Skip to content
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

Support some ops for full quantization #40083

Merged
merged 5 commits into from
Mar 15, 2022
Merged

Support some ops for full quantization #40083

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
423912c
add some op for full_quantization
yghstill Mar 2, 2022
1cc33f7
update ops list
yghstill Mar 14, 2022
7bf00d7
Merge branch 'develop' of github.com:PaddlePaddle/Paddle into add_ful…
yghstill Mar 14, 2022
7e1c8ca
fix fill_constant
yghstill Mar 15, 2022
7c1b447
fix sum op
yghstill Mar 15, 2022
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: 0 additions & 2 deletions python/paddle/fluid/contrib/slim/quantization/post_training_quantization.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -936,8 +936,6 @@ def analysis_and_save_info(op_node, out_var_name):
if op.type in (
self._quantizable_op_type + self._out_scale_op_list):
out_var_names = _get_op_output_var_names(op)
assert len(out_var_names) == 1, "Post training " + \
"quantization only support one output for " + op.type
for var_name in out_var_names:
analysis_and_save_info(op, var_name)

Expand Down
51 changes: 14 additions & 37 deletions python/paddle/fluid/contrib/slim/quantization/quantization_pass.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,42 +47,14 @@
]

_out_scale_op_list = [
"conv2d",
"depthwise_conv2d",
"mul",
"matmul",
"matmul_v2",
"relu",
"leaky_relu",
"relu6",
"sigmoid",
"tanh",
"prelu",
"swish",
"softmax",
"batch_norm",
"layer_norm",
"elementwise_add",
"pool2d",
"reshape2",
"transpose2",
"concat",
"elementwise_mul",
"scale",
"slice",
"hard_swish",
"hard_sigmoid",
"conv2d_transpose",
"gru",
"bilinear_interp",
"nearest_interp",
"trilinear_interp",
"flatten",
"flatten2",
"transpose",
"pad2d",
"reshape",
"layer_norm",
"conv2d", "depthwise_conv2d", "mul", "matmul", "matmul_v2", "relu",
"leaky_relu", "relu6", "sigmoid", "tanh", "prelu", "swish", "softmax",
"batch_norm", "layer_norm", "elementwise_add", "pool2d", "reshape2",
"transpose2", "concat", "elementwise_mul", "scale", "slice", "hard_swish",
"hard_sigmoid", "conv2d_transpose", "gru", "bilinear_interp",
"nearest_interp", "trilinear_interp", "flatten", "flatten2", "transpose",
"pad2d", "reshape", "layer_norm", "split", "flatten_contiguous_range",
"squeeze2", "nearest_interp_v2", "fill_constant_batch_size_like"
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

list后边加个逗号，修改文件后的diff会更直观。

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done

]

# list op real input and output names, to avoid processing input such as AxisTensor.
Expand Down Expand Up @@ -137,6 +109,10 @@
"flatten2": [["X"], ["Out"]],
"unsqueeze2": [["X"], ["Out"]],
"flatten_contiguous_range": [['X'], ["Out"]],
"split": [['X'], ["Out"]],
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

split需要特殊处理下么？类似于concat。

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

out_threshold只输出一个即可，其余暂不需要特殊处理。

"squeeze2": [['X'], ["Out"]],
"nearest_interp_v2": [['X'], ["Out"]],
"fill_constant_batch_size_like": [['Input'], ["Out"]],
}

_conv_ops = ['conv2d', 'depthwise_conv2d', 'conv2d_transpose']
Expand Down Expand Up @@ -1804,7 +1780,8 @@ class AddQuantDequantPass(object):
"squeeze", "elementwise_sub", "mul", "matmul", "relu", "relu6",
"leaky_relu", "tanh", "swish", "scale", "transpose", "transpose2",
"sigmoid", "pad2d", "flatten", "flatten2", "batch_norm", "layer_norm",
"matmul_v2"
"matmul_v2", "split", "flatten_contiguous_range", "squeeze2",
"nearest_interp_v2", "fill_constant_batch_size_like"
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

加逗号

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done

]

# To be compatible with PaddleSlim, not remove _activation_type for now
Expand Down