TransfoXL is deprecated, don't keep it in tested examples! (#37707)

* TransfoXL is deprecated, so we should remove it from examples that get tested

* Remove the tokenizer too

* Trigger tests

examples/pytorch/text-generation/run_generation.py

@@ -38,8 +38,6 @@ from transformers import (
     OpenAIGPTLMHeadModel,
     OpenAIGPTTokenizer,
     OPTForCausalLM,
-    TransfoXLLMHeadModel,
-    TransfoXLTokenizer,
     XLMTokenizer,
     XLMWithLMHeadModel,
     XLNetLMHeadModel,
@@ -62,7 +60,6 @@ MODEL_CLASSES = {
     "ctrl": (CTRLLMHeadModel, CTRLTokenizer),
     "openai-gpt": (OpenAIGPTLMHeadModel, OpenAIGPTTokenizer),
     "xlnet": (XLNetLMHeadModel, XLNetTokenizer),
-    "transfo-xl": (TransfoXLLMHeadModel, TransfoXLTokenizer),
     "xlm": (XLMWithLMHeadModel, XLMTokenizer),
     "gptj": (GPTJForCausalLM, AutoTokenizer),
     "bloom": (BloomForCausalLM, BloomTokenizerFast),
@@ -368,10 +365,7 @@ def main():
         prepare_input = PREPROCESSING_FUNCTIONS.get(args.model_type)
         preprocessed_prompt_text = prepare_input(args, model, tokenizer, prompt_text)
 
-        if model.__class__.__name__ in ["TransfoXLLMHeadModel"]:
-            tokenizer_kwargs = {"add_space_before_punct_symbol": True}
-        else:
-            tokenizer_kwargs = {}
+        tokenizer_kwargs = {}
 
         encoded_prompt = tokenizer.encode(
             preprocessed_prompt_text, add_special_tokens=False, return_tensors="pt", **tokenizer_kwargs