[prompt] remove unused arguments (#3794)

* [prompt] delete unused arguments

* [few-shot] update script for fine-tune

Author: LemonNoel

applications/text_classification/hierarchical/few-shot/README.md

@@ -77,6 +77,7 @@
   - 微调
 
     ```
+    cd ../
     python train.py --dataset_dir "./data/" --save_dir "./checkpoints" --max_seq_length 128 --model_name "ernie-3.0-base-zh" --batch_size 8 --learning_rate 3e-5 --epochs 100 --logging_steps 5 --early_stop --early_stop_num 20
     ```
 

applications/text_classification/multi_class/few-shot/README.md

@@ -76,6 +76,7 @@
 - 微调
 
 ```
+cd ../
 python train.py --dataset_dir "./data/" --save_dir "./checkpoints" --max_seq_length 128 --model_name "ernie-3.0-base-zh" --batch_size 8 --learning_rate 3e-5 --epochs 100 --logging_steps 5 --early_stop
 ```
 

applications/text_classification/multi_label/few-shot/README.md

@@ -82,6 +82,7 @@
   - 微调
 
     ```
+    cd ../
     python train.py --dataset_dir "./data/" --save_dir "./checkpoints" --max_seq_length 128 --model_name "ernie-3.0-base-zh" --batch_size 8 --learning_rate 3e-5 --epochs 100 --logging_steps 5 --early_stop --early_stop_num 10
     ```
 

paddlenlp/prompt/prompt_args.py

@@ -26,29 +26,8 @@ class PromptTuningArguments(TrainingArguments):
     The arguments' subset for training loop during prompt tuning.
     """
 
-    task_type: str = field(
-        default="multi-class",
-        metadata={"help": "The task type of prompt tuning. Support "\
-                          "`multi-class`, `multi-label` and `hierachical`."})
-
     max_seq_length: int = field(
         default=512, metadata={"help": "The maximum length of all input text."})
-    first_max_length: int = field(
-        default=None,
-        metadata={"help": "The maximum length of the first sentence."})
-    other_max_length: int = field(
-        default=None,
-        metadata={"help": "The maximum length of other sentences."})
-    truncate_mode: str = field(
-        default="tail",
-        metadata={
-            "help":
-            "How to truncate input text, `head`, `tail` or "
-            "`manual`. Truncate input text according to "
-            "`first_max_length` and `other_max_length` in the "
-            "manual mode."
-        })
-
     freeze_plm: bool = field(
         default=False,
         metadata={"help": "If True, the pretrained parameters won't be "\