Utilities to save and load only tasks weights

In addition to the private H5IOStore (which we rely on for lora), this
also relies on `_save_state` and `_load_state`. We should probably think
of a better contract with core Keras before we ship, but this will allow
a prototype.

Author: mattdangerw

keras_nlp/models/task.py

@@ -321,6 +321,55 @@ def from_preset(
             config_overrides=kwargs,
         )
 
+    def load_task_weights(self, filepath, skip_mismatch=False):
+        """Load only the tasks specific weights not in the backbone."""
+        if not str(filepath).endswith(".weights.h5"):
+            raise ValueError(
+                "The filename must end in `.weights.h5`. "
+                f"Received: filepath={filepath}"
+            )
+        weights_store = keras.src.saving.saving_lib.H5IOStore(
+            filepath, mode="r"
+        )
+        backbone_layer_ids = set(id(w) for w in self.backbone._flatten_layers())
+        keras.src.saving.saving_lib._load_state(
+            self,
+            weights_store=weights_store,
+            assets_store=None,
+            inner_path="",
+            skip_mismatch=skip_mismatch,
+            visited_trackables=backbone_layer_ids,
+            failed_trackables=set(),
+        )
+        weights_store.close()
+
+    def save_task_weights(self, filepath):
+        """Save only the tasks specific weights not in the backbone."""
+        if not str(filepath).endswith(".weights.h5"):
+            raise ValueError(
+                "The filename must end in `.weights.h5`. "
+                f"Received: filepath={filepath}"
+            )
+        task_weight_ids = set(id(w) for w in self.weights)
+        backbone_weight_ids = set(id(w) for w in self.backbone.weights)
+        backbone_layer_ids = set(id(w) for w in self.backbone._flatten_layers())
+        if set(task_weight_ids) in set(backbone_weight_ids):
+            raise ValueError(
+                f"Task {self} has no weights not in the `backbone`. "
+                "`save_task_weights()` has nothing to save."
+            )
+        weights_store = keras.src.saving.saving_lib.H5IOStore(
+            filepath, mode="w"
+        )
+        keras.src.saving.saving_lib._save_state(
+            self,
+            weights_store=weights_store,
+            assets_store=None,
+            inner_path="",
+            visited_trackables=backbone_layer_ids,
+        )
+        weights_store.close()
+
     @property
     def layers(self):
         # Remove preprocessor from layers so it does not show up in the summary.