Expose CLI flag to disable using GPU for offline chat model

- Offline chat models outputing gibberish when loaded onto some GPU.
  GPU support with Vulkan in GPT4All seems a bit buggy

- This change mitigates the upstream issue by allowing user to
  manually disable using GPU for offline chat

Closes #516

src/khoj/main.py

@@ -94,6 +94,7 @@ def set_state(args):
     state.port = args.port
     state.demo = args.demo
     state.khoj_version = version("khoj-assistant")
+    state.chat_on_gpu = args.chat_on_gpu
 
 
 def start_server(app, host=None, port=None, socket=None):

src/khoj/processor/conversation/gpt4all/utils.py

@@ -1,5 +1,7 @@
 import logging
 
+from khoj.utils import state
+
 
 logger = logging.getLogger(__name__)
 
@@ -13,8 +15,9 @@ def download_model(model_name: str):
 
     # Use GPU for Chat Model, if available
     try:
-        model = GPT4All(model_name=model_name, device="gpu")
+        device = "gpu" if state.chat_on_gpu else "cpu"
-        logger.debug(f"Loaded {model_name} chat model to GPU.")
+        model = GPT4All(model_name=model_name, device=device)
+        logger.debug(f"Loaded {model_name} chat model to {device.upper()}")
     except ValueError:
         model = GPT4All(model_name=model_name)
         logger.debug(f"Loaded {model_name} chat model to CPU.")

src/khoj/utils/cli.py

@@ -34,10 +34,16 @@ def cli(args=None):
         help="Path to UNIX socket for server. Use to run server behind reverse proxy. Default: /tmp/uvicorn.sock",
     )
     parser.add_argument("--version", "-V", action="store_true", help="Print the installed Khoj version and exit")
+    parser.add_argument(
+        "--disable-chat-on-gpu", action="store_true", default=False, help="Disable using GPU for the offline chat model"
+    )
     parser.add_argument("--demo", action="store_true", default=False, help="Run Khoj in demo mode")
 
     args = parser.parse_args(args)
 
+    # Set default values for arguments
+    args.chat_on_gpu = not args.disable_chat_on_gpu
+
     args.version_no = version("khoj-assistant")
     if args.version:
         # Show version of khoj installed and exit

src/khoj/utils/state.py

@@ -31,6 +31,8 @@ telemetry: List[Dict[str, str]] = []
 previous_query: str = None
 demo: bool = False
 khoj_version: str = None
+chat_on_gpu: bool = True
+
 
 if torch.cuda.is_available():
     # Use CUDA GPU