!20050 modify for bert 310infer on r1.3

Merge pull request !20050 from 郑彬/bert0708_r1.3

model_zoo/official/nlp/bert/postprocess.py

@@ -21,11 +21,31 @@ import os
 import argparse
 import numpy as np
 from mindspore import Tensor
-from src.model_utils.config import bert_net_cfg
 from src.assessment_method import Accuracy, F1, MCC, Spearman_Correlation
-from run_ner import eval_result_print
+
+
+def eval_result_print(assessment_method_="accuracy", callback_=None):
+    """print eval result"""
+    if assessment_method_ == "accuracy":
+        print("acc_num {} , total_num {}, accuracy {:.6f}".format(callback_.acc_num, callback_.total_num,
+                                                                  callback_.acc_num / callback_.total_num))
+    elif assessment_method_ == "bf1":
+        print("Precision {:.6f} ".format(callback_.TP / (callback_.TP + callback_.FP)))
+        print("Recall {:.6f} ".format(callback_.TP / (callback_.TP + callback_.FN)))
+        print("F1 {:.6f} ".format(2 * callback_.TP / (2 * callback_.TP + callback_.FP + callback_.FN)))
+    elif assessment_method_ == "mf1":
+        print("F1 {:.6f} ".format(callback_.eval()[0]))
+    elif assessment_method_ == "mcc":
+        print("MCC {:.6f} ".format(callback_.cal()))
+    elif assessment_method_ == "spearman_correlation":
+        print("Spearman Correlation is {:.6f} ".format(callback_.cal()[0]))
+    else:
+        raise ValueError("Assessment method not supported, support: [accuracy, f1, mcc, spearman_correlation]")
+
 
 parser = argparse.ArgumentParser(description="postprocess")
+parser.add_argument("--seq_length", type=int, default=128, help="seq_length, default is 128. You can get this value "
+                                                                "through the relevant'*.yaml' filer")
 parser.add_argument("--batch_size", type=int, default=1, help="Eval batch size, default is 1")
 parser.add_argument("--label_dir", type=str, default="", help="label data dir")
 parser.add_argument("--assessment_method", type=str, default="BF1", choices=["BF1", "clue_benchmark", "MF1"],
@@ -58,21 +78,21 @@ if __name__ == "__main__":
     for f in file_name:
         if use_crf.lower() == "true":
             logits = ()
-            for j in range(bert_net_cfg.seq_length):
+            for j in range(args.seq_length):
                 f_name = f.split('.')[0] + '_' + str(j) + '.bin'
                 data_tmp = np.fromfile(os.path.join(args.result_dir, f_name), np.int32)
                 data_tmp = data_tmp.reshape(args.batch_size, num_class + 2)
                 logits += ((Tensor(data_tmp),),)
-            f_name = f.split('.')[0] + '_' + str(bert_net_cfg.seq_length) + '.bin'
+            f_name = f.split('.')[0] + '_' + str(args.seq_length) + '.bin'
             data_tmp = np.fromfile(os.path.join(args.result_dir, f_name), np.int32).tolist()
             data_tmp = Tensor(data_tmp)
             logits = (logits, data_tmp)
         else:
             f_name = os.path.join(args.result_dir, f.split('.')[0] + '_0.bin')
-            logits = np.fromfile(f_name, np.float32).reshape(bert_net_cfg.seq_length * args.batch_size, num_class)
+            logits = np.fromfile(f_name, np.float32).reshape(args.seq_length * args.batch_size, num_class)
             logits = Tensor(logits)
         label_ids = np.fromfile(os.path.join(args.label_dir, f), np.int32)
-        label_ids = Tensor(label_ids.reshape(args.batch_size, bert_net_cfg.seq_length))
+        label_ids = Tensor(label_ids.reshape(args.batch_size, args.seq_length))
         callback.update(logits, label_ids)
 
     print("==============================================================")

model_zoo/official/nlp/bert/pretrain_config.yaml

@@ -14,7 +14,7 @@ enable_profiling: False
 # ==============================================================================
 description: 'run_pretrain'
 distribute: 'false'
-epoch_size: 1
+epoch_size: 40
 device_id: 0
 device_num: 1
 enable_save_ckpt: 'true'

model_zoo/official/nlp/bert/run_pretrain.py

@@ -196,7 +196,7 @@ def run_pretrain():
             cfg.save_checkpoint_steps *= cfg.accumulation_steps
             logger.info("save checkpoint steps: {}".format(cfg.save_checkpoint_steps))
 
-    ds = create_bert_dataset(device_num, rank, cfg.do_shuffle, cfg.data_dir, cfg.schema_dir)
+    ds = create_bert_dataset(device_num, rank, cfg.do_shuffle, cfg.data_dir, cfg.schema_dir, cfg.batch_size)
     net_with_loss = BertNetworkWithLoss(bert_net_cfg, True)
 
     new_repeat_count = cfg.epoch_size * ds.get_dataset_size() // cfg.data_sink_steps

model_zoo/official/nlp/bert/src/dataset.py

@@ -20,10 +20,9 @@ import mindspore.common.dtype as mstype
 import mindspore.dataset as ds
 import mindspore.dataset.transforms.c_transforms as C
 from mindspore import log as logger
-from .model_utils.config import config as cfg
 
 
-def create_bert_dataset(device_num=1, rank=0, do_shuffle="true", data_dir=None, schema_dir=None):
+def create_bert_dataset(device_num=1, rank=0, do_shuffle="true", data_dir=None, schema_dir=None, batch_size=32):
     """create train dataset"""
     # apply repeat operations
     files = os.listdir(data_dir)
@@ -46,7 +45,7 @@ def create_bert_dataset(device_num=1, rank=0, do_shuffle="true", data_dir=None,
     data_set = data_set.map(operations=type_cast_op, input_columns="input_mask")
     data_set = data_set.map(operations=type_cast_op, input_columns="input_ids")
     # apply batch operations
-    data_set = data_set.batch(cfg.batch_size, drop_remainder=True)
+    data_set = data_set.batch(batch_size, drop_remainder=True)
     logger.info("data size: {}".format(data_set.get_dataset_size()))
     logger.info("repeat count: {}".format(data_set.get_repeat_count()))
     return data_set