Add learnable last-four-layer query fusion

streaming_turn_model_directly.py

@@ -1061,6 +1061,15 @@ def parse_args(argv: Sequence[str] | None = None) -> argparse.Namespace:
             "causal readout token (default: 1 preserves the original result)."
         ),
     )
+    parser.add_argument(
+        "--readout-layer-fusion",
+        choices=("mean", "stack"),
+        default="mean",
+        help=(
+            "Fuse selected query layers by a fixed mean, or preserve [N,D] "
+            "for a trainable downstream layer mixture."
+        ),
+    )
     parser.add_argument("--frame-cache-size", type=int, default=8)
     parser.add_argument(
         "--audio-feature-device",
@@ -1341,6 +1350,7 @@ def main(
         video_min_pixels=args.video_min_pixels,
         video_max_pixels=args.video_max_pixels,
         readout_last_n_layers=args.readout_last_n_layers,
+        readout_layer_fusion=args.readout_layer_fusion,
     )
     config.validate()