Skip to content

Enable _MklFusedBatchMatMulV2 BF16 conversion #147

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
chensuyue merged 1 commit into from
Nov 25, 2022
Merged

Enable _MklFusedBatchMatMulV2 BF16 conversion #147

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions neural_compressor/adaptor/tensorflow.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -648,6 +648,8 @@ def _dump_model_op_stats(self, model_graphdef):
origin_op_type = 'DepthwiseConv2dNative'
if origin_op_type == 'BatchMatMul':
origin_op_type = 'BatchMatMulV2'
if origin_op_type == 'FusedBatchMatMulV2':
origin_op_type = '_MklFusedBatchMatMulV2'
if origin_op_type == 'Deconv2D':
origin_op_type = 'Conv2DBackpropInput'
if origin_op_type == 'Deconv3D':
Expand Down
2 changes: 1 addition & 1 deletion neural_compressor/adaptor/tensorflow.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,7 +30,7 @@
'MaxPool', 'MaxPool3D', 'AvgPool', 'Conv2DBackpropInput', 'Conv3DBackpropInputV2']
bf16: ["_MklLayerNorm", "Conv2D", "Conv2DBackpropFilter", "Conv2DBackpropInput", "Conv3D", "Conv3DBackpropFilterV2", "Conv3DBackpropInputV2",
"DepthwiseConv2dNative", "DepthwiseConv2dNativeBackpropFilter", "DepthwiseConv2dNativeBackpropInput", "GRUBlockCell",
"AUGRUBlockCell", "MklGRU", "MklAUGRU", "MatMul", "BatchMatMul", "BatchMatMulV2", "Einsum", # allow_list
"AUGRUBlockCell", "MklGRU", "MklAUGRU", "MatMul", "BatchMatMul", "BatchMatMulV2", "_MklFusedBatchMatMulV2", "Einsum", # allow_list
"Add", "AddN", "AddV2", "AvgPool", "AvgPool3D", "AvgPool3DGrad", "AvgPoolGrad", "BiasAdd", "BiasAddGrad", "BiasAddV1",
"Erf", "FusedBatchNormV2", "FusedBatchNormGradV2", "FusedBatchNormV3", "FusedBatchNormGradV3", "LeakyRelu", "LeakyReluGrad",
"Mean", "Mul", "Sub", "Elu", "EluGrad", "FloorDiv", "_FusedBatchNormEx", "Log", "Log1p", "LogSoftmax", "Prod", "RealDiv",
Expand Down
1 change: 1 addition & 0 deletions neural_compressor/adaptor/tf_utils/graph_converter.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -517,6 +517,7 @@ def bf16_convert(self):
FP32 + INT8 mixed precision graph.
"""
try:
logger.info("Start BF16 conversion.")
self._tmp_model.graph_def = BF16Convert(
self._tmp_model.graph_def,
self.fp32_ops,
Expand Down
5 changes: 0 additions & 5 deletions neural_compressor/adaptor/tf_utils/graph_rewriter/bf16/bf16_convert.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -243,11 +243,6 @@ def _model_bf16_convert(self):
for bf16_node_name in set(self.bf16_ops):
if bf16_node_name not in self.cur_graph.node_name_details:
self.bf16_ops.remove(bf16_node_name)
continue
else:
if "fused_ops" in self.cur_graph.node_name_details[bf16_node_name].node.attr:
self.bf16_ops.remove(bf16_node_name)
continue
for bf16_node_name in sorted(list(set(self.bf16_ops))):
self._bf16_convert(bf16_node_name)
return self.cur_graph.dump_graph()
Expand Down