Add unit tests for T5 (#3115)

paddlenlp/transformers/generation_utils.py

@@ -422,7 +422,8 @@ def update_model_kwargs_for_generation(outputs,
         # method.
 
         # update cache
-        if isinstance(outputs, tuple):
+        if isinstance(outputs,
+                      tuple) and not isinstance(outputs[1], paddle.Tensor):
             model_kwargs["cache"] = outputs[1]
 
         # update token_type_ids with last value

paddlenlp/transformers/t5/modeling.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
 # Copyright 2018 Mesh TensorFlow authors, T5 Authors and HuggingFace Inc. team.
 #
@@ -31,6 +30,12 @@
     'T5ForConditionalGeneration',
 ]
 
+T5_PRETRAINED_MODEL_ARCHIVE_LIST = [
+    "t5-small",
+    "t5-base",
+    "t5-large",
+]
+
 
 def finfo(dtype):
     if dtype == paddle.float32:
@@ -107,6 +112,27 @@ def forward(self, hidden_states):
         return hidden_states
 
 
+class T5DenseGatedSiluDense(nn.Layer):
+    """
+    Construct a dense-gated_gelu-dense module.
+    """
+
+    def __init__(self, d_model, d_ff, dropout_rate):
+        super().__init__()
+        self.wi_0 = nn.Linear(d_model, d_ff, bias_attr=False)
+        self.wi_1 = nn.Linear(d_model, d_ff, bias_attr=False)
+        self.wo = nn.Linear(d_ff, d_model, bias_attr=False)
+        self.dropout = nn.Dropout(dropout_rate)
+
+    def forward(self, hidden_states):
+        hidden_silu = F.silu(self.wi_0(hidden_states))
+        hidden_linear = self.wi_1(hidden_states)
+        hidden_states = hidden_silu * hidden_linear
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.wo(hidden_states)
+        return hidden_states
+
+
 class T5LayerFF(nn.Layer):
 
     def __init__(self, feed_forward_proj, d_model, d_ff, layer_norm_epsilon,
@@ -117,6 +143,9 @@ def __init__(self, feed_forward_proj, d_model, d_ff, layer_norm_epsilon,
         elif feed_forward_proj == "gated-gelu":
             self.DenseReluDense = T5DenseGatedGeluDense(d_model, d_ff,
                                                         dropout_rate)
+        elif feed_forward_proj == "gated-silu":
+            self.DenseReluDense = T5DenseGatedSiluDense(d_model, d_ff,
+                                                        dropout_rate)
         else:
             raise ValueError(
                 f"{feed_forward_proj} is not supported. Choose between `relu` and `gated-gelu`"
@@ -522,6 +551,7 @@ def forward(
             output_attentions=output_attentions,
         )
         hidden_states, present_key_value_state = self_attention_outputs[:2]
+
         attention_outputs = self_attention_outputs[
             2:]  # Keep self-attention outputs and relative position weights
 
@@ -989,7 +1019,7 @@ def forward(self,
 
             # layer_outputs is a tuple with:
             # hidden-states, key-value-states, (self-attention position bias), (self-attention weights), (cross-attention position bias), (cross-attention weights)
-            if use_cache is False:
+            if not use_cache:
                 layer_outputs = layer_outputs[:1] + (None, ) + layer_outputs[1:]
 
             hidden_states, present_key_value_state = layer_outputs[:2]
@@ -1040,8 +1070,6 @@ def get_extended_attention_mask(self, attention_mask, input_shape):
                 causal_mask = paddle.tile(seq_ids.unsqueeze(axis=[0, 1]),
                                           [batch_size, seq_length, 1
                                            ]) <= seq_ids.unsqueeze(axis=[0, 2])
-                # in case cache are used we need to add a prefix ones mask to the causal mask
-                # causal and attention masks must have same type with pytorch version < 1.3
                 causal_mask = causal_mask.astype(attention_mask.dtype)
 
                 if causal_mask.shape[1] < attention_mask.shape[1]:
@@ -1062,6 +1090,35 @@ def get_extended_attention_mask(self, attention_mask, input_shape):
                     1) * attention_mask.unsqueeze([1, 2])
             else:
                 extended_attention_mask = attention_mask.unsqueeze([1, 2])
+        elif attention_mask.ndim == 4:
+            if self.is_decoder:
+                batch_size, seq_length = input_shape
+                seq_ids = paddle.arange(seq_length)
+                causal_mask = paddle.tile(seq_ids.unsqueeze(axis=[0, 1]),
+                                          [batch_size, seq_length, 1
+                                           ]) <= seq_ids.unsqueeze(axis=[0, 2])
+                # in case cache are used we need to add a prefix ones mask to the causal mask
+                # causal and attention masks must have same type with pytorch version < 1.3
+                causal_mask = causal_mask.astype(attention_mask.dtype)
+
+                if causal_mask.shape[1] < attention_mask.shape[-1]:
+                    prefix_seq_len = attention_mask.shape[
+                        1] - causal_mask.shape[1]
+                    causal_mask = paddle.concat(
+                        [
+                            paddle.ones(
+                                [batch_size, seq_length, prefix_seq_len],
+                                dtype=causal_mask.dtype,
+                            ),
+                            causal_mask,
+                        ],
+                        axis=-1,
+                    )
+
+                extended_attention_mask = causal_mask.unsqueeze(
+                    1) * attention_mask
+            else:
+                extended_attention_mask = attention_mask
         else:
             raise ValueError(
                 f"Wrong shape for input_ids (shape {input_shape}) or attention_mask (shape {attention_mask.shape})"
@@ -1072,10 +1129,12 @@ def get_extended_attention_mask(self, attention_mask, input_shape):
         return extended_attention_mask
 
     def invert_attention_mask(self, encoder_attention_mask):
-        if encoder_attention_mask.ndim == 3:
+        if encoder_attention_mask.ndim == 4:
+            encoder_extended_attention_mask = encoder_attention_mask
+        elif encoder_attention_mask.ndim == 3:
             encoder_extended_attention_mask = encoder_attention_mask.unsqueeze(
                 1)
-        if encoder_attention_mask.ndim == 2:
+        elif encoder_attention_mask.ndim == 2:
             encoder_extended_attention_mask = encoder_attention_mask.unsqueeze(
                 [1, 2])
         encoder_extended_attention_mask = encoder_extended_attention_mask.astype(
@@ -1176,6 +1235,13 @@ def __init__(self,
         self.d_model = d_model
         self.initializer_factor = initializer_factor
 
+        if num_decoder_layers is None and num_layers is None:
+            raise ValueError(
+                "You have to specify either num_decoder_layers or num_layers or both."
+            )
+        elif num_decoder_layers is None:
+            num_decoder_layers = num_layers
+
         self.shared = nn.Embedding(vocab_size, d_model)
         self.encoder = T5Stack(d_model,
                                num_layers,
@@ -1401,9 +1467,10 @@ def set_output_embeddings(self, new_embeddings):
         self.lm_head = new_embeddings
 
     def get_output_embeddings(self):
-        if not self.t5.config["tie_word_embeddings"]:
+        if self.t5.config["tie_word_embeddings"]:
             return self.t5.shared
-        return self.lm_head
+        else:
+            return self.lm_head
 
     def get_encoder(self):
         return self.t5.encoder
@@ -1514,7 +1581,10 @@ def forward(self,
                 output_attentions=output_attentions,
                 output_hidden_states=output_hidden_states)
 
-        hidden_states = encoder_output[0]
+        if isinstance(encoder_output, (tuple, list)):
+            hidden_states = encoder_output[0]
+        else:
+            hidden_states = encoder_output
 
         if labels is not None and decoder_input_ids is None:
             # get decoder inputs from shifting lm labels to the right
@@ -1559,6 +1629,9 @@ def forward(self,
             loss = loss_fct(lm_logits.reshape(shape=[-1, lm_logits.shape[-1]]),
                             labels.flatten())
 
+        if not isinstance(encoder_output, (list, tuple)):
+            encoder_output = (encoder_output, )
+
         output = (lm_logits, ) + decoder_outputs[1:] + encoder_output
         return ((loss, ) + output) if loss is not None else output
 

paddlenlp/transformers/t5/tokenizer.py

@@ -24,6 +24,12 @@
     'T5Tokenizer',
 ]
 
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    "t5-small": 512,
+    "t5-base": 512,
+    "t5-large": 512,
+}
+
 
 class T5Tokenizer(AlbertEnglishTokenizer):
     """
@@ -88,6 +94,8 @@ class T5Tokenizer(AlbertEnglishTokenizer):
         },
     }
 
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+
     def __init__(self,
                  sentencepiece_model_file,
                  do_lower_case=False,
@@ -98,6 +106,7 @@ def __init__(self,
                  pad_token="<pad>",
                  extra_ids=100,
                  additional_special_tokens=[],
+                 sp_model_kwargs=None,
                  **kwargs):
 
         # Add extra_ids to the special token list
@@ -123,28 +132,54 @@ def __init__(self,
         self.extra_ids = extra_ids
         self.sentencepiece_model_file = sentencepiece_model_file
 
-        self.sp_model = spm.SentencePieceProcessor()
+        self.sp_model_kwargs = {} if sp_model_kwargs is None else sp_model_kwargs
+
+        self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
         self.sp_model.Load(sentencepiece_model_file)
 
     def __call__(self,
                  text,
                  text_pair=None,
-                 max_seq_len=None,
+                 max_length=None,
                  stride=0,
                  is_split_into_words=False,
-                 pad_to_max_seq_len=False,
-                 truncation_strategy="longest_first",
+                 padding=None,
+                 truncation="longest_first",
                  return_position_ids=False,
                  return_token_type_ids=False,
                  return_attention_mask=True,
                  return_length=False,
                  return_overflowing_tokens=False,
-                 return_special_tokens_mask=False):
+                 return_special_tokens_mask=False,
+                 **kwargs):
+        if "pad_to_max_seq_len" in kwargs and padding is None:
+            pad_to_max_seq_len = kwargs.pop("pad_to_max_seq_len")
+            padding = "max_length" if pad_to_max_seq_len else False
+        elif padding is None:
+            padding = False
+
+        if "max_seq_len" in kwargs and max_length is None:
+            max_length = kwargs["max_seq_len"]
+
+        if "truncation_strategy" in kwargs and kwargs[
+                "truncation_strategy"] != "longest_first":
+            truncation = kwargs["truncation_strategy"]
+
         return super(T5Tokenizer, self).__call__(
-            text, text_pair, max_seq_len, stride, is_split_into_words,
-            pad_to_max_seq_len, truncation_strategy, return_position_ids,
-            return_token_type_ids, return_attention_mask, return_length,
-            return_overflowing_tokens, return_special_tokens_mask)
+            text=text,
+            text_pair=text_pair,
+            max_length=max_length,
+            stride=stride,
+            is_split_into_words=is_split_into_words,
+            padding=padding,
+            truncation=truncation,
+            return_position_ids=return_position_ids,
+            return_token_type_ids=return_token_type_ids,
+            return_attention_mask=return_attention_mask,
+            return_length=return_length,
+            return_overflowing_tokens=return_overflowing_tokens,
+            return_special_tokens_mask=return_special_tokens_mask,
+            **kwargs)
 
     @property
     def vocab_size(self):
@@ -254,36 +289,6 @@ def convert_tokens_to_string(self, tokens):
         out_string += self.sp_model.decode_pieces(current_sub_tokens)
         return out_string.strip()
 
-    def decode(self,
-               token_ids,
-               skip_special_tokens=False,
-               clean_up_tokenization_spaces=True):
-        """
-        Converts a sequence of ids in a string, using the tokenizer and vocabulary with options to remove special
-        tokens and clean up tokenization spaces.
-
-        Similar to doing ``self.convert_tokens_to_string(self.convert_ids_to_tokens(token_ids))``.
-
-        Args:
-            token_ids (Union[List[int], Tensor]):
-                List of tokenized input ids. 
-            skip_special_tokens (bool, optional):
-                Whether or not to remove special tokens in the decoding. Defaults to `False`.
-            clean_up_tokenization_spaces (bool, optional):
-                Whether or not to clean up the tokenization spaces. Defaults to `True`.
-
-        Returns:
-            str: The decoded sentence.
-        """
-        if hasattr(token_ids, "tolist"):
-            token_ids = token_ids.tolist()
-        text = self.convert_tokens_to_string(
-            self.convert_ids_to_tokens(token_ids,
-                                       skip_special_tokens=skip_special_tokens))
-        if clean_up_tokenization_spaces:
-            text = self.clean_up_tokenization(text)
-        return text
-
     def _convert_token_to_id(self, token):
         """Converts a token (str) in an id using the vocab."""
         if token.startswith("<extra_id_"):
@@ -343,3 +348,18 @@ def clean_up_tokenization(out_string):
                 "n't").replace(" 'm", "'m").replace(" 's", "'s").replace(
                     " 've", "'ve").replace(" 're", "'re"))
         return out_string
+
+    def __getstate__(self):
+        state = self.__dict__.copy()
+        state["sp_model"] = None
+        return state
+
+    def __setstate__(self, d):
+        self.__dict__ = d
+
+        # for backward compatibility
+        if not hasattr(self, "sp_model_kwargs"):
+            self.sp_model_kwargs = {}
+
+        self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
+        self.sp_model.Load(self.sentencepiece_model_file)

tests/transformers/bart/test_modeling.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
 # Copyright 2021, The HuggingFace Inc. team. All rights reserved.
 #

tests/transformers/gpt/test_modeling.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
 # Copyright 2020 The HuggingFace Team. All rights reserved.
 #

tests/transformers/gpt/test_tokenizer.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
 # Copyright 2020 The HuggingFace Team. All rights reserved.
 #

tests/transformers/t5/__init__.py

@@ -0,0 +1,13 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# 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
+#
+#     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.