docker compose for ray only

docker/.env.ray-only.example

@@ -0,0 +1,40 @@
+COMPOSE_PROJECT_NAME=dats-ray-worker
+
+# Which user and group to use for running processes
+# inside containers.
+# Set this to the user and group you're using
+# outside the containers to be able to read
+# and write in folders mapped to volumes.
+UID=1004
+GID=1004
+
+#The root path to repo_root inside docker containers
+SHARED_REPO_ROOT=/tmp/dats
+
+# Docker tag to use for pulling the ray container
+DATS_RAY_DOCKER_VERSION=1.0.5
+
+# Which backend config file to use when running backend or celery
+DATS_BACKEND_CONFIG=configs/default.yaml
+
+# LOG LEVEL ('critical', 'error', 'warning', 'info', 'debug', 'trace')
+LOG_LEVEL=debug
+
+
+# Processing devices can be either "cuda" or "cpu"
+# Spacy only supports cpu at the moment
+RAY_PROCESSING_DEVICE_SPACY=cpu
+RAY_PROCESSING_DEVICE_WHISPER=cuda
+RAY_PROCESSING_DEVICE_DETR=cuda
+RAY_PROCESSING_DEVICE_VIT_GPT2=cuda
+RAY_PROCESSING_DEVICE_BLIP2=cuda
+RAY_PROCESSING_DEVICE_COTA=cuda
+# If operating blip2 on the cpu, this has to be 32
+# Otherwise, you can choose between 8, 16 and 32
+RAY_BLIP2_PRECISION_BIT=8
+RAY_PROCESSING_DEVICE_CLIP=cuda
+
+RAY_API_EXPOSED=43134
+RAY_DASHBOARD_EXPOSED=43135
+
+RAY_GPU_DEVICE_IDS=0

docker/docker-compose-ray-only.yml

@@ -0,0 +1,44 @@
+services:
+  ray:
+    env_file: ".env.ray-only"
+    image: uhhlt/dats_ray:${DATS_RAY_DOCKER_VERSION}
+    command: /dats_code_ray/ray_model_worker_entrypoint.sh
+    user: ${UID:-1004}:${GID:-1004}
+    environment:
+      LOG_LEVEL: ${LOG_LEVEL:-info}
+      DATS_BACKEND_CONFIG: ${DATS_BACKEND_CONFIG:-/dats_code/src/configs/default.yaml}
+      HUGGINGFACE_HUB_CACHE: /models_cache
+      TRANSFORMERS_CACHE: /models_cache
+      TORCH_HOME: /models_cache
+      RAY_PROCESSING_DEVICE_SPACY: ${RAY_PROCESSING_DEVICE_SPACY}
+      RAY_PROCESSING_DEVICE_WHISPER: ${RAY_PROCESSING_DEVICE_WHISPER}
+      RAY_PROCESSING_DEVICE_DETR: ${RAY_PROCESSING_DEVICE_DETR}
+      RAY_PROCESSING_DEVICE_VIT_GPT2: ${RAY_PROCESSING_DEVICE_VIT_GPT2}
+      RAY_PROCESSING_DEVICE_BLIP2: ${RAY_PROCESSING_DEVICE_BLIP2}
+      RAY_PROCESSING_DEVICE_CLIP: ${RAY_PROCESSING_DEVICE_CLIP}
+      RAY_PROCESSING_DEVICE_COTA: ${RAY_PROCESSING_DEVICE_COTA}
+      RAY_BLIP2_PRECISION_BIT: ${RAY_BLIP2_PRECISION_BIT:-32}
+      SHARED_REPO_ROOT: ${SHARED_REPO_ROOT:-/tmp/dats}
+      NUMBA_CACHE_DIR: /numba_cache
+    volumes:
+      #      - ../backend/src/app/preprocessing/ray_model_worker:/dats_code_ray
+      - ./spacy_models:/spacy_models
+      - ./backend_repo:${SHARED_REPO_ROOT:-/tmp/dats}
+      - ./models_cache:/models_cache
+      - ./numba_cache:/numba_cache
+    ports:
+      - "${RAY_API_EXPOSED:-8000}:8000"
+      - "${RAY_DASHBOARD_EXPOSED:-8265}:8265"
+    restart: always
+    shm_size: 12gb
+    networks:
+      - dats_demo_network
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              device_ids: ["${RAY_GPU_DEVICE_IDS:-0}"]
+              capabilities: [gpu]
+networks:
+  dats_demo_network: