wehub-resource-sync
675 lines
25 KiB
YAML
675 lines
25 KiB
YAML
- op : adam_
|
|
version :
|
|
- checkpoint : Upgrade adam add 1 attribute [multi_precision].
|
|
action :
|
|
- add_attr : multi_precision
|
|
comment : (bool) Whether to use multi-precision during weight updating.
|
|
default : "false"
|
|
- checkpoint : Upgrade adam, add 1 dispensable input [EpsilonTensor].
|
|
action :
|
|
- add_input : EpsilonTensor
|
|
comment : If provided, Adam will use this as epsilon, this has a higher priority than attr(epsilon). For better performance in npu kernel.
|
|
- checkpoint : Upgrade adam, add 1 attribute [use_global_beta_pow].
|
|
action :
|
|
- add_attr : use_global_beta_pow
|
|
comment : If true, Adam will use global beta_pow for whole model instead of creating beta_pow for each parameter. In that case, the outputs(Beta1PowOut, Beta2PowOut) will not be used in adam op, and beta_pow will be updated after all adam op in the model.
|
|
default : "false"
|
|
- checkpoint : Upgrade adam, add 1 dispensable input [SkipUpdate].
|
|
action :
|
|
- add_input : SkipUpdate
|
|
comment : If the value is true, Adam will skip the update.
|
|
|
|
- op : affine_grid
|
|
version :
|
|
- checkpoint : Compatible upgrade of affine_grid, add a new attribute [align_corners].
|
|
action :
|
|
- add_attr : align_corners
|
|
comment : Whether to align the corners of input and output.
|
|
default : "true"
|
|
|
|
- op : allclose
|
|
version :
|
|
- checkpoint : Upgrade allclose, add two new inputs [Rtol] and [Atol].
|
|
action:
|
|
- add_input : Rtol
|
|
comment : The added input 'Rtol' is not dispensable.
|
|
- add_input : Atol
|
|
comment : The added input 'Atol' is not dispensable.
|
|
- checkpoint : Delete two float attributes [rtol] and [atol],
|
|
then add 2 string attributes [atol, rtol]. Don't be surprised.
|
|
This is because float cannot represent height-precision
|
|
floating-point values, and our framework doesn't support
|
|
the use of double attributes. As a result, string instead
|
|
of double is used here to represent height-precision
|
|
floating-point values.
|
|
action :
|
|
- add_attr : rtol
|
|
comment : The relative tolerance. Default::math:`1e-5` .
|
|
default : std::string("1e-5")
|
|
- delete_attr : rtol
|
|
comment : The attribute 'rtol' is deleted. The reason why it is deleted is that
|
|
attributes do not support a float64 value and it is changed to a tensor.
|
|
- add_attr : atol
|
|
comment : (string) The absolute tolerance. Default::math:`1e-8` .
|
|
default : std::string("1e-5")
|
|
- delete_attr : atol
|
|
comment : The attribute 'atol' is deleted. The reason why it is deleted is that
|
|
attributes do not support a float64 value and it is changed to a tensor.
|
|
|
|
- op : argsort
|
|
version :
|
|
- checkpoint : Upgrade agsort, add a new attribute [stable]
|
|
action :
|
|
- add_attr : stable
|
|
comment : If true, it will use stable sorting algorithm which preserves the order
|
|
of equivalent elements. Otherwise, the order of equivalent elements will
|
|
not be guaranteed to be preserved.
|
|
default : "false"
|
|
|
|
- op : assign_value
|
|
version :
|
|
- checkpoint : Upgrade assign_value, remove plain attributes in favor of generic attribute.
|
|
action :
|
|
- add_attr : values
|
|
comment : replace generic types with scalar.
|
|
default : std::vector<paddle::experimental::Scalar>()
|
|
- delete_attr : bool_values
|
|
comment : remove plain attributes.
|
|
- delete_attr : fp32_values
|
|
comment : remove plain attributes.
|
|
- delete_attr : int32_values
|
|
comment : remove plain attributes.
|
|
- delete_attr : int64_values
|
|
comment : remove plain attributes.
|
|
|
|
- op : auc
|
|
version :
|
|
- checkpoint : Upgrade auc, add a new input [InsTagWeight].
|
|
action :
|
|
- add_input : ValueTensor
|
|
comment : In order to support multi-tag task.
|
|
|
|
- op : c_split
|
|
version :
|
|
- checkpoint : Upgrade c_split delete 1 attribute[use_calc_stream].
|
|
action :
|
|
- delete_attr : use_calc_stream
|
|
comment : eject CUDA operations to calculation stream.
|
|
default : false
|
|
|
|
- op : clip
|
|
version :
|
|
- checkpoint : Upgrade clip add a new input [Min]
|
|
action :
|
|
- add_input : Min
|
|
comment : Pass the mix, min value as input, not attribute. Min is dispensable.
|
|
- add_input : Max
|
|
comment : Pass the mix, min value as input, not attribute. Max is dispensable.
|
|
|
|
- op : coalesce_tensor
|
|
version :
|
|
- checkpoint : "Upgrade coalesce_tensor: add a new attribute [use_align]."
|
|
action :
|
|
- add_attr : use_align
|
|
comment : In order to optionally take memory alignment into account when
|
|
coalescing tensors. The default value is true to be compatible
|
|
with before.
|
|
default : "true"
|
|
- checkpoint : "Upgrade coalesce_tensor: add a new attribute [align_size]."
|
|
action :
|
|
- add_attr : align_size
|
|
comment : In order to optionally take memory alignment into account when
|
|
coalescing tensors. The default value is -1 and use the default
|
|
align_size
|
|
of each place to be compatible with before.
|
|
default : -1
|
|
|
|
- op : collect_fpn_proposals
|
|
version :
|
|
- checkpoint : Upgrade collect_fpn_proposals add a new input [MultiLevelRoIsNum] and add a new output [RoisNum].
|
|
action :
|
|
- add_input : multi_level_rois_num
|
|
comment : The RoIs' number of each image on multiple levels. The number on each level has the shape of (N), N is the number of images.
|
|
- add_output : rois_num
|
|
comment : The number of RoIs in each image.
|
|
|
|
- op : conv2d
|
|
version :
|
|
- checkpoint : Upgrade conv2d, add a new attribute [use_addto].
|
|
action :
|
|
- add_attr : use_addto
|
|
comment : In order to support new feature (inplace addto strategy) for
|
|
gradient accumulation.
|
|
default : "false"
|
|
|
|
- op : conv2d_transpose
|
|
version :
|
|
- checkpoint : Upgrade convtranspose add a new attribute [output_padding].
|
|
action :
|
|
- add_attr : output_padding
|
|
comment : In order to add additional size to one side of each dimension in the output.
|
|
default : "std::vector<int>{}"
|
|
- checkpoint : Upgrade conv2d transpose to add a new attributes [force_fp32_output, mkldnn_data_type].
|
|
action :
|
|
- add_attr : force_fp32_output
|
|
comment : Force BF16 kernel output FP32, only used in MKL-DNN BF16.
|
|
default : "false"
|
|
- add_attr : mkldnn_data_type
|
|
comment : Data type of mkldnn kernel.
|
|
default : "\"float32\""
|
|
|
|
- op : conv3d
|
|
version :
|
|
- checkpoint : Upgrade conv3d, add a new attribute [use_addto].
|
|
action :
|
|
- add_attr : use_addto
|
|
comment : In order to support new feature (inplace addto strategy) for
|
|
gradient accumulation.
|
|
default : "false"
|
|
|
|
- op : conv3d_transpose
|
|
version :
|
|
- checkpoint : Upgrade convtranspose add a new attribute [output_padding].
|
|
action :
|
|
- add_attr : output_padding
|
|
comment : In order to add additional size to one side of each dimension in the output.
|
|
default : "std::vector<int>{}"
|
|
|
|
- op : conv_transpose
|
|
version :
|
|
- checkpoint : Upgrade convtranspose add a new attribute [output_padding].
|
|
action :
|
|
- add_attr : output_padding
|
|
comment : In order to add additional size to one side of each dimension in the output.
|
|
default : "std::vector<int>{}"
|
|
|
|
- op : cumprod
|
|
version :
|
|
- checkpoint : Upgrade cumprod add 2 new attributes [exlucsive] and [reverse].
|
|
action :
|
|
- add_attr : exlucsive
|
|
comment : In order to perform exclusive cumprod.
|
|
default : "false"
|
|
- add_attr : reverse
|
|
comment : In order to perform cumprod in the opposite direction
|
|
default : "false"
|
|
|
|
- op : cumsum
|
|
version :
|
|
- checkpoint : Upgrade cumsum add a new attribute [flatten].
|
|
action :
|
|
- add_attr : flatten
|
|
comment : In order to compute the cumsum over the flattened array when the argument `axis` in python API is None.
|
|
default : "false"
|
|
|
|
- op : depthwise_conv2d
|
|
version :
|
|
- checkpoint : Upgrade depthwise_conv2d, add a new attribute [use_addto].
|
|
action :
|
|
- add_attr : use_addto
|
|
comment : In order to support new feature (inplace addto strategy) for
|
|
gradient accumulation.
|
|
default : "false"
|
|
|
|
- op : depthwise_conv2d_transpose
|
|
version :
|
|
- checkpoint : Upgrade convtranspose add a new attribute [output_padding].
|
|
action :
|
|
- add_attr : output_padding
|
|
comment : In order to add additional size to one side of each dimension in the output.
|
|
default : "std::vector<int>{}"
|
|
|
|
- op : elementwise_floordiv
|
|
version :
|
|
- checkpoint : Register elementwise_floordiv for adding the attribute of Scale_y
|
|
action :
|
|
- add_attr : Scale_y
|
|
comment : In order to support the function of scaling the input Y when using the operator of elementwise_floordiv.
|
|
default : 1.0
|
|
|
|
- op : elementwise_max
|
|
version :
|
|
- checkpoint : Register elementwise_max for adding the attribute of Scale_y.
|
|
action :
|
|
- add_attr : Scale_y
|
|
comment : In order to support the function of scaling the input Y when using the operator of elementwise_max.
|
|
default : 1.0
|
|
|
|
- op : elementwise_min
|
|
version :
|
|
- checkpoint : Register elementwise_min for adding the attribute of Scale_y.
|
|
action :
|
|
- add_attr : Scale_y
|
|
comment : In order to support the function of scaling the input Y when using the operator of elementwise_min.
|
|
default : 1.0
|
|
|
|
- op : elementwise_mod
|
|
version :
|
|
- checkpoint : Register elementwise_mod for adding the attribute of Scale_y
|
|
action :
|
|
- add_attr : Scale_y
|
|
comment : In order to support the function of scaling the input Y when using the operator of elementwise_mod.
|
|
default : "false"
|
|
|
|
- op : elementwise_pow
|
|
version :
|
|
- checkpoint : Register elementwise_pow for adding the attribute of Scale_y
|
|
action :
|
|
- add_attr : Scale_y
|
|
comment : In order to support the function of scaling the input Y when using the operator of elementwise_pow.
|
|
default : 1.0
|
|
|
|
- op : embedding
|
|
version :
|
|
- checkpoint : Upgrade flip, add new attr [axis] and delete attr [dims]
|
|
action :
|
|
- fix_bug : fix_bug
|
|
comment : lookup_table_v2 support input type `int64`; after support input type `int32/int64`
|
|
|
|
- op : equal
|
|
version :
|
|
- checkpoint : Upgrade compare ops, add a new attribute [force_cpu]
|
|
action :
|
|
- modify_attr : force_cpu
|
|
comment : In order to force fill output variable to gpu memory.
|
|
default : "false"
|
|
|
|
- op : expand_as_v2
|
|
version :
|
|
- checkpoint : fix expand_as_v2 and add new input [Y].
|
|
action :
|
|
- add_input : Y
|
|
comment : Expand X according to the shape of Y.
|
|
|
|
- op : flip
|
|
version :
|
|
- checkpoint : Upgrade flip, add new attr [axis] and delete attr [dims]
|
|
action :
|
|
- add_attr : axis
|
|
comment : The added attr 'axis' doesn't set default value
|
|
default : paddle::none
|
|
- delete_attr : dims
|
|
comment : The attr 'dims' is deleted.
|
|
|
|
- op : gather
|
|
version :
|
|
- checkpoint : Upgrade gather, add a new input [Axis]
|
|
action :
|
|
- add_input : Axis
|
|
comment : Specify the axis of gather operation.
|
|
|
|
- op : gaussian_random
|
|
version :
|
|
- checkpoint : Upgrade gaussian_random add new inputs [ShapeTensor] and [ShapeTensorList]
|
|
and modify the attribute of [shape]
|
|
action :
|
|
- add_input : ShapeTensor
|
|
comment : The output shape supports Tensor type. ShapeTensor is dispensable.
|
|
- add_input : ShapeTensorList
|
|
comment : The output shape supports list filled with Tensor. ShapeTensorList is dispensable.
|
|
- modify_attr : shape
|
|
comment : "The arg 'default_value' of attr 'shape' is changed: from 'None' to '{}'."
|
|
default : std::vector<int64_t>{}
|
|
- checkpoint : Upgrade gaussian_random, change the type of attributes [mean] and [std] from float to double to avoid float32 truncation that breaks float64 output precision.
|
|
action :
|
|
- modify_attr : mean
|
|
comment : "The type of attr 'mean' is changed from float to double, so that the caller's double-precision value is no longer truncated to float32 before being used by the kernel. The numeric default value is unchanged (0.0)."
|
|
default : 0.0
|
|
- modify_attr : std
|
|
comment : "The type of attr 'std' is changed from float to double, so that the caller's double-precision value is no longer truncated to float32 before being used by the kernel. The numeric default value is unchanged (1.0)."
|
|
default : 1.0
|
|
|
|
- op : generate_proposals
|
|
version :
|
|
- checkpoint : Register generate_proposals_v2 for adding the attribute of pixel_offset
|
|
action :
|
|
- add_attr : pixel_offset
|
|
comment : If true, im_shape pixel offset is 1.
|
|
default : "true"
|
|
|
|
- op : global_gather
|
|
version :
|
|
- checkpoint : Upgrade global_gather delete 1 attribute[use_calc_stream].
|
|
action :
|
|
- delete_attr : use_calc_stream
|
|
comment : eject CUDA operations to calculation stream.
|
|
default : false
|
|
|
|
- op : global_scatter
|
|
version :
|
|
- checkpoint : Upgrade global_scatter delete 1 attribute[use_calc_stream].
|
|
action :
|
|
- delete_attr : use_calc_stream
|
|
comment : eject CUDA operations to calculation stream.
|
|
default : false
|
|
|
|
- op : greater_equal
|
|
version :
|
|
- checkpoint : Upgrade compare ops, add a new attribute [force_cpu]
|
|
action :
|
|
- modify_attr : force_cpu
|
|
comment : In order to force fill output variable to gpu memory.
|
|
default : "false"
|
|
|
|
- op : greater_than
|
|
version :
|
|
- checkpoint : Upgrade compare ops, add a new attribute [force_cpu]
|
|
action :
|
|
- modify_attr : force_cpu
|
|
comment : In order to force fill output variable to gpu memory.
|
|
default : "false"
|
|
|
|
- op : grid_sample
|
|
version :
|
|
- checkpoint : Upgrade grid_sampler add a new attribute [mode]
|
|
action :
|
|
- add_attr : mode
|
|
comment : In order to specify interpolation mode
|
|
default : std::string("bilinear")
|
|
|
|
- op : histogram
|
|
version :
|
|
- checkpoint : Upgrade histogram, add a new Input [Weight] and a new attribute [density]
|
|
action :
|
|
- add_input : Weight
|
|
comment : The weight of each value in the input tensor.
|
|
- add_attr : density
|
|
comment : If true, the histogram is normalized to form a probability density.
|
|
default : "false"
|
|
|
|
- op : instance_norm
|
|
version :
|
|
- checkpoint : Change dispensable of attribute from False to True in instance_norm.
|
|
action :
|
|
- modify_attr : Bias
|
|
comment : "The arg 'dispensable' of Input 'Bias' is changed: from 'False' to 'True'."
|
|
default : "true"
|
|
- modify_attr : Scale
|
|
comment : "The arg 'dispensable' of Input 'Scale' is changed: from 'False' to 'True'."
|
|
default : "true"
|
|
|
|
- op : kldiv_loss
|
|
version :
|
|
- checkpoint : Upgrade kldiv_loss, add a new attribute [log_target]
|
|
action :
|
|
- add_attr : log_target
|
|
comment : In order to specify whether 'label' is passed in log space.
|
|
default : "false"
|
|
|
|
- op : lamb
|
|
version :
|
|
- checkpoint : Upgrade lamb, add two new outputs [Beta1PowOut] and [Beta2PowOut].
|
|
action :
|
|
- add_output : Beta1PowOut
|
|
comment : The Output beta1 power accumulator. 'Beta1PowOut' is dispensable.
|
|
- add_output : Beta2PowOut
|
|
comment : The Output beta2 power accumulator. 'Beta2PowOut' is dispensable.
|
|
|
|
- op : legacy_generate_proposals
|
|
version :
|
|
- checkpoint : Incompatible upgrade of output [RpnRoisLod].
|
|
action :
|
|
- delete_output : RpnRoisLod
|
|
comment : Delete RpnRoisLod due to incorrect output name and it is not used in object detection models yet.
|
|
- checkpoint : Upgrade generate_proposals add a new output [RpnRoisNum]
|
|
action :
|
|
- add_output : rpn_rois_num
|
|
comment : The number of Rpn RoIs in each image. RpnRoisNum is dispensable.
|
|
|
|
- op : less_equal
|
|
version :
|
|
- checkpoint : Upgrade compare ops, add a new attribute [force_cpu]
|
|
action :
|
|
- modify_attr : force_cpu
|
|
comment : In order to force fill output variable to gpu memory.
|
|
default : "false"
|
|
|
|
- op : less_than
|
|
version :
|
|
- checkpoint : Upgrade compare ops, add a new attribute [force_cpu]
|
|
action :
|
|
- modify_attr : force_cpu
|
|
comment : In order to force fill output variable to gpu memory.
|
|
default : "false"
|
|
|
|
- op : linspace
|
|
version :
|
|
- checkpoint : Upgrade linspace to add a new attribute [dtype]
|
|
action :
|
|
- add_attr : dtype
|
|
comment : In order to change output data type
|
|
default : 5
|
|
|
|
- op : lstsq
|
|
version :
|
|
- checkpoint : Upgrade lstsq, add 1 outputs [Residuals].
|
|
action :
|
|
- add_output : Residuals
|
|
comment : Output tensor of lstsq operator, meaning the squared residuals of the calculated solutions.
|
|
|
|
- op : matrix_nms
|
|
version :
|
|
- checkpoint : Upgrade matrix_nms, add a new output [RoisNum].
|
|
action :
|
|
- add_output : RoisNum
|
|
comment : The number of RoIs in each image.
|
|
|
|
- op : max_pool2d_with_index
|
|
version :
|
|
- checkpoint : Upgrade max_pool2d_with_index, add a new attribute [ceil_mode].
|
|
action :
|
|
- add_attr : ceil_mode
|
|
comment : When true, will use ceil instead of floor to compute the output shape.
|
|
default : "false"
|
|
|
|
- op : max_pool3d_with_index
|
|
version :
|
|
- checkpoint : Upgrade max_pool3d_with_index, add a new attribute [ceil_mode].
|
|
action :
|
|
- add_attr : ceil_mode
|
|
comment : When true, will use ceil instead of floor to compute the output shape.
|
|
default : "false"
|
|
|
|
- op : momentum
|
|
version :
|
|
- checkpoint : Upgrade momentum add 4 attributes [regularization_method, regularization_coeff, multi_precision, rescale_grad].
|
|
action :
|
|
- add_input : MasterParam
|
|
comment : FP32 master weight for AMP.
|
|
- add_output : MasterParamOut
|
|
comment : The updated FP32 master weight for AMP. It shared memory with Input(MasterParam).
|
|
- add_attr : regularization_method
|
|
comment : (string) regularization_method, right now only support l2decay or none
|
|
default : std::string("")
|
|
- add_attr : regularization_coeff
|
|
comment : (float) regularization_coeff
|
|
default : 0.0
|
|
- add_attr : multi_precision
|
|
comment : (bool) Whether to use multi-precision during weight updating.
|
|
default : "false"
|
|
- add_attr : rescale_grad
|
|
comment : (float) Multiply the gradient with `rescale_grad` before updating. Often choose to be `1.0/batch_size`.
|
|
default : 1.0
|
|
|
|
- op : not_equal
|
|
version :
|
|
- checkpoint : Upgrade compare ops, add a new attribute [force_cpu]
|
|
action :
|
|
- modify_attr : force_cpu
|
|
comment : In order to force fill output variable to gpu memory.
|
|
default : "false"
|
|
|
|
- op : p_norm
|
|
version :
|
|
- checkpoint : Upgrade p_norm, add 1 attribute [asvector].
|
|
action :
|
|
- add_attr : asvector
|
|
comment : Compute as vector when axis is None and input is matrix.
|
|
default : "false"
|
|
|
|
- op : partial_allgather
|
|
version :
|
|
- checkpoint : Upgrade partial_allgather delete 1 attribute[use_calc_stream].
|
|
action :
|
|
- delete_attr : use_calc_stream
|
|
comment : eject CUDA operations to calculation stream.
|
|
default : false
|
|
|
|
- op : partial_send
|
|
version :
|
|
- checkpoint : Upgrade partial_send delete 1 attribute[use_calc_stream].
|
|
action :
|
|
- delete_attr : use_calc_stream
|
|
comment : eject CUDA operations to calculation stream.
|
|
default : false
|
|
|
|
- op : pixel_shuffle
|
|
version :
|
|
- checkpoint : Compatible upgrade of pixel_shuffle, add a new attribute [data_format]
|
|
action :
|
|
- add_attr : data_format
|
|
comment : Specify the data format of the input data
|
|
default : "true"
|
|
|
|
- op : roi_align
|
|
version :
|
|
- checkpoint : Incompatible upgrade of input [RpnRoisLod])
|
|
action :
|
|
- delete_input : RpnRoisLod
|
|
comment : Delete RpnRoisLod due to incorrect input name and it is not used in object detection models yet
|
|
- checkpoint : Upgrade roi_pool add a new input [RoisNum]
|
|
action :
|
|
- add_input : RoisNum
|
|
comment : The number of RoIs in each image. RoisNum is dispensable
|
|
- checkpoint : Upgrade roi_align add a new input [aligned]
|
|
action :
|
|
- add_attr : aligned
|
|
comment : If true, pixel shift it by -0.5 for align more perfectly.
|
|
default : "false"
|
|
|
|
- op : roi_pool
|
|
version :
|
|
- checkpoint : Incompatible upgrade of input [RpnRoisLod]
|
|
action :
|
|
- delete_input : RpnRoisLod
|
|
comment : Delete RpnRoisLod due to incorrect input name and it is not used in object detection models yet.
|
|
- checkpoint : Upgrade roi_pool add a new input [RoisNum]
|
|
action :
|
|
- add_input : RoisNum
|
|
comment : The number of RoIs in each image. RoisNum is dispensable
|
|
|
|
- op : roll
|
|
version :
|
|
- checkpoint : Upgrade roll add 1 attribute [axis], delete 1 attribute[dims].
|
|
action :
|
|
- add_attr : axis
|
|
comment : Axis along which to roll. It must have the same size with shifts, or size = 0.
|
|
default : std::vector<float>()
|
|
- delete_attr : dims
|
|
comment : Dims along which to roll. It must have the same size with shifts, or size = 0
|
|
- checkpoint : Upgrade roll add a dispensable input "ShiftsTensor"
|
|
action :
|
|
- add_input : ShiftsTensor
|
|
comment : The number of places by which the elements of the tensor are shifted.
|
|
|
|
- op : round
|
|
version :
|
|
- checkpoint : Add a new attribute [decimals] to round
|
|
action :
|
|
- add_attr : decimals
|
|
comment : The number of decimal places rounded
|
|
default : 0.0
|
|
|
|
- op : softmax_with_cross_entropy
|
|
version :
|
|
- checkpoint : Add a new attribute [use_softmax]
|
|
action :
|
|
- add_attr : use_softmax
|
|
comment : A flag to indicate whether to do softmax
|
|
default : "true"
|
|
|
|
- op : thresholded_relu
|
|
version :
|
|
- checkpoint : Upgrade thresholded_relu, add a new attribute [value]
|
|
action :
|
|
- add_attr : value
|
|
comment : The threshold value of thresholded_relu.
|
|
default : 0.0
|
|
|
|
- op : trace
|
|
version :
|
|
- checkpoint : Upgrade trace add a new attribute [axis2]
|
|
action :
|
|
- add_attr : axis1
|
|
comment : The added attribute 'axis1' is not yet registered.
|
|
default : std::vector<float>{0.0f}
|
|
- add_attr : axis2
|
|
comment : The added attribute 'axis2' is not yet registered.
|
|
default : std::vector<float>{1.0f}
|
|
- delete_attr : dim1
|
|
comment : The attribute 'dim1' is not recommend according to the specification 2.0.
|
|
- delete_attr : dim2
|
|
comment : The attribute 'dim2' is not recommend according to the specification 2.0.
|
|
|
|
- op : truncated_gaussian_random
|
|
version :
|
|
- checkpoint : Upgrade truncated_gaussian_random add 2 new attribute [a, b]
|
|
action :
|
|
- add_attr : a
|
|
comment : The minimum cutoff value.
|
|
default: -2.0
|
|
- add_attr : b
|
|
comment : The maximum cutoff value.
|
|
default: 2.0
|
|
|
|
- op : unique_consecutive
|
|
version :
|
|
- checkpoint : Upgrade unique_consecutive, add 2 outputs [Indices, Counts] and 3 attribute [return_inverse, return_counts, axis].
|
|
action :
|
|
- add_output : Counts
|
|
comment : The counts for each unique element.
|
|
- add_attr : return_inverse
|
|
comment : If True, also return the indices for where elements in the original input ended up in the returned unique tensor.
|
|
default : "false"
|
|
- add_attr : return_counts
|
|
comment : If True, also return the counts for each unique element.
|
|
default : "false"
|
|
- add_attr : axis
|
|
comment : The axis to apply unique. If None, the input will be flattened.
|
|
default : std::vector<int>{}
|
|
|
|
- op : weight_dequantize
|
|
version :
|
|
- checkpoint : Upgrade weight_dequantize, add a new attribute [group_size]
|
|
action :
|
|
- add_attr : group_size
|
|
comment : The group size of the dequantization scales.
|
|
default : -1
|
|
|
|
- op : weight_only_linear
|
|
version :
|
|
- checkpoint : Upgrade weight_only_linear, add a new attribute [group_size]
|
|
action :
|
|
- add_attr : group_size
|
|
comment : The group size of the dequantization scales.
|
|
default : -1
|
|
|
|
- op : weight_quantize
|
|
version :
|
|
- checkpoint : Upgrade weight_quantize, add a new attribute [group_size]
|
|
action :
|
|
- add_attr : group_size
|
|
comment : The group size of the quantization scales.
|
|
default : -1
|
|
|
|
- op : yolo_box
|
|
version :
|
|
- checkpoint : Upgrade yolo box to add new attribute [iou_aware, iou_aware_factor].
|
|
action :
|
|
- add_attr : iou_aware
|
|
comment : Whether use iou aware.
|
|
default : "false"
|
|
- add_attr : iou_aware_factor
|
|
comment : iou aware factor.
|
|
default : 0.5f
|