[Bert][Gpu]Sync modify of bert script from r1.1 to master

model_zoo/official/nlp/bert/run_pretrain.py

@@ -44,11 +44,18 @@ _current_dir = os.path.dirname(os.path.realpath(__file__))
 
 def _set_bert_all_reduce_split():
     """set bert all_reduce fusion split, support num_hidden_layers is 12 and 24."""
+    device_target = context.get_context('device_target')
+    enable_graph_kernel = context.get_context('enable_graph_kernel')
+    device_num = context.get_auto_parallel_context('device_num')
     if bert_net_cfg.num_hidden_layers == 12:
         if bert_net_cfg.use_relative_positions:
             context.set_auto_parallel_context(all_reduce_fusion_config=[29, 58, 87, 116, 145, 174, 203, 217])
         else:
             context.set_auto_parallel_context(all_reduce_fusion_config=[28, 55, 82, 109, 136, 163, 190, 205])
+            if device_target == 'GPU' and enable_graph_kernel and device_num == 8:
+                context.set_auto_parallel_context(all_reduce_fusion_config=[180, 205])
+            elif device_target == 'GPU' and enable_graph_kernel and device_num == 16:
+                context.set_auto_parallel_context(all_reduce_fusion_config=[120, 205])
     elif bert_net_cfg.num_hidden_layers == 24:
         if bert_net_cfg.use_relative_positions:
             context.set_auto_parallel_context(all_reduce_fusion_config=[30, 90, 150, 210, 270, 330, 390, 421])
@@ -99,8 +106,7 @@ def _get_optimizer(args_opt, network):
 def _auto_enable_graph_kernel(device_target, graph_kernel_mode):
     """Judge whether is suitable to enable graph kernel."""
     return graph_kernel_mode in ("auto", "true") and device_target == 'GPU' and \
-        cfg.bert_network == 'base' and (cfg.batch_size == 32 or cfg.batch_size == 64) and \
-        cfg.optimizer == 'AdamWeightDecay'
+        cfg.bert_network == 'base' and cfg.optimizer == 'AdamWeightDecay'
 
 
 def _set_graph_kernel_context(device_target, enable_graph_kernel, is_auto_enable_graph_kernel):
@@ -111,10 +117,15 @@ def _set_graph_kernel_context(device_target, enable_graph_kernel, is_auto_enable
             logger.warning('Graph kernel only supports GPU back-end now, run with graph kernel off.')
 
 
-def _check_compute_type(device_target, is_auto_enable_graph_kernel):
-    if device_target == 'GPU' and bert_net_cfg.compute_type != mstype.float32 and not is_auto_enable_graph_kernel:
-        logger.warning('Gpu only support fp32 temporarily, run with fp32.')
+def _check_compute_type(args_opt, is_auto_enable_graph_kernel):
+    if args_opt.device_target == 'GPU' and bert_net_cfg.compute_type != mstype.float32 and \
+       not is_auto_enable_graph_kernel:
+        warning_message = 'Gpu only support fp32 temporarily, run with fp32.'
         bert_net_cfg.compute_type = mstype.float32
+        if args_opt.enable_lossscale == "true":
+            args_opt.enable_lossscale = "false"
+            warning_message = 'Gpu only support fp32 temporarily, run with fp32 and disable lossscale.'
+        logger.warning(warning_message)
 
 
 def argparse_init():
@@ -160,6 +171,8 @@ def run_pretrain():
     args_opt = parser.parse_args()
     context.set_context(mode=context.GRAPH_MODE, device_target=args_opt.device_target, device_id=args_opt.device_id)
     context.set_context(reserve_class_name_in_scope=False)
+    is_auto_enable_graph_kernel = _auto_enable_graph_kernel(args_opt.device_target, args_opt.enable_graph_kernel)
+    _set_graph_kernel_context(args_opt.device_target, args_opt.enable_graph_kernel, is_auto_enable_graph_kernel)
     ckpt_save_dir = args_opt.save_checkpoint_path
     if args_opt.distribute == "true":
         if args_opt.device_target == 'Ascend':
@@ -175,15 +188,12 @@ def run_pretrain():
         context.reset_auto_parallel_context()
         context.set_auto_parallel_context(parallel_mode=ParallelMode.DATA_PARALLEL, gradients_mean=True,
                                           device_num=device_num)
-        if args_opt.device_target == 'Ascend':
-            _set_bert_all_reduce_split()
+        _set_bert_all_reduce_split()
     else:
         rank = 0
         device_num = 1
 
-    is_auto_enable_graph_kernel = _auto_enable_graph_kernel(args_opt.device_target, args_opt.enable_graph_kernel)
-    _set_graph_kernel_context(args_opt.device_target, args_opt.enable_graph_kernel, is_auto_enable_graph_kernel)
-    _check_compute_type(args_opt.device_target, is_auto_enable_graph_kernel)
+    _check_compute_type(args_opt, is_auto_enable_graph_kernel)
 
     if args_opt.accumulation_steps > 1:
         logger.info("accumulation steps: {}".format(args_opt.accumulation_steps))

model_zoo/official/nlp/bert/scripts/run_distributed_pretrain_for_gpu.sh

@@ -32,7 +32,7 @@ mpirun --allow-run-as-root -n $RANK_SIZE --output-filename log_output --merge-st
     --distribute="true"        \
     --epoch_size=$EPOCH_SIZE    \
     --enable_save_ckpt="true"    \
-    --enable_lossscale="false"    \
+    --enable_lossscale="true"    \
     --do_shuffle="true"        \
     --enable_data_sink="true"    \
     --data_sink_steps=20        \

model_zoo/official/nlp/bert/scripts/run_standalone_pretrain_for_gpu.sh

@@ -36,7 +36,7 @@ python run_pretrain.py  \
     --distribute="false" \
     --epoch_size=$EPOCH_SIZE \
     --enable_save_ckpt="true" \
-    --enable_lossscale="false" \
+    --enable_lossscale="true" \
     --do_shuffle="true" \
     --enable_data_sink="true" \
     --data_sink_steps=20 \