Update usage example with trust_remote_code parameter

README.md

@@ -98,9 +98,12 @@ Training and inference on M2 Pro (measured at checkpoint 20000):
 ```python
 from transformers import AutoTokenizer, AutoModelForCausalLM
 
-# Load model and tokenizer
+# Load model and tokenizer (requires trust_remote_code for custom architecture)
 tokenizer = AutoTokenizer.from_pretrained("jacksuuuu/nanogpt-mlx-53m-finewebedu")
-model = AutoModelForCausalLM.from_pretrained("jacksuuuu/nanogpt-mlx-53m-finewebedu")
+model = AutoModelForCausalLM.from_pretrained(
+    "jacksuuuu/nanogpt-mlx-53m-finewebedu",
+    trust_remote_code=True
+)
 
 # Generate text
 prompt = "Once upon a time"