Set Gemma 2 2B MLX as default local model and add --mlx-model flag

apple_llm.py

@@ -92,7 +92,7 @@ class MacOSLLM(FrameProcessor):
     def __init__(
         self,
         *,
-        model: str = "mlx-community/Llama-3.2-1B-Instruct-4bit",
+        model: str = "mlx-community/gemma-2-2b-it-4bit",
         system_prompt: str | None = None,
         observer=None,
         **kwargs,
@@ -194,20 +194,20 @@ class MacOSLLM(FrameProcessor):
                 loop = asyncio.get_running_loop()
                 def _gen():
                     from mlx_lm import generate
+                    from mlx_lm.sample_utils import make_sampler
                     # Keep history concise to avoid context drift and repetition
                     recent_history = self._history[-8:]
                     messages = [{"role": "system", "content": self._system_prompt}] + recent_history
                     prompt = self._mlx_tokenizer.apply_chat_template(
                         messages, add_generation_prompt=True, tokenize=False
                     )
+                    sampler = make_sampler(temp=0.7)
                     return generate(
                         self._mlx_model,
                         self._mlx_tokenizer,
                         prompt=prompt,
                         max_tokens=150,
-                        temp=0.7,
-                        repetition_penalty=1.2,
-                        repetition_context_size=30,
+                        sampler=sampler,
                         verbose=False,
                     )