mejora en el modelo de traduccion

docker-compose.yml

@@ -1,3 +1,16 @@
+# ==================================================================================
+# CONFIGURACIÓN DE SECRETOS - Se cargan desde .env
+# ==================================================================================
+# Las variables POSTGRES_PASSWORD, REDIS_PASSWORD, DB_PASS, SECRET_KEY, GRAFANA_PASSWORD
+# deben estar definidas en un archivo .env antes de ejecutar docker compose up
+#
+# Para generar credenciales seguras automáticamente:
+#   ./generate_secure_credentials.sh --force
+#
+# Despliegue simple:
+#   docker compose up -d
+# ==================================================================================
+
 services:
   db:
     image: postgres:18-alpine
@@ -25,6 +38,7 @@ services:
     restart: unless-stopped
     healthcheck:
       test: [ "CMD-SHELL", "pg_isready -h 127.0.0.1 -p 5432 -U $$POSTGRES_USER -d $$POSTGRES_DB || exit 1" ]
+      # NOTA: Si falla, verifica que POSTGRES_PASSWORD esté definido en .env
       interval: 5s
       timeout: 5s
       retries: 30
@@ -43,7 +57,7 @@ services:
       TZ: Europe/Madrid
     # SEGURIDAD: Redis con autenticación
     command: >
-      redis-server --appendonly yes --maxmemory 512mb --maxmemory-policy allkeys-lru --requirepass ${REDIS_PASSWORD}
+      redis-server --appendonly yes --maxmemory 512mb --maxmemory-policy allkeys-lru --requirepass ${REDIS_PASSWORD:-!ERROR!REDIS_PASSWORD_NEEDED!}
     volumes:
       - ./data/redis-data:/data
       - /etc/timezone:/etc/timezone:ro
@@ -55,6 +69,7 @@ services:
           - rss2_redis
     restart: unless-stopped
     healthcheck:
+      # VALIDACIÓN: Si REDIS_PASSWORD está vacío, esto fallará y detendrá el despliegue
       test: [ "CMD", "redis-cli", "--no-auth-warning", "-a", "${REDIS_PASSWORD}", "ping" ]
       interval: 5s
       timeout: 3s
@@ -282,8 +297,6 @@ services:
       dockerfile: Dockerfile.translator
     image: rss2-translator:latest
     command: bash -lc "python -m workers.ctranslator_worker"
-    security_opt:
-      - seccomp=unconfined
     environment:
       DB_HOST: db
       DB_PORT: 5432
@@ -299,14 +312,60 @@ services:
       HF_HOME: /app/hf_cache
       TZ: Europe/Madrid
       TRANSLATOR_ID: ${TRANSLATOR_ID:-}
+      PYTORCH_ENABLE_MPS_FALLBACK: 1
     volumes:
       - ./workers:/app/workers
       - ./hf_cache:/app/hf_cache
       - ./models:/app/models
     networks:
       - backend
-    profiles:
-      - cpu-only
+    depends_on:
+      db:
+        condition: service_healthy
+    restart: unless-stopped
+
+  # ==================================================================================
+  # TRANSLATOR GPU - Multiple instances with --scale for parallel GPU processing
+  # Configure multiple GPU workers: docker compose up -d --scale translator-gpu=2
+  # ==================================================================================
+  translator-gpu:
+    build:
+      context: .
+      dockerfile: Dockerfile.translator-gpu
+    image: rss2-translator-gpu:latest
+    command: bash -lc "python -m workers.ctranslator_worker"
+    environment:
+      DB_HOST: db
+      DB_PORT: 5432
+      DB_NAME: ${DB_NAME:-rss}
+      DB_USER: ${DB_USER:-rss}
+      DB_PASS: ${DB_PASS}
+      TARGET_LANGS: es
+      TRANSLATOR_BATCH: 256
+      CT2_MODEL_PATH: /app/models/nllb-ct2-1.3b
+      CT2_DEVICE: cuda
+      CT2_COMPUTE_TYPE: float16
+      UNIVERSAL_MODEL: facebook/nllb-200-1.3B
+      HF_HOME: /app/hf_cache
+      TZ: Europe/Madrid
+      TRANSLATOR_ID: ${TRANSLATOR_ID:-}
+      PYTORCH_CUDA_ALLOC_CONF: max_split_size_mb:512
+      NCCL_DEBUG: INFO
+    volumes:
+      - ./workers:/app/workers
+      - ./hf_cache:/app/hf_cache
+      - ./models:/app/models
+      - /dev/nvidia:/dev/nvidia
+      - /usr/local/nvidia:/usr/local/nvidia:ro
+    networks:
+      - backend
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: 1
+              capabilities: [gpu]
     depends_on:
       db:
         condition: service_healthy
@@ -335,8 +394,6 @@ services:
       - ./models:/app/models
     networks:
       - backend
-    profiles:
-      - remote
     deploy:
       resources:
         limits:
@@ -381,52 +438,6 @@ services:
         condition: service_healthy
     restart: unless-stopped
 
-  # ==================================================================================
-  # TRANSLATOR GPU (CTranslate2 with CUDA)
-  # ==================================================================================
-  translator-gpu:
-    build:
-      context: .
-      dockerfile: Dockerfile.translator-gpu
-    image: rss2-translator-gpu:latest
-    container_name: rss2_translator_gpu
-    command: bash -lc "python -m workers.ctranslator_worker"
-    security_opt:
-      - seccomp=unconfined
-    environment:
-      DB_HOST: db
-      DB_PORT: 5432
-      DB_NAME: ${DB_NAME:-rss}
-      DB_USER: ${DB_USER:-rss}
-      DB_PASS: ${DB_PASS}
-      TARGET_LANGS: es
-      TRANSLATOR_BATCH: 64
-      CT2_MODEL_PATH: /app/models/nllb-ct2
-      CT2_DEVICE: cuda
-      CT2_COMPUTE_TYPE: float16
-      UNIVERSAL_MODEL: facebook/nllb-200-distilled-600M
-      HF_HOME: /app/hf_cache
-      TZ: Europe/Madrid
-    volumes:
-      - ./workers:/app/workers
-      - ./hf_cache:/app/hf_cache
-      - ./models:/app/models
-    networks:
-      - backend
-    deploy:
-      resources:
-        limits:
-          memory: 4G
-        reservations:
-          devices:
-            - driver: nvidia
-              count: 1
-              capabilities: [ gpu ]
-    depends_on:
-      db:
-        condition: service_healthy
-    restart: unless-stopped
-
   embeddings:
     build:
       context: .