Merge pull request #1323 from gzrp/dev-postgresql

Add the implementations of tuner for the PEFT example

Author: lzjpaul

examples/singa_peft/src/singa_peft/tuners/linear_lora/tuner.py

@@ -0,0 +1,58 @@
+#
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+#
+
+from singa import model, layer
+from singa_peft.peft_registry import PeftRegistry
+from singa_peft.tuners.base_tuner import BaseTuner
+from singa_peft.tuners.linear_lora.config import LinearLoraConfig
+from singa_peft.tuners.linear_lora.layer import LinearLoRALayer
+
+
+@PeftRegistry.register("linear_lora")
+class LinearLoraTuner(BaseTuner):
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.targeted_layers = []
+
+    def inject(self, base_model: model.Model) -> model.Model:
+        # freeze base_model parameters
+        if self.config.freeze_base_model:
+            self.freeze_base_parameters(base_model)
+        return self._inject_linear_lora(base_model, self.config)
+
+    def _inject_linear_lora(self, base_model, config: LinearLoraConfig) -> model.Model:
+        target_layers = config.target_layers
+        r = config.r
+        lora_alpha = config.lora_alpha
+        lora_dropout = config.lora_dropout
+        for target_layer in target_layers:
+            base_layer = getattr(base_model, target_layer)
+            if base_layer is not None and isinstance(base_layer, layer.Linear):
+                self.targeted_layers.append(target_layer)
+                new_layer = LinearLoRALayer(base_layer, r, lora_alpha, lora_dropout)
+                setattr(base_model, target_layer, new_layer)
+        return base_model
+
+    def merge_weights(self, base_model: model.Model, mode: bool = True) -> model.Model:
+        for target_layer in self.targeted_layers:
+            base_layer = getattr(base_model, target_layer)
+            if base_layer is not None:
+                base_layer.merge_weights(mode)
+        return base_model