diff --git a/toolchain/cli/download.py b/toolchain/cli/download.py
index 10889992e..ab8e96b7f 100644
--- a/toolchain/cli/download.py
+++ b/toolchain/cli/download.py
@@ -7,9 +7,10 @@ from huggingface_hub import snapshot_download
 from huggingface_hub.utils import GatedRepoError, RepositoryNotFoundError
 
 from toolchain.cli.subcommand import Subcommand
+from toolchain.utils import DEFAULT_DUMP_DIR
 
 
-DEFAULT_CHECKPOINT_DIR = f"{os.path.expanduser('~')}/.llama/checkpoints/"
+DEFAULT_CHECKPOINT_DIR = os.path.join(DEFAULT_DUMP_DIR, "checkpoints")
 
 
 class Download(Subcommand):
@@ -61,10 +62,8 @@ class Download(Subcommand):
 
     def _run_download_cmd(self, args: argparse.Namespace):
         model_name = args.repo_id.split("/")[-1]
-
+        output_dir = Path(DEFAULT_CHECKPOINT_DIR) / model_name
         os.makedirs(output_dir, exist_ok=True)
-
-        output_dir = Path(output_dir) / model_name
         try:
             true_output_dir = snapshot_download(
                 args.repo_id,
diff --git a/toolchain/cli/inference/configure.py b/toolchain/cli/inference/configure.py
index 9c728f6d8..df97ebf04 100644
--- a/toolchain/cli/inference/configure.py
+++ b/toolchain/cli/inference/configure.py
@@ -5,9 +5,11 @@ import textwrap
 from pathlib import Path
 
 from toolchain.cli.subcommand import Subcommand
+from toolchain.utils import DEFAULT_DUMP_DIR
 
 
-CONFIGS_BASE_DIR = f"{os.path.expanduser('~')}/.llama/configs/"
+CONFIGS_BASE_DIR = os.path.join(DEFAULT_DUMP_DIR, "configs")
+
 
 class InferenceConfigure(Subcommand):
     """Llama cli for configuring llama toolchain configs"""
diff --git a/toolchain/cli/inference/start.py b/toolchain/cli/inference/start.py
index ab447b644..c105d4198 100644
--- a/toolchain/cli/inference/start.py
+++ b/toolchain/cli/inference/start.py
@@ -43,6 +43,7 @@ class InferenceStart(Subcommand):
             "--config",
             type=str,
             help="Path to config file",
+            default="inference"
         )
 
     def _run_inference_start_cmd(self, args: argparse.Namespace) -> None:
diff --git a/toolchain/inference/server.py b/toolchain/inference/server.py
index 52aac3dda..a2846f136 100644
--- a/toolchain/inference/server.py
+++ b/toolchain/inference/server.py
@@ -10,7 +10,7 @@ from fastapi.responses import StreamingResponse
 
 from omegaconf import OmegaConf
 
-from toolchain.utils import get_config_dir, parse_config
+from toolchain.utils import get_default_config_dir, parse_config
 from .api.config import ModelInferenceHydraConfig
 from .api.endpoints import ChatCompletionRequest, ChatCompletionResponseStreamChunk
 
@@ -100,7 +100,7 @@ def chat_completion(request: Request, exec_request: ChatCompletionRequest):
 
 def main(config_path: str, port: int = 5000, disable_ipv6: bool = False):
     global GLOBAL_CONFIG
-    config_dir = get_config_dir()
+    config_dir = get_default_config_dir()
     GLOBAL_CONFIG = parse_config(config_dir, config_path)
 
     signal.signal(signal.SIGINT, handle_sigint)
diff --git a/toolchain/utils.py b/toolchain/utils.py
index b8c91f529..dbc72b0c8 100644
--- a/toolchain/utils.py
+++ b/toolchain/utils.py
@@ -8,6 +8,9 @@ from hydra.core.global_hydra import GlobalHydra
 from omegaconf import OmegaConf
 
 
+DEFAULT_DUMP_DIR = os.path.expanduser("~/.llama/")
+
+
 def get_root_directory():
     current_dir = os.path.dirname(os.path.abspath(__file__))
     while os.path.isfile(os.path.join(current_dir, "__init__.py")):
@@ -16,8 +19,8 @@ def get_root_directory():
     return current_dir
 
 
-def get_config_dir():
-    return os.path.join(get_root_directory(), "toolchain", "configs")
+def get_default_config_dir():
+    return os.path.join(DEFAULT_DUMP_DIR, "configs")
 
 
 def parse_config(config_dir: str, config_path: Optional[str] = None) -> str: