Remove n_ctx from configs (#14165)

* Remove n_ctx from configs

* Fix GPTJ and OpenAIGPT, both are acceptable breaking changes as there are no configs such that it breaks

* Remove unecessary n_positions from TFOpenAIGPT

tests/test_modeling_tf_openai.py

@@ -98,7 +98,6 @@ class TFOpenAIGPTModelTester:
             # hidden_dropout_prob=self.hidden_dropout_prob,
             # attention_probs_dropout_prob=self.attention_probs_dropout_prob,
             n_positions=self.max_position_embeddings,
-            n_ctx=self.max_position_embeddings,
             # type_vocab_size=self.type_vocab_size,
             # initializer_range=self.initializer_range,
             pad_token_id=self.pad_token_id,

tests/test_trainer.py

@@ -490,7 +490,7 @@ class TrainerIntegrationTest(TestCasePlus, TrainerIntegrationCommon):
         _ = trainer.predict(eval_dataset)
 
     def test_evaluation_with_keys_to_drop(self):
-        config = GPT2Config(vocab_size=100, n_positions=128, n_ctx=128, n_embd=32, n_layer=3, n_head=4)
+        config = GPT2Config(vocab_size=100, n_positions=128, n_embd=32, n_layer=3, n_head=4)
         tiny_gpt2 = GPT2LMHeadModel(config)
         x = torch.randint(0, 100, (128,))
         eval_dataset = RepeatDataset(x)
@@ -531,7 +531,7 @@ class TrainerIntegrationTest(TestCasePlus, TrainerIntegrationCommon):
         self.assertEqual(train_output.global_step, 10)
 
     def test_logging_inf_nan_filter(self):
-        config = GPT2Config(vocab_size=100, n_positions=128, n_ctx=128, n_embd=32, n_layer=3, n_head=4)
+        config = GPT2Config(vocab_size=100, n_positions=128, n_embd=32, n_layer=3, n_head=4)
         tiny_gpt2 = GPT2LMHeadModel(config)
         x = torch.randint(0, 100, (128,))
         train_dataset = RepeatDataset(x)