Implement --free_gpu_mem for server/backend.py

solves #813
solves #831

backend/modules/create_cmd_parser.py

@@ -45,5 +45,11 @@ def create_cmd_parser():
         help=f'Set model precision. Defaults to auto selected based on device. Options: {", ".join(PRECISION_CHOICES)}',
         default="auto",
     )
+    parser.add_argument(
+        '--free_gpu_mem',
+        dest='free_gpu_mem',
+        action='store_true',
+        help='Force free gpu memory before final decoding',
+    )
 
     return parser

backend/server.py

@@ -50,6 +50,7 @@ host = opt.host  # Web & socket.io host
 port = opt.port  # Web & socket.io port
 verbose = opt.verbose  # enables copious socket.io logging
 precision = opt.precision
+free_gpu_mem = opt.free_gpu_mem
 embedding_path = opt.embedding_path
 additional_allowed_origins = (
     opt.cors if opt.cors else []
@@ -148,6 +149,7 @@ generate = Generate(
     precision=precision,
     embedding_path=embedding_path,
 )
+generate.free_gpu_mem = free_gpu_mem
 generate.load_model()