Separate RAG content and embeddings images, use CPU version of torch …

…library

.tekton/ansible-chatbot-service-pull-request.yaml

@@ -258,7 +258,7 @@ spec:
           - name: COMMIT_SHA
             value: $(tasks.clone-repository.results.commit)
           - name: BUILD_ARGS
-            value: ["IMAGE_TAGS=pr-{{pull_request_number}} pr-{{pull_request_number}}.$(tasks.git-metadata.results.commit-timestamp)", "GIT_COMMIT=$(tasks.clone-repository.results.commit)", "RAG_CONTENTS_SUB_FOLDER=vector_db/aap_product_docs", "LIGHTSPEED_RAG_CONTENT_IMAGE=quay.io/ansible/aap-rag-content:latest"]
+            value: ["IMAGE_TAGS=pr-{{pull_request_number}} pr-{{pull_request_number}}.$(tasks.git-metadata.results.commit-timestamp)", "GIT_COMMIT=$(tasks.clone-repository.results.commit)", "RAG_CONTENTS_SUB_FOLDER=vector_db/aap_product_docs", "LIGHTSPEED_RAG_CONTENT_IMAGE=quay.io/ansible/aap-rag-content:latest", "LIGHTSPEED_RAG_EMBEDDINGS_IMAGE=quay.io/ansible/aap-rag-embeddings-image:latest"]
           - name: BUILD_ARGS_FILE
             value: $(params.build-args-file)
         runAfter:

.tekton/ansible-chatbot-service-push.yaml

@@ -274,7 +274,7 @@ spec:
           - name: TARGET_STAGE
             value: production
           - name: BUILD_ARGS
-            value: [ "IMAGE_TAGS=latest 1.0.$(tasks.git-metadata.results.commit-timestamp)", "GIT_COMMIT=$(tasks.clone-repository.results.commit)", "RAG_CONTENTS_SUB_FOLDER=vector_db/aap_product_docs", "LIGHTSPEED_RAG_CONTENT_IMAGE=quay.io/ansible/aap-rag-content:latest"]
+            value: [ "IMAGE_TAGS=latest 1.0.$(tasks.git-metadata.results.commit-timestamp)", "GIT_COMMIT=$(tasks.clone-repository.results.commit)", "RAG_CONTENTS_SUB_FOLDER=vector_db/aap_product_docs", "LIGHTSPEED_RAG_CONTENT_IMAGE=quay.io/ansible/aap-rag-content:latest", "LIGHTSPEED_RAG_EMBEDDINGS_IMAGE=quay.io/ansible/aap-rag-embeddings-image:latest"]
         runAfter:
           - prefetch-dependencies
         taskRef:

Containerfile

@@ -28,12 +28,13 @@ ENV PYTHONDONTWRITEBYTECODE=1 \
 WORKDIR /app-root
 
 COPY --from=lightspeed-rag-content /rag/${RAG_CONTENTS_SUB_FOLDER} ${APP_ROOT}/${RAG_CONTENTS_SUB_FOLDER}
-COPY --from=lightspeed-rag-content /rag/embeddings_model ./embeddings_model
+COPY --from=lightspeed-rag-embeddings /rag/embeddings_model ./embeddings_model
 
 # Add explicit files and directories
 # (avoid accidental inclusion of local directories or env files or credentials)
 COPY runner.py requirements.txt ./
 
+RUN pip3.11 install --upgrade pip
 RUN pip3.11 install --no-cache-dir -r requirements.txt
 
 COPY ols ./ols

pyproject.toml

@@ -89,7 +89,8 @@ description = "Road-core service is an AI powered assistant that runs on OpenShi
 authors = []
 dependencies = [
     "pdm==2.21.0",
-    "torch==2.5.1",
+    'torch@http://download.pytorch.org/whl/cpu/torch-2.5.1%2Bcpu-cp311-cp311-linux_x86_64.whl ; platform_system != "Darwin"',
+    'torch-macos@http://download.pytorch.org/whl/cpu/torch-2.5.1-cp311-none-macosx_11_0_arm64.whl ; platform_system == "Darwin"',
     "pandas==2.1.4",
     "httpx==0.27.2",
     "fastapi==0.115.6",

scripts/build-container-aap.sh

@@ -3,7 +3,8 @@
 # Build an ansible-chatbot-service image locally
 
 AAP_VERSION=v2.5
-LIGHTSPEED_RAG_CONTENT_IMAGE=quay.io/ttakamiy/aap-rag-content:latest
+LIGHTSPEED_RAG_CONTENT_IMAGE=quay.io/ansible/aap-rag-content:latest
+LIGHTSPEED_RAG_EMBEDDINGS_IMAGE=quay.io/ansible/aap-rag-embeddings-image:latest
 RAG_CONTENTS_SUB_FOLDER=vector_db/aap_product_docs
 
 CACHE_OPTS=""
@@ -16,6 +17,7 @@ podman build \
   ${CACHE_OPTS} \
   --build-arg=VERSION="${AAP_VERSION}" \
   --build-arg=LIGHTSPEED_RAG_CONTENT_IMAGE="${LIGHTSPEED_RAG_CONTENT_IMAGE}" \
+  --build-arg=LIGHTSPEED_RAG_EMBEDDINGS_IMAGE="${LIGHTSPEED_RAG_EMBEDDINGS_IMAGE}" \
   --build-arg=RAG_CONTENTS_SUB_FOLDER="${RAG_CONTENTS_SUB_FOLDER}" \
   -t "${AAP_API_IMAGE:-quay.io/ansible/ansible-chatbot-service:latest}" \
   -f Containerfile