Use the Apple Silicon GPU when available

Pick cuda, then mps, then cpu.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

embedlab.py

@@ -31,7 +31,12 @@ LAST_MODEL_FILE = os.path.join(os.path.dirname(os.path.abspath(__file__)), ".cac
 HISTORY_FILE = os.path.join(os.path.dirname(os.path.abspath(__file__)), ".cache", "history")
 COMMANDS = ["/select", "/sim", "/tokens", "/params", "/set", "/reset", "/lsmod", "/help", "/quit", "/exit"]
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
+if torch.cuda.is_available():
+    device = "cuda"
+elif torch.backends.mps.is_available():  # Apple Silicon GPU
+    device = "mps"
+else:
+    device = "cpu"
 _loaded = {}      # loaded models: name -> (tokenizer, model)
 _params = {name: dict(d) for name, d in MODELS.items()}  # live, per-model parameters
 current = None    # currently selected model