Add finetuned ernie gram (#1814)

* add finetuned ernie-gram model

* add finetuned ernie-gram model

Co-authored-by: tianxin <[email protected]>

Author: smallv0221

docs/model_zoo/transformers.rst

@@ -329,6 +329,11 @@ Transformer预训练模型汇总
 |ERNIE-GRAM_         |``ernie-gram-zh``                                                                 | Chinese      | 12-layer, 768-hidden,                   |
 |                    |                                                                                  |              | 12-heads, 108M parameters.              |
 |                    |                                                                                  |              | Trained on Chinese text.                |
++                    +----------------------------------------------------------------------------------+--------------+-----------------------------------------+
+|                    |``ernie-gram-zh-finetuned-dureader-robust``                                       | Chinese      | 12-layer, 768-hidden,                   |
+|                    |                                                                                  |              | 12-heads, 108M parameters.              |
+|                    |                                                                                  |              | Trained on Chinese text.                |
+|                    |                                                                                  |              | Then finetuned on dreader-robust        |
 +--------------------+----------------------------------------------------------------------------------+--------------+-----------------------------------------+
 |GPT_                |``gpt-cpm-large-cn``                                                              | Chinese      | 32-layer, 2560-hidden,                  |
 |                    |                                                                                  |              | 32-heads, 2.6B parameters.              |

paddlenlp/transformers/ernie_gram/modeling.py

@@ -98,12 +98,27 @@ class ErnieGramPretrainedModel(PretrainedModel):
             "type_vocab_size": 2,
             "vocab_size": 18018
         },
+        "ernie-gram-zh-finetuned-dureader-robust": {
+            "attention_probs_dropout_prob": 0.1,
+            "emb_size": 768,
+            "hidden_act": "gelu",
+            "hidden_dropout_prob": 0.1,
+            "hidden_size": 768,
+            "initializer_range": 0.02,
+            "max_position_embeddings": 512,
+            "num_attention_heads": 12,
+            "num_hidden_layers": 12,
+            "type_vocab_size": 2,
+            "vocab_size": 18018
+        },
     }
     resource_files_names = {"model_state": "model_state.pdparams"}
     pretrained_resource_files_map = {
         "model_state": {
             "ernie-gram-zh":
             "https://bj.bcebos.com/paddlenlp/models/transformers/ernie_gram_zh/ernie_gram_zh.pdparams",
+            "ernie-gram-zh-finetuned-dureader-robust":
+            "https://bj.bcebos.com/paddlenlp/models/transformers/ernie-gram-zh-finetuned-dureader-robust/model_state.pdparams",
         },
     }
     base_model_prefix = "ernie_gram"

paddlenlp/transformers/ernie_gram/tokenizer.py

@@ -77,9 +77,18 @@ class ErnieGramTokenizer(ErnieTokenizer):
         "vocab_file": {
             "ernie-gram-zh":
             "https://bj.bcebos.com/paddlenlp/models/transformers/ernie_gram_zh/vocab.txt",
+            "ernie-gram-zh-finetuned-dureader-robust":
+            "https://bj.bcebos.com/paddlenlp/models/transformers/ernie_gram_zh/vocab.txt",
         }
     }
-    pretrained_init_configuration = {"ernie-gram-zh": {"do_lower_case": True}, }
+    pretrained_init_configuration = {
+        "ernie-gram-zh": {
+            "do_lower_case": True
+        },
+        "ernie-gram-zh-finetuned-dureader-robust": {
+            "do_lower_case": True
+        },
+    }
 
     def __init__(self,
                  vocab_file,