plt add nlp case, test=model (#2998)

* plt update, test=model

* plt update, test=model

* plt update, test=model

* plt update testing reporter, test=model

* plt update testing reporter, test=model

* plt update testing reporter, test=model

* plt update test reporter, test=model

* plt update test reporter, test=model

* plt update test reporter, test=model

* plt update test reporter, test=model

* plt update test reporter, test=model

* plt update test reporter, test=model

* plt add pr_info

* plt fix ocr case, test=model

* plt fix ocr case, test=model

* plt add nlp case, test=model

* plt add nlp case, test=model

* plt add nlp case, test=model

* plt add nlp case, test=model

* plt add nlp case, test=model

* plt add more nlp case, test=model

* plt add more nlp case, test=model

* plt add nlp, test=model

* plt fix case, test=model

Author: Zeref996

framework/e2e/PaddleLT_new/engine/paddle_train.py

@@ -119,7 +119,8 @@ def _get_data_grad(self, data):
         """记录list[inputs...]中的input.grad并生成list[input.grad...]"""
         data_grad = []
         for i in data:
-            data_grad.append(i.grad)
+            if isinstance(i, paddle.Tensor):
+                data_grad.append(i.grad)
         return data_grad
 
     def dy_train(self):

framework/e2e/PaddleLT_new/generator/builder_data.py

@@ -16,8 +16,11 @@
     import layercase
 
     if os.environ.get("USE_PADDLE_MODEL", "None") == "PaddleOCR":
-        import layerModelcase
+        import layerOCRcase
         import PaddleOCR
+    elif os.environ.get("USE_PADDLE_MODEL", "None") == "PaddleNLP":
+        import layerNLPcase
+        import paddlenlp
 elif os.environ.get("FRAMEWORK") == "torch":
     import torch
     import layerTorchcase
@@ -40,16 +43,33 @@ def get_single_data(self):
             data = []
             # for i in eval(dataname):
             for i in getattr(self.layer_module, "create_numpy_inputs")():
-                if os.environ.get("FRAMEWORK") == "paddle":
-                    if i.dtype == np.int64 or i.dtype == np.int32:
-                        data.append(paddle.to_tensor(i, stop_gradient=True))
-                    else:
-                        data.append(paddle.to_tensor(i, stop_gradient=False))
-                elif os.environ.get("FRAMEWORK") == "torch":
-                    if i.dtype == np.int64 or i.dtype == np.int32:
-                        data.append(torch.tensor(i, requires_grad=False))
-                    else:
-                        data.append(torch.tensor(i, requires_grad=True))
+                if isinstance(i, (tuple, list)):  # 为了适配list输入的模型子图
+                    tmp = []
+                    for j in i:
+                        if os.environ.get("FRAMEWORK") == "paddle":
+                            if j.dtype == np.int64 or j.dtype == np.int32:
+                                tmp.append(paddle.to_tensor(j, stop_gradient=True))
+                            else:
+                                tmp.append(paddle.to_tensor(j, stop_gradient=False))
+                        elif os.environ.get("FRAMEWORK") == "torch":
+                            if j.dtype == np.int64 or j.dtype == np.int32:
+                                tmp.append(torch.tensor(j, requires_grad=False))
+                            else:
+                                tmp.append(torch.tensor(j, requires_grad=True))
+                    data.append(tmp)
+                elif isinstance(i, np.ndarray):
+                    if os.environ.get("FRAMEWORK") == "paddle":
+                        if i.dtype == np.int64 or i.dtype == np.int32:
+                            data.append(paddle.to_tensor(i, stop_gradient=True))
+                        else:
+                            data.append(paddle.to_tensor(i, stop_gradient=False))
+                    elif os.environ.get("FRAMEWORK") == "torch":
+                        if i.dtype == np.int64 or i.dtype == np.int32:
+                            data.append(torch.tensor(i, requires_grad=False))
+                        else:
+                            data.append(torch.tensor(i, requires_grad=True))
+                else:
+                    data.append(i)
         else:
             data = self.get_single_tensor()
 

framework/e2e/PaddleLT_new/generator/builder_layer.py

@@ -14,8 +14,11 @@
     import layercase
 
     if os.environ.get("USE_PADDLE_MODEL", "None") == "PaddleOCR":
-        import layerModelcase
+        import layerOCRcase
         import PaddleOCR
+    elif os.environ.get("USE_PADDLE_MODEL", "None") == "PaddleNLP":
+        import layerNLPcase
+        import paddlenlp
 elif os.environ.get("FRAMEWORK") == "torch":
     import torch
     import layerTorchcase

…/PaddleLT_new/layerModelcase/__init__.py …2e/PaddleLT_new/layerNLPcase/__init__.pyframework/e2e/PaddleLT_new/layerModelcase/__init__.py renamed to framework/e2e/PaddleLT_new/layerNLPcase/__init__.py framework/e2e/PaddleLT_new/layerModelcase/__init__.py renamed to framework/e2e/PaddleLT_new/layerNLPcase/__init__.py

@@ -15,4 +15,4 @@
     # 获取文件名（不含扩展名）
     module_name = os.path.basename(folder)
     # 导入模块
-    __import__('layerModelcase.' + module_name, globals(), locals(), [])
+    __import__('layerNLPcase.' + module_name, globals(), locals(), [])

framework/e2e/PaddleLT_new/layerNLPcase/debug/case_bug/transformers/layoutlmv2/__init__.py

@@ -0,0 +1,15 @@
+import os
+import glob
+
+# 获取当前文件所在目录
+current_dir = os.path.dirname(__file__)
+
+# 获取当前目录下所有的 .py 文件路径
+py_files = glob.glob(os.path.join(current_dir, "*.py"))
+
+# 动态导入所有 .py 文件
+for py_file in py_files:
+    # 获取文件名（不含扩展名）
+    module_name = os.path.basename(py_file)[:-3]
+    # 导入模块
+    __import__("layerNLPcase.transformers.layoutlmv2." + module_name, globals(), locals(), [])

framework/e2e/PaddleLT_new/layerNLPcase/debug/case_bug/transformers/layoutlmv2/layoutlmv2_model.py

@@ -0,0 +1,41 @@
+import paddle
+import numpy as np
+from paddlenlp.transformers import LayoutLMv2Model, LayoutLMv2Tokenizer
+
+def LayerCase():
+    """模型库中间态"""
+    model = LayoutLMv2Model.from_pretrained('layoutlmv2-base-uncased')
+    return model
+
+def create_inputspec():
+    inputspec = (
+        paddle.static.InputSpec(shape=(-1, 13), dtype=paddle.float32, stop_gradient=False),
+        None,
+        None,
+        paddle.static.InputSpec(shape=(-1, 13), dtype=paddle.float32, stop_gradient=False),
+        )
+    return inputspec
+
+
+def create_tensor_inputs():
+    tokenizer = LayoutLMv2Tokenizer.from_pretrained('layoutlmv2-base-uncased')
+    inputs_dict = tokenizer("Welcome to use PaddlePaddle and PaddleNLP!")
+    inputs = (
+        paddle.to_tensor(inputs_dict['input_ids'], stop_gradient=False),
+        None,
+        None,
+        paddle.to_tensor(inputs_dict['token_type_ids'], stop_gradient=False),
+    )
+    return inputs
+
+
+def create_numpy_inputs():
+    tokenizer = LayoutLMv2Tokenizer.from_pretrained('layoutlmv2-base-uncased')
+    inputs_dict = tokenizer("Welcome to use PaddlePaddle and PaddleNLP!")
+    inputs = (
+        np.array([inputs_dict['input_ids']]),
+        None,
+        None,
+        np.array([inputs_dict['token_type_ids']]),
+    )
+    return inputs

framework/e2e/PaddleLT_new/layerNLPcase/debug/case_bug/transformers/layoutxlm/__init__.py

@@ -0,0 +1,15 @@
+import os
+import glob
+
+# 获取当前文件所在目录
+current_dir = os.path.dirname(__file__)
+
+# 获取当前目录下所有的 .py 文件路径
+py_files = glob.glob(os.path.join(current_dir, "*.py"))
+
+# 动态导入所有 .py 文件
+for py_file in py_files:
+    # 获取文件名（不含扩展名）
+    module_name = os.path.basename(py_file)[:-3]
+    # 导入模块
+    __import__("layerNLPcase.transformers.layoutxlm." + module_name, globals(), locals(), [])

framework/e2e/PaddleLT_new/layerNLPcase/debug/case_bug/transformers/layoutxlm/layoutxlm_model.py

@@ -0,0 +1,41 @@
+import paddle
+import numpy as np
+from paddlenlp.transformers import LayoutXLMModel, LayoutXLMTokenizer
+
+def LayerCase():
+    """模型库中间态"""
+    model = LayoutXLMModel.from_pretrained('layoutxlm-base-uncased')
+    return model
+
+def create_inputspec():
+    inputspec = (
+        paddle.static.InputSpec(shape=(-1, 13), dtype=paddle.float32, stop_gradient=False),
+        None,
+        None,
+        paddle.static.InputSpec(shape=(-1, 13), dtype=paddle.float32, stop_gradient=False),
+        )
+    return inputspec
+
+
+def create_tensor_inputs():
+    tokenizer = LayoutXLMTokenizer.from_pretrained('layoutxlm-base-uncased')
+    inputs_dict = tokenizer("Welcome to use PaddlePaddle and PaddleNLP!")
+    inputs = (
+        paddle.to_tensor(inputs_dict['input_ids'], stop_gradient=False),
+        None,
+        None,
+        paddle.to_tensor(inputs_dict['token_type_ids'], stop_gradient=False),
+    )
+    return inputs
+
+
+def create_numpy_inputs():
+    tokenizer = LayoutXLMTokenizer.from_pretrained('layoutxlm-base-uncased')
+    inputs_dict = tokenizer("Welcome to use PaddlePaddle and PaddleNLP!")
+    inputs = (
+        np.array([inputs_dict['input_ids']]),
+        None,
+        None,
+        np.array([inputs_dict['token_type_ids']]),
+    )
+    return inputs

…odelcase/Ocr_cases/backbones/__init__.py …se_bug/transformers/reformer/__init__.pyframework/e2e/PaddleLT_new/layerModelcase/Ocr_cases/backbones/__init__.py renamed to framework/e2e/PaddleLT_new/layerNLPcase/debug/case_bug/transformers/reformer/__init__.py framework/e2e/PaddleLT_new/layerModelcase/Ocr_cases/backbones/__init__.py renamed to framework/e2e/PaddleLT_new/layerNLPcase/debug/case_bug/transformers/reformer/__init__.py

@@ -12,4 +12,4 @@
     # 获取文件名（不含扩展名）
     module_name = os.path.basename(py_file)[:-3]
     # 导入模块
-    __import__("layerModelcase.Ocr_cases.backbones." + module_name, globals(), locals(), [])
+    __import__("layerNLPcase.transformers.reformer." + module_name, globals(), locals(), [])

framework/e2e/PaddleLT_new/layerNLPcase/debug/case_bug/transformers/reformer/reformer_model.py

@@ -0,0 +1,29 @@
+import paddle
+import numpy as np
+from paddlenlp.transformers import ReformerModel, ReformerTokenizer
+
+def LayerCase():
+    """模型库中间态"""
+    model = ReformerModel.from_pretrained('reformer-crime-and-punishment')
+    return model
+
+def create_inputspec():
+    inputspec = (
+        paddle.static.InputSpec(shape=(-1, 15), dtype=paddle.float32, stop_gradient=False),
+        paddle.static.InputSpec(shape=(-1, 15), dtype=paddle.float32, stop_gradient=False),
+        )
+    return inputspec
+
+
+def create_tensor_inputs():
+    tokenizer = ReformerTokenizer.from_pretrained('reformer-crime-and-punishment')
+    inputs_dict = tokenizer("Welcome to use PaddlePaddle and PaddleNLP!")
+    inputs = tuple(paddle.to_tensor([v], stop_gradient=False) for (k, v) in inputs_dict.items())
+    return inputs
+
+
+def create_numpy_inputs():
+    tokenizer = ReformerTokenizer.from_pretrained('reformer-crime-and-punishment')
+    inputs_dict = tokenizer("Welcome to use PaddlePaddle and PaddleNLP!")
+    inputs = tuple(np.array([v]) for (k, v) in inputs_dict.items())
+    return inputs