version bump + some fixes (#792)

llms/mlx_lm/lora.py

@@ -80,7 +80,7 @@ def build_parser():
     parser.add_argument(
         "--lora-layers",
         type=int,
-        help="Number of layers to fine-tune",
+        help="Number of layers to fine-tune. Default is 16, use -1 for all.",
     )
     parser.add_argument("--batch-size", type=int, help="Minibatch size.")
     parser.add_argument("--iters", type=int, help="Iterations to train for.")
@@ -143,7 +143,7 @@ def build_parser():
         help="Use gradient checkpointing to reduce memory use.",
         default=None,
     )
-    parser.add_argument("--seed", type=int, default=0, help="The PRNG seed")
+    parser.add_argument("--seed", type=int, default=None, help="The PRNG seed")
     parser.add_argument(
         "--use-dora", action="store_true", default=None, help="Use DoRA to finetune."
     )
@@ -268,7 +268,7 @@ def main():
             config = yaml.load(file, yaml_loader)
         # Prefer parameters from command-line arguments
         for k, v in config.items():
-            if args.get(k, None) is not None:
+            if args.get(k, None) is None:
                 args[k] = v
 
     # Update defaults for unspecified parameters

llms/mlx_lm/requirements.txt

@@ -1,4 +1,4 @@
-mlx>=0.11
+mlx>=0.13.1
 numpy
 transformers>=4.39.3
 protobuf

llms/mlx_lm/tuner/utils.py

@@ -54,6 +54,10 @@ def linear_to_lora_layers(
     """
 
     num_layers = len(model.layers)
+
+    if num_lora_layers < 0:
+        num_lora_layers = num_layers
+
     if num_lora_layers > num_layers:
         raise ValueError(
             f"Requested {num_lora_layers} LoRA layers "

llms/mlx_lm/version.py

@@ -1,3 +1,3 @@
 # Copyright © 2023-2024 Apple Inc.
 
-__version__ = "0.13.1"
+__version__ = "0.14.0"