GenAIExamples/ChatQnA/chatqna.yaml

# Copyright (C) 2024 Intel Corporation
# SPDX-License-Identifier: Apache-2.0

opea_micro_services:
  redis-vector-db:
    host: ${REDIS_SERVICE_HOST_IP}
    ports:
      - "6379:6379"
      - "8001:8001"
    image: redis/redis-stack:7.2.0-v9
  dataprep-redis-service:
    host: ${DATAPREP_SERVICE_HOST_IP}
    ports: ${DATAPREP_SERVICE_PORT}
    image: opea/dataprep-redis:latest
    environment:
      REDIS_URL: ${REDIS_URL}
      INDEX_NAME: ${INDEX_NAME}
  tei-embedding-service:
    host: ${TEI_EMBEDDING_SERVICE_IP}
    ports: ${TEI_EMBEDDING_SERVICE_PORT}
    image: ghcr.io/huggingface/tei-gaudi:latest
    volumes:
      - "./data:/data"
    runtime: habana
    cap_add:
      - SYS_NICE
    ipc: host
    environment:
      HABANA_VISIBLE_DEVICES: all
      OMPI_MCA_btl_vader_single_copy_mechanism: none
    model-id: ${EMBEDDING_MODEL_ID}
  retrieval:
    host: ${RETRIEVER_SERVICE_HOST_IP}
    ports: ${RETRIEVER_SERVICE_PORT}
    image: opea/retriever-redis:latest
    endpoint: /v1/retrieval
  tgi-service:
    host: ${TGI_SERVICE_IP}
    ports: ${TGI_SERVICE_PORT}
    image: ghcr.io/huggingface/tgi-gaudi:2.0.6
    volumes:
      - "./data:/data"
    runtime: habana
    cap_add:
      - SYS_NICE
    ipc: host
    environment:
      HUGGING_FACE_HUB_TOKEN: ${HUGGINGFACEHUB_API_TOKEN}
      HABANA_VISIBLE_DEVICES: all
      OMPI_MCA_btl_vader_single_copy_mechanism: none
      ENABLE_HPU_GRAPH: true
      LIMIT_HPU_GRAPH: true
      USE_FLASH_ATTENTION: true
      FLASH_ATTENTION_RECOMPUTE: true
    model-id: ${LLM_MODEL_ID}
  ui:
    host: ${UI_SERVICE_HOST_IP}
    ports:
      - "5173:5173"
    image: opea/chatqna-ui:latest
    environment:
      - CHAT_BASE_URL=${BACKEND_SERVICE_ENDPOINT}
      - UPLOAD_FILE_BASE_URL=${DATAPREP_SERVICE_ENDPOINT}

opea_mega_service:
  host: ${MEGA_SERVICE_HOST_IP}
  ports: ${MEGA_SERVICE_PORT}
  image: opea/chatqna:latest
  endpoint: /v1/chatqna
  mega_flow:
    - embedding >> retrieval >> reranking >> llm