Merge pull request #1111 from TeslaZhao/develop

Update paddle predictor API verson to 2.0

python/examples/pipeline/ocr/web_service.py

@@ -48,7 +48,7 @@ class DetOp(Op):
         imgs = []
         for key in input_dict.keys():
             data = base64.b64decode(input_dict[key].encode('utf8'))
-            data = np.fromstring(data, np.uint8)
+            data = np.frombuffer(data, np.uint8)
             self.im = cv2.imdecode(data, cv2.IMREAD_COLOR)
             self.ori_h, self.ori_w, _ = self.im.shape
             det_img = self.det_preprocess(self.im)
@@ -57,7 +57,7 @@ class DetOp(Op):
         return {"image": np.concatenate(imgs, axis=0)}, False, None, ""
 
     def postprocess(self, input_dicts, fetch_dict, log_id):
-#        print(fetch_dict)
+        #        print(fetch_dict)
         det_out = fetch_dict["concat_1.tmp_0"]
         ratio_list = [
             float(self.new_h) / self.ori_h, float(self.new_w) / self.ori_w
@@ -114,5 +114,5 @@ class OcrService(WebService):
 
 
 uci_service = OcrService(name="ocr")
-uci_service.prepare_pipeline_config("config2.yml")
+uci_service.prepare_pipeline_config("config.yml")
 uci_service.run_service()

python/examples/pipeline/simple_web_service/web_service.py

@@ -14,7 +14,7 @@
 try:
     from paddle_serving_server.web_service import WebService, Op
 except ImportError:
-    from paddle_serving_server.web_service import WebService, Op
+    from paddle_serving_server_gpu.web_service import WebService, Op
 import logging
 import numpy as np
 import sys
@@ -34,8 +34,11 @@ class UciOp(Op):
         x_value = input_dict["x"].split(self.batch_separator)
         x_lst = []
         for x_val in x_value:
-            x_lst.append(np.array([float(x.strip()) for x in x_val.split(self.separator)]).reshape(1, 13))
-        input_dict["x"] = np.concatenate(x_lst, axis=0) 
+            x_lst.append(
+                np.array([
+                    float(x.strip()) for x in x_val.split(self.separator)
+                ]).reshape(1, 13))
+        input_dict["x"] = np.concatenate(x_lst, axis=0)
         proc_dict = {}
         return input_dict, False, None, ""
 
@@ -53,5 +56,5 @@ class UciService(WebService):
 
 
 uci_service = UciService(name="uci")
-uci_service.prepare_pipeline_config("config2.yml")
+uci_service.prepare_pipeline_config("config.yml")
 uci_service.run_service()

python/paddle_serving_app/local_predict.py

@@ -19,16 +19,12 @@ import os
 import google.protobuf.text_format
 import numpy as np
 import argparse
-import paddle.fluid as fluid
-import paddle.inference as inference
 from .proto import general_model_config_pb2 as m_config
-from paddle.fluid.core import PaddleTensor
-from paddle.fluid.core import AnalysisConfig
-from paddle.fluid.core import create_paddle_predictor
+import paddle.inference as paddle_infer
 import logging
 
 logging.basicConfig(format="%(asctime)s - %(levelname)s - %(message)s")
-logger = logging.getLogger("fluid")
+logger = logging.getLogger("LocalPredictor")
 logger.setLevel(logging.INFO)
 
 
@@ -62,7 +58,7 @@ class LocalPredictor(object):
                           use_xpu=False,
                           use_feed_fetch_ops=False):
         """
-        Load model config and set the engine config for the paddle predictor
+        Load model configs and create the paddle predictor by Paddle Inference API.
    
         Args:
             model_path: model config path.
@@ -83,14 +79,18 @@ class LocalPredictor(object):
         model_conf = google.protobuf.text_format.Merge(
             str(f.read()), model_conf)
         if os.path.exists(os.path.join(model_path, "__params__")):
-            config = AnalysisConfig(os.path.join(model_path, "__model__"), os.path.join(model_path, "__params__")) 
+            config = paddle_infer.Config(
+                os.path.join(model_path, "__model__"),
+                os.path.join(model_path, "__params__"))
         else:
-            config = AnalysisConfig(model_path) 
-        logger.info("load_model_config params: model_path:{}, use_gpu:{},\
+            config = paddle_infer.Config(model_path)
+
+        logger.info(
+            "LocalPredictor load_model_config params: model_path:{}, use_gpu:{},\
             gpu_id:{}, use_profile:{}, thread_num:{}, mem_optim:{}, ir_optim:{},\
             use_trt:{}, use_lite:{}, use_xpu: {}, use_feed_fetch_ops:{}".format(
-            model_path, use_gpu, gpu_id, use_profile, thread_num, mem_optim,
-            ir_optim, use_trt, use_lite, use_xpu, use_feed_fetch_ops))
+                model_path, use_gpu, gpu_id, use_profile, thread_num, mem_optim,
+                ir_optim, use_trt, use_lite, use_xpu, use_feed_fetch_ops))
 
         self.feed_names_ = [var.alias_name for var in model_conf.feed_var]
         self.fetch_names_ = [var.alias_name for var in model_conf.fetch_var]
@@ -129,7 +129,7 @@ class LocalPredictor(object):
 
         if use_lite:
             config.enable_lite_engine(
-                precision_mode=inference.PrecisionType.Float32,
+                precision_mode=paddle_infer.PrecisionType.Float32,
                 zero_copy=True,
                 passes_filter=[],
                 ops_filter=[])
@@ -138,11 +138,11 @@ class LocalPredictor(object):
             # 2MB l3 cache
             config.enable_xpu(8 * 1024 * 1024)
 
-        self.predictor = create_paddle_predictor(config)
+        self.predictor = paddle_infer.create_predictor(config)
 
     def predict(self, feed=None, fetch=None, batch=False, log_id=0):
         """
-        Predict locally
+        Run model inference by Paddle Inference API.
 
         Args:
             feed: feed var
@@ -155,14 +155,16 @@ class LocalPredictor(object):
             fetch_map: dict 
         """
         if feed is None or fetch is None:
-            raise ValueError("You should specify feed and fetch for prediction")
+            raise ValueError("You should specify feed and fetch for prediction.\
+                log_id:{}".format(log_id))
         fetch_list = []
         if isinstance(fetch, str):
             fetch_list = [fetch]
         elif isinstance(fetch, list):
             fetch_list = fetch
         else:
-            raise ValueError("Fetch only accepts string and list of string")
+            raise ValueError("Fetch only accepts string and list of string.\
+                log_id:{}".format(log_id))
 
         feed_batch = []
         if isinstance(feed, dict):
@@ -170,27 +172,21 @@ class LocalPredictor(object):
         elif isinstance(feed, list):
             feed_batch = feed
         else:
-            raise ValueError("Feed only accepts dict and list of dict")
-
-        int_slot_batch = []
-        float_slot_batch = []
-        int_feed_names = []
-        float_feed_names = []
-        int_shape = []
-        float_shape = []
-        fetch_names = []
-        counter = 0
-        batch_size = len(feed_batch)
+            raise ValueError("Feed only accepts dict and list of dict.\
+                log_id:{}".format(log_id))
 
+        fetch_names = []
+        # Filter invalid fetch names
         for key in fetch_list:
             if key in self.fetch_names_:
                 fetch_names.append(key)
 
         if len(fetch_names) == 0:
             raise ValueError(
-                "Fetch names should not be empty or out of saved fetch list.")
-            return {}
+                "Fetch names should not be empty or out of saved fetch list.\
+                    log_id:{}".format(log_id))
 
+        # Assemble the input data of paddle predictor 
         input_names = self.predictor.get_input_names()
         for name in input_names:
             if isinstance(feed[name], list):
@@ -204,27 +200,31 @@ class LocalPredictor(object):
                 feed[name] = feed[name].astype("int32")
             else:
                 raise ValueError("local predictor receives wrong data type")
-            input_tensor = self.predictor.get_input_tensor(name)
+            input_tensor_handle = self.predictor.get_input_handle(name)
             if "{}.lod".format(name) in feed:
-                input_tensor.set_lod([feed["{}.lod".format(name)]])
+                input_tensor_handle.set_lod([feed["{}.lod".format(name)]])
             if batch == False:
-                input_tensor.copy_from_cpu(feed[name][np.newaxis, :])
+                input_tensor_handle.copy_from_cpu(feed[name][np.newaxis, :])
             else:
-                input_tensor.copy_from_cpu(feed[name])
-        output_tensors = []
+                input_tensor_handle.copy_from_cpu(feed[name])
+        output_tensor_handles = []
         output_names = self.predictor.get_output_names()
         for output_name in output_names:
-            output_tensor = self.predictor.get_output_tensor(output_name)
-            output_tensors.append(output_tensor)
+            output_tensor_handle = self.predictor.get_output_handle(output_name)
+            output_tensor_handles.append(output_tensor_handle)
+
+        # Run inference 
+        self.predictor.run()
+
+        # Assemble output data of predict results
         outputs = []
-        self.predictor.zero_copy_run()
-        for output_tensor in output_tensors:
-            output = output_tensor.copy_to_cpu()
+        for output_tensor_handle in output_tensor_handles:
+            output = output_tensor_handle.copy_to_cpu()
             outputs.append(output)
         fetch_map = {}
         for i, name in enumerate(fetch):
             fetch_map[name] = outputs[i]
-            if len(output_tensors[i].lod()) > 0:
-                fetch_map[name + ".lod"] = np.array(output_tensors[i].lod()[
-                    0]).astype('int32')
+            if len(output_tensor_handles[i].lod()) > 0:
+                fetch_map[name + ".lod"] = np.array(output_tensor_handles[i]
+                                                    .lod()[0]).astype('int32')
         return fetch_map