[chatglm] set default max_seq_length to 2048 (PaddlePaddle#5970)

w5688414 · web-flow · commit a0cdff3cfca4 · 2023-05-19T17:46:00.000+08:00
* set max_seq_length to 2048

* Add THUDM/chatglm-6b-v1.1

* Add chatglm v1.1 config
diff --git a/paddlenlp/taskflow/taskflow.py b/paddlenlp/taskflow/taskflow.py
@@ -473,8 +473,12 @@
                 "task_class": ChatGLMTask,
                 "task_flag": "text_generation-tiny-random-chatglm",
             },
+            "THUDM/chatglm-6b-v1.1": {
+                "task_class": ChatGLMTask,
+                "task_flag": "text_generation-THUDM/chatglm-6b-v1.1",
+            },
         },
-        "default": {"model": "THUDM/chatglm-6b"},
+        "default": {"model": "THUDM/chatglm-6b-v1.1"},
     },
     "zero_shot_text_classification": {
         "models": {
@@ -727,6 +731,7 @@
     "__internal_testing__/tiny-random-uie-m",
     "__internal_testing__/tiny-random-uie-x",
     "THUDM/chatglm-6b",
+    "THUDM/chatglm-6b-v1.1",
 ]
 
 
diff --git a/paddlenlp/taskflow/text2text_generation.py b/paddlenlp/taskflow/text2text_generation.py
@@ -35,9 +35,9 @@ def __init__(self, task, model, **kwargs):
         self._static_mode = False
         self._dtype = kwargs.get("dtype", "float16")
         self.kwargs["generation_task"] = task
-        self._tgt_length = kwargs.get("tgt_length", 128)
+        self._tgt_length = kwargs.get("tgt_length", 2048)
         # Token max length
-        self._max_seq_length = kwargs.get("max_seq_length", 128)
+        self._max_seq_length = kwargs.get("max_seq_length", 2048)
         self._top_k = kwargs.get("top_k", 1)
         self._top_p = kwargs.get("top_p", 1.0)
         self._temperature = kwargs.get("temperature", 1.0)
diff --git a/paddlenlp/transformers/chatglm/configuration.py b/paddlenlp/transformers/chatglm/configuration.py
@@ -24,7 +24,8 @@
 
 CHATGLM_PRETRAINED_RESOURCE_FILES_MAP = {
     "model_state": {
-        "THUDM/chatglm-6b": "https://paddlenlp.bj.bcebos.com/models/community/THUDM/chatglm-6b/model_state.pdparams"
+        "THUDM/chatglm-6b": "https://paddlenlp.bj.bcebos.com/models/community/THUDM/chatglm-6b/model_state.pdparams",
+        "THUDM/chatglm-6b-v1.1": "https://paddlenlp.bj.bcebos.com/models/community/THUDM/chatglm-6b-v1.1/model_state.pdparams",
     }
 }
 
diff --git a/paddlenlp/transformers/chatglm/tokenizer.py b/paddlenlp/transformers/chatglm/tokenizer.py
@@ -165,11 +165,12 @@ class ChatGLMTokenizer(PretrainedTokenizer):
     """
 
     resource_files_names = {"vocab_file": "ice_text.model"}
-    max_model_input_sizes = {"THUDM/chatglm-6b": 2048}
+    max_model_input_sizes = {"THUDM/chatglm-6b": 2048, "THUDM/chatglm-6b-v1.1": 2048}
     model_input_names = ["input_ids", "attention_mask", "position_ids"]
     pretrained_resource_files_map = {
         "model_file": {
-            "THUDM/chatglm-6b": "https://paddlenlp.bj.bcebos.com/models/community/THUDM/chatglm-6b/ice_text.model"
+            "THUDM/chatglm-6b": "https://paddlenlp.bj.bcebos.com/models/community/THUDM/chatglm-6b/ice_text.model",
+            "THUDM/chatglm-6b-v1.1": "https://paddlenlp.bj.bcebos.com/models/community/THUDM/chatglm-6b-v1.1/ice_text.model",
         }
     }
 

Original file line number	Diff line number	Diff line change
`@@ -24,7 +24,8 @@`
`24`	`24`
`25`	`25`	`CHATGLM_PRETRAINED_RESOURCE_FILES_MAP = {`
`26`	`26`	`"model_state": {`
`27`		`- "THUDM/chatglm-6b": "https://paddlenlp.bj.bcebos.com/models/community/THUDM/chatglm-6b/model_state.pdparams"`
	`27`	`+ "THUDM/chatglm-6b": "https://paddlenlp.bj.bcebos.com/models/community/THUDM/chatglm-6b/model_state.pdparams",`
	`28`	`+ "THUDM/chatglm-6b-v1.1": "https://paddlenlp.bj.bcebos.com/models/community/THUDM/chatglm-6b-v1.1/model_state.pdparams",`
`28`	`29`	`}`
`29`	`30`	`}`
`30`	`31`
Original file line number	Diff line number	Diff line change
`@@ -165,11 +165,12 @@ class ChatGLMTokenizer(PretrainedTokenizer):`
`165`	`165`	`"""`
`166`	`166`
`167`	`167`	`resource_files_names = {"vocab_file": "ice_text.model"}`
`168`		`- max_model_input_sizes = {"THUDM/chatglm-6b": 2048}`
	`168`	`+ max_model_input_sizes = {"THUDM/chatglm-6b": 2048, "THUDM/chatglm-6b-v1.1": 2048}`
`169`	`169`	`model_input_names = ["input_ids", "attention_mask", "position_ids"]`
`170`	`170`	`pretrained_resource_files_map = {`
`171`	`171`	`"model_file": {`
`172`		`- "THUDM/chatglm-6b": "https://paddlenlp.bj.bcebos.com/models/community/THUDM/chatglm-6b/ice_text.model"`
	`172`	`+ "THUDM/chatglm-6b": "https://paddlenlp.bj.bcebos.com/models/community/THUDM/chatglm-6b/ice_text.model",`
	`173`	`+ "THUDM/chatglm-6b-v1.1": "https://paddlenlp.bj.bcebos.com/models/community/THUDM/chatglm-6b-v1.1/ice_text.model",`
`173`	`174`	`}`
`174`	`175`	`}`
`175`	`176`