support simple serving for ernie-m text classification (#4435)

* support ernie m

* add simple serving for ernie m

* add handler in library

* change class name

Author: lugimzzz

applications/text_classification/hierarchical/deploy/simple_serving/README.md

@@ -16,7 +16,10 @@ pip install paddlenlp --upgrade
 ```bash
 paddlenlp server server:app --host 0.0.0.0 --port 8189
 ```
-
+如果是ERNIE-M模型则启动
+```bash
+paddlenlp server ernie_m_server:app --host 0.0.0.0 --port 8189
+```
 #### 分类任务发送服务
 ```bash
 python client.py

applications/text_classification/hierarchical/deploy/simple_serving/ernie_m_server.py

@@ -0,0 +1,25 @@
+# coding:utf-8
+# Copyright (c) 2022  PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from paddlenlp import SimpleServer
+from paddlenlp.server import ERNIEMHandler, MultiLabelClassificationPostHandler
+
+app = SimpleServer()
+app.register(
+    "models/cls_hierarchical",
+    model_path="../../export",
+    tokenizer_name="ernie-m-base",
+    model_handler=ERNIEMHandler,
+    post_handler=MultiLabelClassificationPostHandler,
+)

applications/text_classification/multi_class/deploy/simple_serving/README.md

@@ -16,7 +16,10 @@ pip install paddlenlp >= 2.4.4
 ```bash
 paddlenlp server server:app --host 0.0.0.0 --port 8189
 ```
-
+如果是ERNIE-M模型则启动
+```bash
+paddlenlp server ernie_m_server:app --host 0.0.0.0 --port 8189
+```
 #### 启动分类 Client 服务
 ```bash
 python client.py

applications/text_classification/multi_class/deploy/simple_serving/ernie_m_server.py

@@ -0,0 +1,26 @@
+# coding:utf-8
+# Copyright (c) 2022  PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddlenlp import SimpleServer
+from paddlenlp.server import ERNIEMHandler, MultiClassificationPostHandler
+
+app = SimpleServer()
+app.register(
+    "models/cls_multi_class",
+    model_path="../../export",
+    tokenizer_name="ernie-m-base",
+    model_handler=ERNIEMHandler,
+    post_handler=MultiClassificationPostHandler,
+)

applications/text_classification/multi_label/deploy/simple_serving/README.md

@@ -16,7 +16,10 @@ pip install paddlenlp --upgrade
 ```bash
 paddlenlp server server:app --host 0.0.0.0 --port 8189
 ```
-
+如果是ERNIE-M模型则启动
+```bash
+paddlenlp server ernie_m_server:app --host 0.0.0.0 --port 8189
+```
 #### 分类任务发送服务
 ```bash
 python client.py

applications/text_classification/multi_label/deploy/simple_serving/ernie_m_server.py

@@ -0,0 +1,26 @@
+# coding:utf-8
+# Copyright (c) 2022  PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddlenlp import SimpleServer
+from paddlenlp.server import ERNIEMHandler, MultiLabelClassificationPostHandler
+
+app = SimpleServer()
+app.register(
+    "models/cls_multi_label",
+    model_path="../../export",
+    tokenizer_name="ernie-m-base",
+    model_handler=ERNIEMHandler,
+    post_handler=MultiLabelClassificationPostHandler,
+)

paddlenlp/server/handlers/__init__.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+# Copyright (c) 2023 PaddlePaddle Authors. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -17,7 +17,7 @@
     MultiClassificationPostHandler,
     MultiLabelClassificationPostHandler,
 )
-from .custom_model_handler import CustomModelHandler
+from .custom_model_handler import CustomModelHandler, ERNIEMHandler
 from .qa_model_handler import QAModelHandler
 from .taskflow_handler import TaskflowHandler
 from .token_model_handler import TokenClsModelHandler

paddlenlp/server/handlers/custom_model_handler.py

@@ -1,5 +1,5 @@
 # coding:utf-8
-# Copyright (c) 2022  PaddlePaddle Authors. All Rights Reserved.
+# Copyright (c) 2023  PaddlePaddle Authors. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License"
 # you may not use this file except in compliance with the License.
@@ -13,10 +13,9 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import numpy as np
+
+from ...data import Pad, Tuple
 from .base_handler import BaseModelHandler
-from ...transformers import AutoTokenizer
-from ...data import Tuple, Pad
-from ...utils.log import logger
 
 
 class CustomModelHandler(BaseModelHandler):
@@ -59,10 +58,11 @@ def process(cls, predictor, tokenizer, data, parameters):
         # Seperates data into some batches.
         batches = [examples[i : i + batch_size] for i in range(0, len(examples), batch_size)]
 
-        batchify_fn = lambda samples, fn=Tuple(
-            Pad(axis=0, pad_val=tokenizer.pad_token_id, dtype="int64"),  # input
-            Pad(axis=0, pad_val=tokenizer.pad_token_type_id, dtype="int64"),  # segment
-        ): fn(samples)
+        def batchify_fn(samples):
+            return Tuple(
+                Pad(axis=0, pad_val=tokenizer.pad_token_id, dtype="int64"),
+                Pad(axis=0, pad_val=tokenizer.pad_token_type_id, dtype="int64"),
+            )(samples)
 
         results = [[]] * predictor._output_num
         for batch in batches:
@@ -75,7 +75,74 @@ def process(cls, predictor, tokenizer, data, parameters):
                 for i, out in enumerate(output):
                     results[i].append(out)
             else:
-                outputs = predictor._predictor.run(None, {"input_ids": input_ids, "token_type_ids": token_type_ids})
+                predictor._predictor.run(None, {"input_ids": input_ids, "token_type_ids": token_type_ids})
+                for i, out in enumerate(output):
+                    results[i].append(out)
+
+        # Resolve the logits result and get the predict label and confidence
+        results_concat = []
+        for i in range(0, len(results)):
+            results_concat.append(np.concatenate(results[i], axis=0))
+        out_dict = {"logits": results_concat[0].tolist(), "data": data}
+        for i in range(1, len(results_concat)):
+            out_dict[f"logits_{i}"] = results_concat[i].tolist()
+        return out_dict
+
+
+class ERNIEMHandler(BaseModelHandler):
+    def __init__(self):
+        super().__init__()
+
+    @classmethod
+    def process(cls, predictor, tokenizer, data, parameters):
+        max_seq_len = 128
+        batch_size = 1
+        if "max_seq_len" not in parameters:
+            max_seq_len = parameters["max_seq_len"]
+        if "batch_size" not in parameters:
+            batch_size = parameters["batch_size"]
+        text = None
+        if "text" in data:
+            text = data["text"]
+        if text is None:
+            return {}
+        if isinstance(text, str):
+            text = [text]
+        has_pair = False
+        if "text_pair" in data and data["text_pair"] is not None:
+            text_pair = data["text_pair"]
+            if isinstance(text_pair, str):
+                text_pair = [text_pair]
+            if len(text) != len(text_pair):
+                raise ValueError("The length of text and text_pair must be same.")
+            has_pair = True
+
+        # Get the result of tokenizer
+        examples = []
+        for idx, data in enumerate(text):
+            if has_pair:
+                result = tokenizer(text=text[idx], text_pair=text_pair[idx], max_length=max_seq_len)
+            else:
+                result = tokenizer(text=text[idx], max_length=max_seq_len)
+            examples.append(result["input_ids"])
+
+        # Seperates data into some batches.
+        batches = [examples[i : i + batch_size] for i in range(0, len(examples), batch_size)]
+
+        def batchify_fn(samples):
+            return Pad(axis=0, pad_val=tokenizer.pad_token_id, dtype="int64")(samples)
+
+        results = [[]] * predictor._output_num
+        for batch in batches:
+            input_ids = batchify_fn(batch)
+            if predictor._predictor_type == "paddle_inference":
+                predictor._input_handles[0].copy_from_cpu(input_ids)
+                predictor._predictor.run()
+                output = [output_handle.copy_to_cpu() for output_handle in predictor._output_handles]
+                for i, out in enumerate(output):
+                    results[i].append(out)
+            else:
+                predictor._predictor.run(None, {"input_ids": input_ids})
                 for i, out in enumerate(output):
                     results[i].append(out)