Port distilbert transformer checkpoint (#1736)

* port distilbert

* update test

* resolve comments

* fixed embedding matching

keras_nlp/src/utils/transformers/convert.py

@@ -16,6 +16,12 @@
 
 from keras_nlp.src.utils.transformers.convert_bert import load_bert_backbone
 from keras_nlp.src.utils.transformers.convert_bert import load_bert_tokenizer
+from keras_nlp.src.utils.transformers.convert_distilbert import (
+    load_distilbert_backbone,
+)
+from keras_nlp.src.utils.transformers.convert_distilbert import (
+    load_distilbert_tokenizer,
+)
 from keras_nlp.src.utils.transformers.convert_gemma import load_gemma_backbone
 from keras_nlp.src.utils.transformers.convert_gemma import load_gemma_tokenizer
 from keras_nlp.src.utils.transformers.convert_gpt2 import load_gpt2_backbone
@@ -56,6 +62,8 @@ def load_transformers_backbone(cls, preset, load_weights):
         return load_pali_gemma_backbone(cls, preset, load_weights)
     if cls.__name__ == "GPT2Backbone":
         return load_gpt2_backbone(cls, preset, load_weights)
+    if cls.__name__ == "DistilBertBackbone":
+        return load_distilbert_backbone(cls, preset, load_weights)
     raise ValueError(
         f"{cls} has not been ported from the Hugging Face format yet. "
         "Please check Hugging Face Hub for the Keras model. "
@@ -85,6 +93,8 @@ def load_transformers_tokenizer(cls, preset):
         return load_pali_gemma_tokenizer(cls, preset)
     if cls.__name__ == "GPT2Tokenizer":
         return load_gpt2_tokenizer(cls, preset)
+    if cls.__name__ == "DistilBertTokenizer":
+        return load_distilbert_tokenizer(cls, preset)
     raise ValueError(
         f"{cls} has not been ported from the Hugging Face format yet. "
         "Please check Hugging Face Hub for the Keras model. "

keras_nlp/src/utils/transformers/convert_distilbert.py

@@ -0,0 +1,184 @@
+# Copyright 2024 The KerasNLP Authors
+#
+# Licensed 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
+#
+#     https://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.
+import numpy as np
+
+from keras_nlp.src.utils.preset_utils import HF_CONFIG_FILE
+from keras_nlp.src.utils.preset_utils import HF_TOKENIZER_CONFIG_FILE
+from keras_nlp.src.utils.preset_utils import get_file
+from keras_nlp.src.utils.preset_utils import jax_memory_cleanup
+from keras_nlp.src.utils.preset_utils import load_config
+from keras_nlp.src.utils.transformers.safetensor_utils import SafetensorLoader
+
+
+def convert_backbone_config(transformers_config):
+    return {
+        "vocabulary_size": transformers_config["vocab_size"],
+        "num_layers": transformers_config["n_layers"],
+        "num_heads": transformers_config["n_heads"],
+        "hidden_dim": transformers_config["dim"],
+        "intermediate_dim": transformers_config["hidden_dim"],
+        "dropout": transformers_config["dropout"],
+        "max_sequence_length": transformers_config["max_position_embeddings"],
+    }
+
+
+def convert_weights(backbone, loader):
+    # Embeddings
+    loader.port_weight(
+        keras_variable=backbone.get_layer(
+            "token_and_position_embedding"
+        ).token_embedding.embeddings,
+        hf_weight_key="distilbert.embeddings.word_embeddings.weight",
+    )
+    loader.port_weight(
+        keras_variable=backbone.get_layer(
+            "token_and_position_embedding"
+        ).position_embedding.position_embeddings,
+        hf_weight_key="distilbert.embeddings.position_embeddings.weight",
+    )
+
+    # Attention blocks
+    for index in range(backbone.num_layers):
+        decoder_layer = backbone.transformer_layers[index]
+
+        # Norm layers
+        loader.port_weight(
+            keras_variable=decoder_layer._self_attention_layer_norm.gamma,
+            hf_weight_key=f"distilbert.transformer.layer.{index}.sa_layer_norm.weight",
+        )
+        loader.port_weight(
+            keras_variable=decoder_layer._self_attention_layer_norm.beta,
+            hf_weight_key=f"distilbert.transformer.layer.{index}.sa_layer_norm.bias",
+        )
+        loader.port_weight(
+            keras_variable=decoder_layer._feedforward_layer_norm.gamma,
+            hf_weight_key=f"distilbert.transformer.layer.{index}.output_layer_norm.weight",
+        )
+        loader.port_weight(
+            keras_variable=decoder_layer._feedforward_layer_norm.beta,
+            hf_weight_key=f"distilbert.transformer.layer.{index}.output_layer_norm.bias",
+        )
+
+        # Attention layers
+        # Query
+        loader.port_weight(
+            keras_variable=decoder_layer._self_attention_layer.query_dense.kernel,
+            hf_weight_key=f"distilbert.transformer.layer.{index}.attention.q_lin.weight",
+            hook_fn=lambda hf_tensor, keras_shape: np.reshape(
+                np.transpose(hf_tensor), keras_shape
+            ),
+        )
+        loader.port_weight(
+            keras_variable=decoder_layer._self_attention_layer.query_dense.bias,
+            hf_weight_key=f"distilbert.transformer.layer.{index}.attention.q_lin.bias",
+            hook_fn=lambda hf_tensor, keras_shape: np.reshape(
+                hf_tensor, keras_shape
+            ),
+        )
+
+        # Key
+        loader.port_weight(
+            keras_variable=decoder_layer._self_attention_layer.key_dense.kernel,
+            hf_weight_key=f"distilbert.transformer.layer.{index}.attention.k_lin.weight",
+            hook_fn=lambda hf_tensor, keras_shape: np.reshape(
+                np.transpose(hf_tensor), keras_shape
+            ),
+        )
+        loader.port_weight(
+            keras_variable=decoder_layer._self_attention_layer.key_dense.bias,
+            hf_weight_key=f"distilbert.transformer.layer.{index}.attention.k_lin.bias",
+            hook_fn=lambda hf_tensor, keras_shape: np.reshape(
+                hf_tensor, keras_shape
+            ),
+        )
+
+        # Value
+        loader.port_weight(
+            keras_variable=decoder_layer._self_attention_layer.value_dense.kernel,
+            hf_weight_key=f"distilbert.transformer.layer.{index}.attention.v_lin.weight",
+            hook_fn=lambda hf_tensor, keras_shape: np.reshape(
+                np.transpose(hf_tensor), keras_shape
+            ),
+        )
+        loader.port_weight(
+            keras_variable=decoder_layer._self_attention_layer.value_dense.bias,
+            hf_weight_key=f"distilbert.transformer.layer.{index}.attention.v_lin.bias",
+            hook_fn=lambda hf_tensor, keras_shape: np.reshape(
+                hf_tensor, keras_shape
+            ),
+        )
+
+        # Output
+        loader.port_weight(
+            keras_variable=decoder_layer._self_attention_layer.output_dense.kernel,
+            hf_weight_key=f"distilbert.transformer.layer.{index}.attention.out_lin.weight",
+            hook_fn=lambda hf_tensor, keras_shape: np.reshape(
+                np.transpose(hf_tensor), keras_shape
+            ),
+        )
+        loader.port_weight(
+            keras_variable=decoder_layer._self_attention_layer.output_dense.bias,
+            hf_weight_key=f"distilbert.transformer.layer.{index}.attention.out_lin.bias",
+        )
+
+        # MLP layers
+        loader.port_weight(
+            keras_variable=decoder_layer._feedforward_intermediate_dense.kernel,
+            hf_weight_key=f"distilbert.transformer.layer.{index}.ffn.lin1.weight",
+            hook_fn=lambda hf_tensor, _: np.transpose(hf_tensor, axes=(1, 0)),
+        )
+        loader.port_weight(
+            keras_variable=decoder_layer._feedforward_intermediate_dense.bias,
+            hf_weight_key=f"distilbert.transformer.layer.{index}.ffn.lin1.bias",
+        )
+        loader.port_weight(
+            keras_variable=decoder_layer._feedforward_output_dense.kernel,
+            hf_weight_key=f"distilbert.transformer.layer.{index}.ffn.lin2.weight",
+            hook_fn=lambda hf_tensor, _: np.transpose(hf_tensor, axes=(1, 0)),
+        )
+        loader.port_weight(
+            keras_variable=decoder_layer._feedforward_output_dense.bias,
+            hf_weight_key=f"distilbert.transformer.layer.{index}.ffn.lin2.bias",
+        )
+
+    # Normalization
+    loader.port_weight(
+        keras_variable=backbone.embeddings_layer_norm.gamma,
+        hf_weight_key="distilbert.embeddings.LayerNorm.weight",
+    )
+    loader.port_weight(
+        keras_variable=backbone.embeddings_layer_norm.beta,
+        hf_weight_key="distilbert.embeddings.LayerNorm.bias",
+    )
+
+    return backbone
+
+
+def load_distilbert_backbone(cls, preset, load_weights):
+    transformers_config = load_config(preset, HF_CONFIG_FILE)
+    keras_config = convert_backbone_config(transformers_config)
+    backbone = cls(**keras_config)
+    if load_weights:
+        jax_memory_cleanup(backbone)
+        with SafetensorLoader(preset) as loader:
+            convert_weights(backbone, loader)
+    return backbone
+
+
+def load_distilbert_tokenizer(cls, preset):
+    transformers_config = load_config(preset, HF_TOKENIZER_CONFIG_FILE)
+    return cls(
+        get_file(preset, "vocab.txt"),
+        lowercase=transformers_config["do_lower_case"],
+    )

keras_nlp/src/utils/transformers/convert_distilbert_test.py

@@ -0,0 +1,31 @@
+# Copyright 2024 The KerasNLP Authors
+#
+# Licensed 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
+#
+#     https://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.
+import pytest
+
+from keras_nlp.src.models.distil_bert.distil_bert_classifier import (
+    DistilBertClassifier,
+)
+from keras_nlp.src.tests.test_case import TestCase
+
+
+class TestTask(TestCase):
+    @pytest.mark.large
+    def test_convert_tiny_preset(self):
+        model = DistilBertClassifier.from_preset(
+            "hf://distilbert/distilbert-base-uncased", num_classes=2
+        )
+        prompt = "That movies was terrible."
+        model.predict([prompt])
+
+    # TODO: compare numerics with huggingface model