/llm.yaml b/ChatQnA/kubernetes/manifests/llm.yaml index 5405240c1..6d31e2c0f 100644 --- a/ChatQnA/kubernetes/manifests/llm.yaml +++ b/ChatQnA/kubernetes/manifests/llm.yaml @@ -6,7 +6,6 @@ apiVersion: apps/v1 kind: Deployment metadata: name: llm-deploy - namespace: default spec: replicas: 1 selector: diff --git a/ChatQnA/kubernetes/manifests/qna_configmap_gaudi.yaml b/ChatQnA/kubernetes/manifests/qna_configmap_gaudi.yaml index d0fe2fd65..dbd67872c 100644 --- a/ChatQnA/kubernetes/manifests/qna_configmap_gaudi.yaml +++ b/ChatQnA/kubernetes/manifests/qna_configmap_gaudi.yaml @@ -5,11 +5,10 @@ apiVersion: v1 kind: ConfigMap metadata: name: qna-config - namespace: default data: EMBEDDING_MODEL_ID: "BAAI/bge-base-en-v1.5" RERANK_MODEL_ID: "BAAI/bge-reranker-large" - LLM_MODEL_ID: "/data/neural-chat-7b-v3-3" + LLM_MODEL_ID: "Intel/neural-chat-7b-v3-3" TEI_EMBEDDING_ENDPOINT: "http://tei-embedding-gaudi-svc.default.svc.cluster.local:6006" TEI_RERANKING_ENDPOINT: "http://tei-reranking-svc.default.svc.cluster.local:8808" TGI_LLM_ENDPOINT: "http://tgi-gaudi-svc.default.svc.cluster.local:9009" diff --git a/ChatQnA/kubernetes/manifests/qna_configmap_xeon.yaml b/ChatQnA/kubernetes/manifests/qna_configmap_xeon.yaml index 33b4cf1f7..b5dd22303 100644 --- a/ChatQnA/kubernetes/manifests/qna_configmap_xeon.yaml +++ b/ChatQnA/kubernetes/manifests/qna_configmap_xeon.yaml @@ -5,11 +5,10 @@ apiVersion: v1 kind: ConfigMap metadata: name: qna-config - namespace: default data: EMBEDDING_MODEL_ID: "BAAI/bge-base-en-v1.5" RERANK_MODEL_ID: "BAAI/bge-reranker-large" - LLM_MODEL_ID: "/data/neural-chat-7b-v3-3" + LLM_MODEL_ID: "Intel/neural-chat-7b-v3-3" TEI_EMBEDDING_ENDPOINT: "http://tei-embedding-svc.default.svc.cluster.local:6006" TEI_RERANKING_ENDPOINT: "http://tei-reranking-svc.default.svc.cluster.local:8808" TGI_LLM_ENDPOINT: "http://tgi-svc.default.svc.cluster.local:9009" diff --git a/ChatQnA/kubernetes/manifests/reranking.yaml b/ChatQnA/kubernetes/manifests/reranking.yaml index 9a6e11e47..7c3ca8b2c 100644 --- a/ChatQnA/kubernetes/manifests/reranking.yaml +++ b/ChatQnA/kubernetes/manifests/reranking.yaml @@ -6,7 +6,6 @@ apiVersion: apps/v1 kind: Deployment metadata: name: reranking-deploy - namespace: default spec: replicas: 1 selector: diff --git a/ChatQnA/kubernetes/manifests/retriever.yaml b/ChatQnA/kubernetes/manifests/retriever.yaml index 0a4c84956..04157ba14 100644 --- a/ChatQnA/kubernetes/manifests/retriever.yaml +++ b/ChatQnA/kubernetes/manifests/retriever.yaml @@ -6,7 +6,6 @@ apiVersion: apps/v1 kind: Deployment metadata: name: retriever-deploy - namespace: default spec: replicas: 1 selector: diff --git a/ChatQnA/kubernetes/manifests/tei_embedding_gaudi_service.yaml b/ChatQnA/kubernetes/manifests/tei_embedding_gaudi_service.yaml index bb029727d..60081154b 100644 --- a/ChatQnA/kubernetes/manifests/tei_embedding_gaudi_service.yaml +++ b/ChatQnA/kubernetes/manifests/tei_embedding_gaudi_service.yaml @@ -6,7 +6,6 @@ apiVersion: apps/v1 kind: Deployment metadata: name: tei-embedding-gaudi-service-deploy - namespace: default spec: replicas: 1 selector: diff --git a/ChatQnA/kubernetes/manifests/tei_embedding_service.yaml b/ChatQnA/kubernetes/manifests/tei_embedding_service.yaml index edbe66fde..5051e78fa 100644 --- a/ChatQnA/kubernetes/manifests/tei_embedding_service.yaml +++ b/ChatQnA/kubernetes/manifests/tei_embedding_service.yaml @@ -6,7 +6,6 @@ apiVersion: apps/v1 kind: Deployment metadata: name: tei-embedding-service-deploy - namespace: default spec: replicas: 1 selector: diff --git a/ChatQnA/kubernetes/manifests/tei_reranking_service.yaml b/ChatQnA/kubernetes/manifests/tei_reranking_service.yaml index 49f3f382f..7528d01d4 100644 --- a/ChatQnA/kubernetes/manifests/tei_reranking_service.yaml +++ b/ChatQnA/kubernetes/manifests/tei_reranking_service.yaml @@ -6,7 +6,6 @@ apiVersion: apps/v1 kind: Deployment metadata: name: tei-reranking-service-deploy - namespace: default spec: replicas: 1 selector: diff --git a/ChatQnA/kubernetes/manifests/tgi_gaudi_service.yaml b/ChatQnA/kubernetes/manifests/tgi_gaudi_service.yaml index 45910cf11..039b39079 100644 --- a/ChatQnA/kubernetes/manifests/tgi_gaudi_service.yaml +++ b/ChatQnA/kubernetes/manifests/tgi_gaudi_service.yaml @@ -6,7 +6,6 @@ apiVersion: apps/v1 kind: Deployment metadata: name: tgi-gaudi-service-deploy - namespace: default spec: replicas: 1 selector: diff --git a/ChatQnA/kubernetes/manifests/tgi_service.yaml b/ChatQnA/kubernetes/manifests/tgi_service.yaml index 9ab23f556..7a858c7b3 100644 --- a/ChatQnA/kubernetes/manifests/tgi_service.yaml +++ b/ChatQnA/kubernetes/manifests/tgi_service.yaml @@ -6,7 +6,6 @@ apiVersion: apps/v1 kind: Deployment metadata: name: tgi-service-deploy - namespace: default spec: replicas: 1 selector: