cleaning up configuration classes

transformers/modeling_transfo_xl.py

@@ -592,14 +592,14 @@ class TransfoXLModel(TransfoXLPreTrainedModel):
         self.output_attentions = config.output_attentions
         self.output_hidden_states = config.output_hidden_states
 
-        self.n_token = config.n_token
+        self.n_token = config.vocab_size
 
         self.d_embed = config.d_embed
         self.d_model = config.d_model
         self.n_head = config.n_head
         self.d_head = config.d_head
 
-        self.word_emb = AdaptiveEmbedding(config.n_token, config.d_embed, config.d_model, config.cutoffs,
+        self.word_emb = AdaptiveEmbedding(config.vocab_size, config.d_embed, config.d_model, config.cutoffs,
                                           div_val=config.div_val)
 
         self.drop = nn.Dropout(config.dropout)
@@ -836,11 +836,11 @@ class TransfoXLLMHeadModel(TransfoXLPreTrainedModel):
         self.sample_softmax = config.sample_softmax
         # use sampled softmax
         if config.sample_softmax > 0:
-            self.out_layer = nn.Linear(config.d_model, config.n_token)
-            self.sampler = LogUniformSampler(config.n_token, config.sample_softmax)
+            self.out_layer = nn.Linear(config.d_model, config.vocab_size)
+            self.sampler = LogUniformSampler(config.vocab_size, config.sample_softmax)
         # use adaptive softmax (including standard softmax)
         else:
-            self.crit = ProjectedAdaptiveLogSoftmax(config.n_token, config.d_embed, config.d_model,
+            self.crit = ProjectedAdaptiveLogSoftmax(config.vocab_size, config.d_embed, config.d_model,
                                                     config.cutoffs, div_val=config.div_val)
         self.init_weights()
 

transformers/modeling_xlnet.py

@@ -609,7 +609,7 @@ class XLNetModel(XLNetPreTrainedModel):
         self.clamp_len = config.clamp_len
         self.n_layer = config.n_layer
 
-        self.word_embedding = nn.Embedding(config.n_token, config.d_model)
+        self.word_embedding = nn.Embedding(config.vocab_size, config.d_model)
         self.mask_emb = nn.Parameter(torch.FloatTensor(1, 1, config.d_model))
         self.layer = nn.ModuleList([XLNetLayer(config) for _ in range(config.n_layer)])
         self.dropout = nn.Dropout(config.dropout)
@@ -940,7 +940,7 @@ class XLNetLMHeadModel(XLNetPreTrainedModel):
         self.same_length = config.same_length
 
         self.transformer = XLNetModel(config)
-        self.lm_loss = nn.Linear(config.d_model, config.n_token, bias=True)
+        self.lm_loss = nn.Linear(config.d_model, config.vocab_size, bias=True)
 
         self.init_weights()
 

transformers/tests/modeling_albert_test.py

@@ -110,7 +110,7 @@ class AlbertModelTest(CommonTestCases.CommonModelTester):
                 choice_labels = ids_tensor([self.batch_size], self.num_choices)
 
             config = AlbertConfig(
-                vocab_size_or_config_json_file=self.vocab_size,
+                vocab_size=self.vocab_size,
                 hidden_size=self.hidden_size,
                 num_hidden_layers=self.num_hidden_layers,
                 num_attention_heads=self.num_attention_heads,

transformers/tests/modeling_bert_test.py

@@ -109,7 +109,7 @@ class BertModelTest(CommonTestCases.CommonModelTester):
                 choice_labels = ids_tensor([self.batch_size], self.num_choices)
 
             config = BertConfig(
-                vocab_size_or_config_json_file=self.vocab_size,
+                vocab_size=self.vocab_size,
                 hidden_size=self.hidden_size,
                 num_hidden_layers=self.num_hidden_layers,
                 num_attention_heads=self.num_attention_heads,

transformers/tests/modeling_common_test.py

@@ -633,7 +633,7 @@ class CommonTestCases:
                 mc_token_ids = ids_tensor([self.batch_size, self.n_choices], self.seq_length)
 
             config = self.config_class(
-                vocab_size_or_config_json_file=self.vocab_size,
+                vocab_size=self.vocab_size,
                 n_positions=self.n_positions,
                 n_embd=self.hidden_size,
                 n_layer=self.num_hidden_layers,

transformers/tests/modeling_ctrl_test.py

@@ -114,7 +114,7 @@ class CTRLModelTest(CommonTestCases.CommonModelTester):
                 choice_labels = ids_tensor([self.batch_size], self.num_choices)
 
             config = CTRLConfig(
-                vocab_size_or_config_json_file=self.vocab_size,
+                vocab_size=self.vocab_size,
                 n_embd=self.hidden_size,
                 n_layer=self.num_hidden_layers,
                 n_head=self.num_attention_heads,

transformers/tests/modeling_distilbert_test.py

@@ -105,7 +105,7 @@ class DistilBertModelTest(CommonTestCases.CommonModelTester):
                 choice_labels = ids_tensor([self.batch_size], self.num_choices)
 
             config = DistilBertConfig(
-                vocab_size_or_config_json_file=self.vocab_size,
+                vocab_size=self.vocab_size,
                 dim=self.hidden_size,
                 n_layers=self.num_hidden_layers,
                 n_heads=self.num_attention_heads,

transformers/tests/modeling_gpt2_test.py

@@ -110,7 +110,7 @@ class GPT2ModelTest(CommonTestCases.CommonModelTester):
                 choice_labels = ids_tensor([self.batch_size], self.num_choices)
 
             config = GPT2Config(
-                vocab_size_or_config_json_file=self.vocab_size,
+                vocab_size=self.vocab_size,
                 n_embd=self.hidden_size,
                 n_layer=self.num_hidden_layers,
                 n_head=self.num_attention_heads,

transformers/tests/modeling_openai_test.py

@@ -98,7 +98,7 @@ class OpenAIGPTModelTest(CommonTestCases.CommonModelTester):
                 choice_labels = ids_tensor([self.batch_size], self.num_choices)
 
             config = OpenAIGPTConfig(
-                vocab_size_or_config_json_file=self.vocab_size,
+                vocab_size=self.vocab_size,
                 n_embd=self.hidden_size,
                 n_layer=self.num_hidden_layers,
                 n_head=self.num_attention_heads,

transformers/tests/modeling_roberta_test.py

@@ -106,7 +106,7 @@ class RobertaModelTest(CommonTestCases.CommonModelTester):
                 choice_labels = ids_tensor([self.batch_size], self.num_choices)
 
             config = RobertaConfig(
-                vocab_size_or_config_json_file=self.vocab_size,
+                vocab_size=self.vocab_size,
                 hidden_size=self.hidden_size,
                 num_hidden_layers=self.num_hidden_layers,
                 num_attention_heads=self.num_attention_heads,

transformers/tests/modeling_tf_albert_test.py

@@ -118,7 +118,7 @@ class TFAlbertModelTest(TFCommonTestCases.TFCommonModelTester):
                 choice_labels = ids_tensor([self.batch_size], self.num_choices)
 
             config = AlbertConfig(
-                vocab_size_or_config_json_file=self.vocab_size,
+                vocab_size=self.vocab_size,
                 hidden_size=self.hidden_size,
                 num_hidden_layers=self.num_hidden_layers,
                 num_attention_heads=self.num_attention_heads,

transformers/tests/modeling_tf_bert_test.py

@@ -114,7 +114,7 @@ class TFBertModelTest(TFCommonTestCases.TFCommonModelTester):
                 choice_labels = ids_tensor([self.batch_size], self.num_choices)
 
             config = BertConfig(
-                vocab_size_or_config_json_file=self.vocab_size,
+                vocab_size=self.vocab_size,
                 hidden_size=self.hidden_size,
                 num_hidden_layers=self.num_hidden_layers,
                 num_attention_heads=self.num_attention_heads,

transformers/tests/modeling_tf_ctrl_test.py

@@ -112,7 +112,7 @@ class TFCTRLModelTest(TFCommonTestCases.TFCommonModelTester):
                 choice_labels = ids_tensor([self.batch_size], self.num_choices)
 
             config = CTRLConfig(
-                vocab_size_or_config_json_file=self.vocab_size,
+                vocab_size=self.vocab_size,
                 n_embd=self.hidden_size,
                 n_layer=self.num_hidden_layers,
                 n_head=self.num_attention_heads,

transformers/tests/modeling_tf_distilbert_test.py

@@ -107,7 +107,7 @@ class TFDistilBertModelTest(TFCommonTestCases.TFCommonModelTester):
                 choice_labels = ids_tensor([self.batch_size], self.num_choices)
 
             config = DistilBertConfig(
-                vocab_size_or_config_json_file=self.vocab_size,
+                vocab_size=self.vocab_size,
                 dim=self.hidden_size,
                 n_layers=self.num_hidden_layers,
                 n_heads=self.num_attention_heads,

transformers/tests/modeling_tf_gpt2_test.py

@@ -115,7 +115,7 @@ class TFGPT2ModelTest(TFCommonTestCases.TFCommonModelTester):
                 choice_labels = ids_tensor([self.batch_size], self.num_choices)
 
             config = GPT2Config(
-                vocab_size_or_config_json_file=self.vocab_size,
+                vocab_size=self.vocab_size,
                 n_embd=self.hidden_size,
                 n_layer=self.num_hidden_layers,
                 n_head=self.num_attention_heads,

transformers/tests/modeling_tf_openai_gpt_test.py

@@ -114,7 +114,7 @@ class TFOpenAIGPTModelTest(TFCommonTestCases.TFCommonModelTester):
                 choice_labels = ids_tensor([self.batch_size], self.num_choices)
 
             config = OpenAIGPTConfig(
-                vocab_size_or_config_json_file=self.vocab_size,
+                vocab_size=self.vocab_size,
                 n_embd=self.hidden_size,
                 n_layer=self.num_hidden_layers,
                 n_head=self.num_attention_heads,

transformers/tests/modeling_tf_roberta_test.py

@@ -109,7 +109,7 @@ class TFRobertaModelTest(TFCommonTestCases.TFCommonModelTester):
                 choice_labels = ids_tensor([self.batch_size], self.num_choices)
 
             config = RobertaConfig(
-                vocab_size_or_config_json_file=self.vocab_size,
+                vocab_size=self.vocab_size,
                 hidden_size=self.hidden_size,
                 num_hidden_layers=self.num_hidden_layers,
                 num_attention_heads=self.num_attention_heads,

transformers/tests/modeling_tf_transfo_xl_test.py

@@ -92,7 +92,7 @@ class TFTransfoXLModelTest(TFCommonTestCases.TFCommonModelTester):
                 lm_labels = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
 
             config = TransfoXLConfig(
-                vocab_size_or_config_json_file=self.vocab_size,
+                vocab_size=self.vocab_size,
                 mem_len=self.mem_len,
                 clamp_len=self.clamp_len,
                 cutoffs=self.cutoffs,

transformers/tests/modeling_tf_xlm_test.py

@@ -125,7 +125,7 @@ class TFXLMModelTest(TFCommonTestCases.TFCommonModelTester):
                 is_impossible_labels = ids_tensor([self.batch_size], 2, dtype=tf.float32)
 
             config = XLMConfig(
-                 vocab_size_or_config_json_file=self.vocab_size,
+                 vocab_size=self.vocab_size,
                  n_special=self.n_special,
                  emb_dim=self.hidden_size,
                  n_layers=self.num_hidden_layers,

transformers/tests/modeling_tf_xlnet_test.py

@@ -64,7 +64,6 @@ class TFXLNetModelTest(TFCommonTestCases.TFCommonModelTester):
                      num_attention_heads=4,
                      d_inner=128,
                      num_hidden_layers=5,
-                     max_position_embeddings=10,
                      type_sequence_label_size=2,
                      untie_r=True,
                      bi_data=False,
@@ -88,7 +87,6 @@ class TFXLNetModelTest(TFCommonTestCases.TFCommonModelTester):
             self.num_attention_heads = num_attention_heads
             self.d_inner = d_inner
             self.num_hidden_layers = num_hidden_layers
-            self.max_position_embeddings = max_position_embeddings
             self.bi_data = bi_data
             self.untie_r = untie_r
             self.same_length = same_length
@@ -122,13 +120,12 @@ class TFXLNetModelTest(TFCommonTestCases.TFCommonModelTester):
                 is_impossible_labels = ids_tensor([self.batch_size], 2, dtype=tf.float32)
 
             config = XLNetConfig(
-                vocab_size_or_config_json_file=self.vocab_size,
+                vocab_size=self.vocab_size,
                 d_model=self.hidden_size,
                 n_head=self.num_attention_heads,
                 d_inner=self.d_inner,
                 n_layer=self.num_hidden_layers,
                 untie_r=self.untie_r,
-                max_position_embeddings=self.max_position_embeddings,
                 mem_len=self.mem_len,
                 clamp_len=self.clamp_len,
                 same_length=self.same_length,