Handle offline chat model retrieval when no internet

Offline chat shouldn't fail on retrieve_model when no internet,
if model was previously downloaded and usable offline

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

@@ -12,11 +12,12 @@ def download_model(model_name: str):
         logger.info("There was an error importing GPT4All. Please run pip install gpt4all in order to install it.")
         raise e
 
-    # Download the chat model
-    chat_model_config = gpt4all.GPT4All.retrieve_model(model_name=model_name, allow_download=True)
-
     # Decide whether to load model to GPU or CPU
+    chat_model_config = None
     try:
+        # Download the chat model and its config
+        chat_model_config = gpt4all.GPT4All.retrieve_model(model_name=model_name, allow_download=True)
+
         # Try load chat model to GPU if:
         # 1. Loading chat model to GPU isn't disabled via CLI and
         # 2. Machine has GPU
@@ -26,6 +27,12 @@ def download_model(model_name: str):
         )
     except ValueError:
         device = "cpu"
+    except Exception as e:
+        if chat_model_config is None:
+            device = "cpu"  # Fallback to CPU as can't determine if GPU has enough memory
+            logger.debug(f"Unable to download model config from gpt4all website: {e}")
+        else:
+            raise e
 
     # Now load the downloaded chat model onto appropriate device
     chat_model = gpt4all.GPT4All(model_name=model_name, device=device, allow_download=False)