SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Fix TF Flaubert and XLM (#9661)

Browse Source 
* Fix Flaubert and XLM

* Fix Flaubert and XLM

* Apply style
This commit is contained in:
Julien Plu
2021-01-19 18:02:57 +01:00
committed by GitHub
parent 11ec74905a
commit fa876aee2a
2 changed files with 22 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
src/transformers/models/flaubert/modeling_tf_flaubert.py
View File

@@ -214,10 +214,13 @@ class TFFlaubertPreTrainedModel(TFPreTrainedModel):
inputs_list = tf.constant([[7, 6, 0, 0, 1], [1, 2, 3, 0, 0], [0, 0, 0, 4, 5]]) inputs_list = tf.constant([[7, 6, 0, 0, 1], [1, 2, 3, 0, 0], [0, 0, 0, 4, 5]])
attns_list = tf.constant([[1, 1, 0, 0, 1], [1, 1, 1, 0, 0], [1, 0, 0, 1, 1]]) attns_list = tf.constant([[1, 1, 0, 0, 1], [1, 1, 1, 0, 0], [1, 0, 0, 1, 1]])
if self.config.use_lang_emb and self.config.n_langs > 1: if self.config.use_lang_emb and self.config.n_langs > 1:
langs_list = tf.constant([[1, 1, 0, 0, 1], [1, 1, 1, 0, 0], [1, 0, 0, 1, 1]]) return {
"input_ids": inputs_list,
"attention_mask": attns_list,
"langs": tf.constant([[1, 1, 0, 0, 1], [1, 1, 1, 0, 0], [1, 0, 0, 1, 1]]),
}
else: else:
langs_list = None return {"input_ids": inputs_list, "attention_mask": attns_list}
return {"input_ids": inputs_list, "attention_mask": attns_list, "langs": langs_list}
@add_start_docstrings( @add_start_docstrings(

15
src/transformers/models/xlm/modeling_tf_xlm.py
View File

@@ -536,10 +536,13 @@ class TFXLMPreTrainedModel(TFPreTrainedModel):
inputs_list = tf.constant([[7, 6, 0, 0, 1], [1, 2, 3, 0, 0], [0, 0, 0, 4, 5]]) inputs_list = tf.constant([[7, 6, 0, 0, 1], [1, 2, 3, 0, 0], [0, 0, 0, 4, 5]])
attns_list = tf.constant([[1, 1, 0, 0, 1], [1, 1, 1, 0, 0], [1, 0, 0, 1, 1]]) attns_list = tf.constant([[1, 1, 0, 0, 1], [1, 1, 1, 0, 0], [1, 0, 0, 1, 1]])
if self.config.use_lang_emb and self.config.n_langs > 1: if self.config.use_lang_emb and self.config.n_langs > 1:
langs_list = tf.constant([[1, 1, 0, 0, 1], [1, 1, 1, 0, 0], [1, 0, 0, 1, 1]]) return {
"input_ids": inputs_list,
"attention_mask": attns_list,
"langs": tf.constant([[1, 1, 0, 0, 1], [1, 1, 1, 0, 0], [1, 0, 0, 1, 1]]),
}
else: else:
langs_list = None return {"input_ids": inputs_list, "attention_mask": attns_list}
return {"input_ids": inputs_list, "attention_mask": attns_list, "langs": langs_list}
# Remove when XLMWithLMHead computes loss like other LM models # Remove when XLMWithLMHead computes loss like other LM models
@@ -1045,10 +1048,16 @@ class TFXLMForMultipleChoice(TFXLMPreTrainedModel, TFMultipleChoiceLoss):
Returns: Returns:
tf.Tensor with dummy inputs tf.Tensor with dummy inputs
""" """
# Sometimes XLM has language embeddings so don't forget to build them as well if needed
if self.config.use_lang_emb and self.config.n_langs > 1:
return { return {
"input_ids": tf.constant(MULTIPLE_CHOICE_DUMMY_INPUTS), "input_ids": tf.constant(MULTIPLE_CHOICE_DUMMY_INPUTS),
"langs": tf.constant(MULTIPLE_CHOICE_DUMMY_INPUTS), "langs": tf.constant(MULTIPLE_CHOICE_DUMMY_INPUTS),
} }
else:
return {
"input_ids": tf.constant(MULTIPLE_CHOICE_DUMMY_INPUTS),
}
@add_start_docstrings_to_model_forward(XLM_INPUTS_DOCSTRING.format("batch_size, num_choices, sequence_length")) @add_start_docstrings_to_model_forward(XLM_INPUTS_DOCSTRING.format("batch_size, num_choices, sequence_length"))
@add_code_sample_docstrings( @add_code_sample_docstrings(