ai.onnx

13

1, 6

X (differentiable) : T — Input tensor

Y (differentiable) : T — Output tensor

None.

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input and output types to all numeric tensors.

test_abs

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

7

input (differentiable) : T — Input tensor

output (differentiable) : T — The arccosine of the input tensor computed element-wise

None.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_acos_example, test_acos

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

9

input (differentiable) : T — Input tensor

output (differentiable) : T — The hyperbolic arccosine values of the input tensor computed element-wise

None.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_acosh_example, test_acosh

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

14

1, 6, 7, 13

A (differentiable) : T — First operand.
B (differentiable) : T — Second operand.

C (differentiable) : T — Result, has same element type as two inputs

None.

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input and output types to all numeric tensors.

test_add, test_add_int8, test_add_int16, test_add_uint8, test_add_uint16, test_add_uint32, test_add_uint64, test_add_bcast

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

20

theta (non-differentiable) : T1 — input batch of affine matrices with shape (N, 2, 3) for 2D or (N, 3, 4) for 3D
size (non-differentiable) : T2 — the target output image size (N, C, H, W) for 2D or (N, C, D, H, W) for 3D

grid (differentiable) : T1 — output tensor of shape (N, H, W, 2) of 2D sample coordinates or (N, D, H, W, 3) of 3D sample coordinates.

align_corners : int (default is 0) — if align_corners=1, consider -1 and 1 to refer to the centers of the corner pixels. if align_corners=0, consider -1 and 1 to refer to the outer edge the corner pixels.

T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain grid types to float tensors.
T2 : tensor(int64) — Constrain size’s type to int64 tensors.

None published for this operator.

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

7

1

A (non-differentiable) : T — First input operand for the logical operator.
B (non-differentiable) : T — Second input operand for the logical operator.

C (non-differentiable) : T1 — Result tensor.

None.

T : tensor(bool) — Constrain input to boolean tensor.
T1 : tensor(bool) — Constrain output to boolean tensor.

test_and2d, test_and3d, test_and4d, test_and_bcast3v1d, test_and_bcast3v2d, test_and_bcast4v2d, test_and_bcast4v3d, test_and_bcast4v4d

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

1, 11, 12

data (non-differentiable) : T — An input tensor.

reduced (non-differentiable) : tensor(int64) — Reduced output tensor with integer data type.

axis : int (default is 0) — The axis in which to compute the arg indices. Accepted range is [-r, r-1] where r = rank(data).
keepdims : int (default is 1) — Keep the reduced dimension or not, default 1 means keep reduced dimension.
select_last_index : int (default is 0) — Whether to select the last index or the first index if the {name} appears in multiple indices, default is False (first index).

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input and output types to all numeric tensors.

test_argmax_default_axis_example, test_argmax_default_axis_random, test_argmax_default_axis_example_select_last_index, test_argmax_default_axis_random_select_last_index, test_argmax_keepdims_example, test_argmax_keepdims_random, test_argmax_keepdims_example_select_last_index, test_argmax_keepdims_random_select_last_index, test_argmax_negative_axis_keepdims_example, test_argmax_negative_axis_keepdims_random, test_argmax_negative_axis_keepdims_example_select_last_index, test_argmax_negative_axis_keepdims_random_select_last_index, test_argmax_no_keepdims_example, test_argmax_no_keepdims_random, test_argmax_no_keepdims_example_select_last_index, test_argmax_no_keepdims_random_select_last_index

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

1, 11, 12

data (non-differentiable) : T — An input tensor.

reduced (non-differentiable) : tensor(int64) — Reduced output tensor with integer data type.

axis : int (default is 0) — The axis in which to compute the arg indices. Accepted range is [-r, r-1] where r = rank(data).
keepdims : int (default is 1) — Keep the reduced dimension or not, default 1 means keep reduced dimension.
select_last_index : int (default is 0) — Whether to select the last index or the first index if the {name} appears in multiple indices, default is False (first index).

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input and output types to all numeric tensors.

test_argmin_default_axis_example, test_argmin_default_axis_random, test_argmin_default_axis_example_select_last_index, test_argmin_default_axis_random_select_last_index, test_argmin_keepdims_example, test_argmin_keepdims_random, test_argmin_keepdims_example_select_last_index, test_argmin_keepdims_random_select_last_index, test_argmin_negative_axis_keepdims_example, test_argmin_negative_axis_keepdims_random, test_argmin_negative_axis_keepdims_example_select_last_index, test_argmin_negative_axis_keepdims_random_select_last_index, test_argmin_no_keepdims_example, test_argmin_no_keepdims_random, test_argmin_no_keepdims_example_select_last_index, test_argmin_no_keepdims_random_select_last_index

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

7

input (differentiable) : T — Input tensor

output (differentiable) : T — The arcsine of the input tensor computed element-wise

None.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_asin_example, test_asin

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

9

input (differentiable) : T — Input tensor

output (differentiable) : T — The hyperbolic arcsine values of the input tensor computed element-wise

None.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_asinh_example, test_asinh

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

7

input (differentiable) : T — Input tensor

output (differentiable) : T — The arctangent of the input tensor computed element-wise

None.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_atan_example, test_atan

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

9

input (differentiable) : T — Input tensor

output (differentiable) : T — The hyperbolic arctangent values of the input tensor computed element-wise

None.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_atanh_example, test_atanh

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

25

23, 24

Q : T1 — Query tensor. 4D tensor with shape (batch_size, q_num_heads, q_sequence_length, head_size) or 3D tensor with shape (batch_size, q_sequence_length, q_hidden_size). For cases with a 3D input tensor, q_hidden_size = q_num_heads * head_size
K : T1 — Key tensor. 4D tensor with shape (batch_size, kv_num_heads, kv_sequence_length, head_size) or 3D tensor with shape (batch_size, kv_sequence_length, k_hidden_size). For cases with a 3D input tensor, k_hidden_size = kv_num_heads * head_size
V : T2 — Value tensor. 4D tensor with shape (batch_size, kv_num_heads, kv_sequence_length, v_head_size) or 3D tensor with shape (batch_size, kv_sequence_length, v_hidden_size). For cases with a 3D input tensor, v_hidden_size = kv_num_heads * v_head_size
attn_mask (optional) : U — Attention mask. Shape must be broadcastable to (batch_size, q_num_heads, q_sequence_length, total_sequence_length) where total_sequence_length = past_sequence_length + kv_sequence_length. The last dimension can also be shorter than total_sequence_length and will be padded to total_sequence_length with negative infinity. Two types of masks are supported: a boolean mask where a value of True indicates that the element should take part in attention, or a float mask of the same type as query, key, value that is added to the attention score.
past_key (optional) : T1 — Past state for key with shape (batch_size, kv_num_heads, past_sequence_length, head_size). Must be used together with past_value input.
past_value (optional) : T2 — Past state for value with shape (batch_size, kv_num_heads, past_sequence_length, v_head_size). Must be used together with past_key input.
nonpad_kv_seqlen (optional) : tensor(int64) — A vector of integers of shape (batch_size,) that indicates the number of valid (i.e., non-padding) tokens in each sample. A padding mask can be derived from this. This should not be used together with past_key and past_value inputs or present_key and present_value outputs (see the KV cache use cases in the operator description).

Y : T1 — The output tensor. 4D tensor with shape (batch_size, q_num_heads, q_sequence_length, v_head_size) or 3D tensor with shape (batch_size, q_sequence_length, hidden_size). For cases with a 3D input tensor, hidden_size = q_num_heads * v_head_size
present_key (optional) : T1 — Updated key cache with shape (batch_size, kv_num_heads, total_sequence_length, head_size) where total_sequence_length = past_sequence_length + kv_sequence_length.
present_value (optional) : T2 — Updated value cache with shape (batch_size, kv_num_heads, total_sequence_length, v_head_size) where total_sequence_length = past_sequence_length + kv_sequence_length.
qk_matmul_output (optional) : T1 — The output of QK matmul. 4D tensor with shape (batch_size, q_num_heads, q_sequence_length, total_sequence_length) where total_sequence_length = past_sequence_length + kv_sequence_length.

is_causal : int (default is 0) — If set to 1, causal masking is applied. For a square Q/K (no cache offset) this is a lower-triangular matrix. In general the mask is bottom-right (offset-aware): query in-block index i attends key j iff j <= i + offset, where offset is the count of valid keys preceding the query block (past_sequence_length for an internal past_key cache, or nonpad_kv_seqlen - q_sequence_length per batch for an external cache). When offset = 0 this reduces to the lower-triangular (top-left) mask.
kv_num_heads : int — Number of heads of key and value. Must be used with 3D inputs of Q, K and V.
left_window_size : int (default is -1) — Maximum number of positions to the left of the current absolute query position that may be attended. A value of 0 allows the current position but no preceding position, while -1 leaves the left side unbounded. This bound is composed with is_causal and attn_mask.
q_num_heads : int — Number of heads of query. Must be used with 3D inputs of Q, K and V.
qk_matmul_output_mode : int (default is 0) — Determines what the optional 4th output contains: 0 (default): raw QK matmul result; 1: after softcap (before bias addition); 2: QK + softcap + bias; 3: post-softmax probabilities (after fully-masked-row guard). In mode 3, a fully-masked query row (every key disallowed) is a zero row, consistent with the corresponding row of the primary output Y. The mode-3 output is emitted at the operator’s output precision (T1); when softmax_precision differs from T1 this is a cast of the softmax result to T1.
right_window_size : int (default is -1) — Maximum number of positions to the right of the current absolute query position that may be attended. A value of 0 allows the current position but no following position, while -1 leaves the right side unbounded. Set is_causal=0 to use a positive right window.
scale : float — Scaling factor applied to $Q*K^T$. Default value is 1/sqrt(head_size). To prevent numerical overflow, scale Q, K by sqrt(scale) before matmul.
softcap : float (default is 0.0) — Soft cap for attention logits, applied as softcap * tanh(logits / softcap). Default value of 0.0 means no soft capping is applied. The soft cap is applied before mask / bias addition and softmax.
softmax_precision : int — Specifies the precision for softmax computation. If provided, the attention weights will be cast to this type before softmax and then cast back to the original type. Supported values are: 1 (FLOAT), 10 (FLOAT16), 11 (DOUBLE), 16 (BFLOAT16).

T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain Q and K inputs types to float tensors.
T2 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain V input types to float tensors.
U : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(bool) — Constrain output ‘mask’ types to boolean tensors and input types.

test_attention_4d, test_attention_23_boolmask_fullymasked_row_nan_robustness, test_attention_23_fullymasked_qk_matmul_output_mode3_zero, test_attention_24_fullymasked_qk_matmul_output_mode3_zero, test_attention_24_qk_matmul_output_mode3_softmax_precision, test_attention_3d, test_attention_3d_attn_mask, test_attention_3d_causal, test_attention_3d_causal_bf16, test_attention_3d_diff_heads_sizes, test_attention_3d_diff_heads_sizes_attn_mask, test_attention_3d_diff_heads_sizes_causal, test_attention_3d_diff_heads_sizes_scaled, test_attention_3d_diff_heads_sizes_softcap, test_attention_3d_diff_heads_with_past_and_present, test_attention_3d_gqa, test_attention_3d_gqa_attn_mask, test_attention_3d_gqa_causal, test_attention_3d_gqa_scaled, test_attention_3d_gqa_softcap, test_attention_3d_gqa_with_past_and_present, test_attention_3d_local_window, test_attention_3d_scaled, test_attention_3d_softcap, test_attention_3d_transpose_verification, test_attention_3d_with_past_and_present, test_attention_3d_with_past_and_present_qk_matmul, test_attention_3d_with_past_and_present_qk_matmul_bias, test_attention_3d_with_past_and_present_qk_matmul_softcap, test_attention_3d_with_past_and_present_qk_matmul_softmax, test_attention_4d_causal_nonpad_attn_mask_composition, test_attention_4d_causal_nonpad_batch_prefill, test_attention_4d_causal_nonpad_continued_prefill, test_attention_4d_causal_nonpad_negative_offset_structural_empty, test_attention_4d_causal_with_past_and_present, test_attention_4d_diff_heads_mask4d_padded_kv, test_attention_4d_gqa_causal_nonpad_decode, test_attention_4d_gqa_causal_nonpad_decode_fp16, test_attention_4d_attn_mask_3d, test_attention_4d_attn_mask_3d_causal, test_attention_4d_attn_mask_4d, test_attention_4d_attn_mask_4d_causal, test_attention_4d_attn_mask, test_attention_4d_attn_mask_bool, test_attention_4d_attn_mask_bool_4d, test_attention_4d_attn_mask_causal_bf16, test_attention_bidirectional_window, test_attention_4d_causal, test_attention_4d_causal_bf16, test_attention_causal_boolmask_nan_robustness, test_attention_4d_causal_fp16, test_attention_4d_causal_padded_kv_bf16, test_attention_4d_diff_heads_sizes, test_attention_4d_diff_heads_sizes_attn_mask, test_attention_4d_diff_heads_sizes_causal, test_attention_4d_diff_heads_sizes_scaled, test_attention_4d_diff_heads_sizes_softcap, test_attention_4d_diff_heads_with_past_and_present, test_attention_4d_diff_heads_with_past_and_present_mask3d, test_attention_4d_diff_heads_with_past_and_present_mask4d, test_attention_4d_fp16, test_attention_4d_gqa, test_attention_4d_gqa_attn_mask, test_attention_4d_gqa_causal, test_attention_4d_gqa_scaled, test_attention_4d_gqa_softcap, test_attention_4d_gqa_with_past_and_present, test_attention_4d_gqa_with_past_and_present_fp16, test_attention_local_window, test_attention_local_window_default, test_attention_local_window_ext_cache_float16_mask, test_attention_local_window_ext_cache_rank2_mask, test_attention_local_window_ext_cache_rank3_head_mask, test_attention_local_window_ext_cache_rank4_batch_mask, test_attention_local_window_gqa_rank4_mask, test_attention_local_window_rank1_boolean_mask, test_attention_local_window_with_past, test_attention_4d_padded_kv_bf16, test_attention_4d_scaled, test_attention_4d_softcap, test_attention_4d_softcap_neginf_mask, test_attention_4d_softcap_neginf_mask_poison, test_attention_4d_with_past_and_present, test_attention_4d_with_past_and_present_qk_matmul, test_attention_4d_with_past_and_present_qk_matmul_bias, test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask, test_attention_4d_with_past_and_present_qk_matmul_bias_3d_mask_causal, test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask, test_attention_4d_with_past_and_present_qk_matmul_bias_4d_mask_causal, test_attention_4d_with_qk_matmul, test_attention_4d_with_qk_matmul_bias, test_attention_4d_with_qk_matmul_softcap, test_attention_4d_with_qk_matmul_softmax

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

1, 7, 10, 11, 19

X (differentiable) : T — Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 …​ Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …​].

Y (differentiable) : T — Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used

auto_pad : string (default is NOTSET) — auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
ceil_mode : int (default is 0) — Whether to use ceil or floor (default) to compute the output shape.
count_include_pad : int (default is 0) — Whether include pad pixels when calculating values for the edges. Default is 0, doesn’t count include pad.
dilations : list of ints — Dilation value along each spatial axis of filter. If not present, the dilation defaults to 1 along each spatial axis.
kernel_shape : list of ints (required) — The size of the kernel along each axis.
pads : list of ints — Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…​x1_end, x2_end,…​], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides : list of ints — Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_averagepool_1d_default, test_averagepool_2d_ceil, test_averagepool_2d_ceil_last_window_starts_on_pad, test_averagepool_2d_default, test_averagepool_2d_dilations, test_averagepool_2d_pads, test_averagepool_2d_pads_count_include_pad, test_averagepool_2d_precomputed_pads, test_averagepool_2d_precomputed_pads_count_include_pad, test_averagepool_2d_precomputed_same_upper, test_averagepool_2d_precomputed_strides, test_averagepool_2d_same_lower, test_averagepool_2d_same_upper, test_averagepool_2d_strides, test_averagepool_3d_default, test_averagepool_3d_dilations_small

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

15

1, 6, 7, 9, 14

X (differentiable) : T — Input data tensor from the previous operator; dimensions are in the form of (N x C x D1 x D2 …​ Dn), where N is the batch size, C is the number of channels. Statistics are computed for every channel of C over N and D1 to Dn dimensions. For image data, input dimensions become (N x C x H x W). The op also accepts single dimension input of size N in which case C is assumed to be 1
scale (differentiable) : T1 — Scale tensor of shape ©.
B (differentiable) : T1 — Bias tensor of shape ©.
input_mean (differentiable) : T2 — running (training) or estimated (testing) mean tensor of shape ©.
input_var (differentiable) : T2 — running (training) or estimated (testing) variance tensor of shape ©.

Y (differentiable) : T — The output tensor of the same shape as X
running_mean (optional, non-differentiable) : T2 — The running mean after the BatchNormalization operator.
running_var (optional, non-differentiable) : T2 — The running variance after the BatchNormalization operator. This op uses the population size (N) for calculating variance, and not the sample size N-1.

epsilon : float (default is 1e-05) — The epsilon value to use to avoid division by zero.
momentum : float (default is 0.9) — Factor used in computing the running mean and variance.e.g., running_mean = running_mean * momentum + mean * (1 — momentum).
training_mode : int (default is 0) — If set to true, it indicates BatchNormalization is being used for training, and outputs 1 and 2 are to be computed.

T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input and output types to float tensors.
T1 : tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain scale and bias types to float tensors.
T2 : tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain mean and variance types to float tensors.

test_batchnorm_example, test_batchnorm_epsilon, test_batchnorm_example_training_mode, test_batchnorm_epsilon_training_mode

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

15

input : T1 — All values in input have to be in the range:[0, 1].

output : T2 — The returned output tensor only has values 0 or 1, same shape as input tensor.

dtype : int — The data type for the elements of the output tensor. if not specified, we will use the data type of the input tensor.
seed : float — (Optional) Seed to the random generator, if not specified we will auto generate one.

T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input types to float tensors.
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(bool) — Constrain output types to all numeric tensors and bool tensors.

test_bernoulli_double, test_bernoulli_seed, test_bernoulli

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

26

input (non-differentiable) : T1 — Input tensor to be bitcast.

output (non-differentiable) : T2 — Output tensor with the same shape as the input.

to : int (required) — The data type to which the input tensor is bitwise reinterpreted. Must be one of the non-string types from DataType enum in TensorProto. The target type must have the same bit-width as the input type.

T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2) — Constrain input types. Bitcasting from string is not supported.
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2) — Constrain output types. Bitcasting to string is not supported.

test_bitcast_2d_float32_to_int32, test_bitcast_bool_to_uint8, test_bitcast_float32_to_int32, test_bitcast_float64_to_int64, test_bitcast_int32_to_float32, test_bitcast_int64_to_float64, test_bitcast_int8_to_uint8, test_bitcast_scalar_float32_to_int32, test_bitcast_uint16_to_int16, test_bitcast_uint32_to_int32

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

28

11

X (non-differentiable) : T — First operand, input to be shifted.
Y (non-differentiable) : T — Second operand, amounts of shift.

Z (non-differentiable) : T — Output tensor

direction : string (required) — Direction of moving bits. It can be either “RIGHT” (for right shift) or “LEFT” (for left shift).

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64) — Constrain input and output types to integer tensors.

test_bitshift_left_int16, test_bitshift_left_int32, test_bitshift_left_int32_negative_shift, test_bitshift_left_int32_overflow, test_bitshift_left_int32_shift_ge_width, test_bitshift_left_int64, test_bitshift_left_int8, test_bitshift_left_int8_negative_shift, test_bitshift_left_int8_overflow, test_bitshift_left_int8_shift_ge_width, test_bitshift_left_uint16, test_bitshift_left_uint32, test_bitshift_left_uint64, test_bitshift_left_uint8, test_bitshift_right_int16, test_bitshift_right_int32, test_bitshift_right_int32_negative_input, test_bitshift_right_int32_negative_shift, test_bitshift_right_int32_shift_ge_width, test_bitshift_right_int64, test_bitshift_right_int8, test_bitshift_right_int8_negative_input, test_bitshift_right_int8_negative_shift, test_bitshift_right_int8_shift_ge_width, test_bitshift_right_uint16, test_bitshift_right_uint32, test_bitshift_right_uint64, test_bitshift_right_uint8

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

18

A (non-differentiable) : T — First input operand for the bitwise operator.
B (non-differentiable) : T — Second input operand for the bitwise operator.

C (non-differentiable) : T — Result tensor.

None.

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64) — Constrain input to integer tensors.

test_bitwise_and_i32_2d, test_bitwise_and_i16_3d, test_bitwise_and_ui64_bcast_3v1d, test_bitwise_and_ui8_bcast_4v3d

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

18

X (non-differentiable) : T — Input tensor

Y (non-differentiable) : T — Output tensor

None.

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64) — Constrain input/output to integer tensors.

test_bitwise_not_2d, test_bitwise_not_3d, test_bitwise_not_4d

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

18

A (non-differentiable) : T — First input operand for the bitwise operator.
B (non-differentiable) : T — Second input operand for the bitwise operator.

C (non-differentiable) : T — Result tensor.

None.

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64) — Constrain input to integer tensors.

test_bitwise_or_i32_2d, test_bitwise_or_i16_4d, test_bitwise_or_ui64_bcast_3v1d, test_bitwise_or_ui8_bcast_4v3d

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

18

A (non-differentiable) : T — First input operand for the bitwise operator.
B (non-differentiable) : T — Second input operand for the bitwise operator.

C (non-differentiable) : T — Result tensor.

None.

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64) — Constrain input to integer tensors.

test_bitwise_xor_ui64_bcast_3v1d, test_bitwise_xor_ui8_bcast_4v3d, test_bitwise_xor_i32_2d, test_bitwise_xor_i16_3d

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

17

size (non-differentiable) : T1 — A scalar value indicating the length of the window.

output (non-differentiable) : T2 — A Blackman window with length: size. The output has the shape: [size].

output_datatype : int (default is 1) — The data type of the output tensor. Strictly must be one of the values from DataType enum in TensorProto whose values correspond to T2. The default value is 1 = FLOAT.
periodic : int (default is 1) — If 1, returns a window to be used as periodic function. If 0, return a symmetric window. When ‘periodic’ is specified, hann computes a window of length size + 1 and returns the first size points. The default value is 1.

T1 : tensor(int32), tensor(int64) — Constrain the input size to int32_t or int64_t.
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain output types to numeric tensors.

test_blackmanwindow, test_blackmanwindow_symmetric

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

28

1, 6, 9, 13, 19, 21, 23, 24, 25

input (differentiable) : T1 — Input tensor to be cast.

output (differentiable) : T2 — Output tensor with the same shape as input with type specified by the ‘to’ argument

round_mode : string (default is up) — Rounding mode for conversion to float8e8m0. It only applies to casting to float8e8m0 and is up by default. up: round to nearest value away from zero, down: round to nearest value towards zero, nearest: round to nearest value and ties round up.
saturate : int (default is 1) — The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 conversion (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz, float8e8m0). It is true by default. All cases are fully described in the tables inserted in the operator description.
to : int (required) — The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto

T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2), tensor(float6e2m3), tensor(float6e3m2) — Constrain input types. Casting from complex is not supported.
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2), tensor(float6e2m3), tensor(float6e3m2) — Constrain output types. Casting to complex is not supported.

test_cast_, test_cast_e8m0_, test_cast_no_saturate_

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

25

15, 19, 21, 23, 24

input (differentiable) : T1 — Input tensor to be cast.
target_type (non-differentiable) : T2 — The (first) input tensor will be cast to produce a tensor of the same type as this (second input) tensor.

output (differentiable) : T2 — Output tensor produced by casting the first input tensor to have the same type as the second input tensor.

round_mode : string (default is up) — Rounding mode for conversion to float8e8m0. It only applies to casting to float8e8m0 and is up by default. up: round to nearest value away from zero, down: round to nearest value towards zero, nearest: round to nearest value and ties round up. Please refer to operator Cast description for further details.
saturate : int (default is 1) — The parameter defines how the conversion behaves if an input value is out of range of the destination type. It only applies for float 8 conversion (float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz, float8e8m0). It is true by default. Please refer to operator Cast description for further details.

T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2) — Constrain input types. Casting from complex is not supported.
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2) — Constrain output types. Casting to complex is not supported.

test_castlike_, test_castlike_no_saturate_

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

27

input (differentiable) : T — Input tensor with shape (batch_size, channels, length). Channels-first layout.
weight (differentiable) : T — Depthwise convolution kernel with shape (channels, 1, k) where k is the kernel size. The middle dim of size 1 follows the ONNX Conv weight layout (M, C/group, k1, ..., kn): since this op is always depthwise, group = channels, so C/group = 1. Keeping this layout makes the weight tensor a drop-in for a depthwise Conv(group=channels) weight, so Conv CausalConvWithState rewrites require no reshape.
bias (optional, differentiable) : T — Optional per-channel bias with shape (channels).
past_state (optional, non-differentiable) : T — Carry state from previous step with shape (batch_size, channels, k — 1). If not provided, padding is zero.

output (differentiable) : T — Convolution output with same shape as input.
present_state (non-differentiable) : T — Updated carry state with shape (batch_size, channels, k — 1). Contains the last (k — 1) values of the effective padded/concatenated sequence along the causal axis, including any values from past_state or zero-padding when the current input is shorter than k — 1.

activation : string (default is none) — Fused activation function. One of: ‘silu’, ‘swish’, ‘none’. Default is ‘none’.

T : tensor(float), tensor(float16), tensor(bfloat16) — Constrain input and output types to float tensors.

test_causal_conv_with_state_b1_c1_degenerate, test_causal_conv_with_state_basic, test_causal_conv_with_state_decode_step, test_causal_conv_with_state_fp16, test_causal_conv_with_state_kernel_size_one, test_causal_conv_with_state_short_input_no_past_state, test_causal_conv_with_state_silu, test_causal_conv_with_state_silu_fp16, test_causal_conv_with_state_silu_with_past_state, test_causal_conv_with_state_swish_alias, test_causal_conv_with_state_with_bias, test_causal_conv_with_state_with_bias_and_past_state, test_causal_conv_with_state_with_past_state

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

1, 6

X (non-differentiable) : T — Input tensor

Y (non-differentiable) : T — Output tensor

None.

T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input and output types to float tensors.

test_ceil_example, test_ceil

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

28

12

X (differentiable) : T — Input tensor

Y (differentiable) : T — Output tensor

alpha : float (default is 1.0) — The Alpha value in Celu formula which control the shape of the unit. The default value is 1.0.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_celu, test_celu_bfloat16, test_celu_float16

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

18

input_data (differentiable) : T — Input to extract the centered crop from.
shape (non-differentiable) : Tind — 1-D tensor representing the cropping window dimensions.

output_data (differentiable) : T — Output data.

axes : list of ints — If provided, it specifies a subset of axes that ‘shape’ refer to. If not provided, all axes are assumed [0, 1, …​, r-1], where r = rank(data). Negative value means counting dimensions from the back. Accepted range is [-r, r-1], where r = rank(data). Behavior is undefined if an axis is repeated.

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128) — Constrain input and output types to all tensor types.
Tind : tensor(int32), tensor(int64) — Constrain indices to integer types

test_center_crop_pad_crop, test_center_crop_pad_crop_and_pad, test_center_crop_pad_crop_axes_chw, test_center_crop_pad_crop_axes_hwc, test_center_crop_pad_crop_negative_axes_hwc, test_center_crop_pad_pad

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

1, 6, 11, 12

input (differentiable) : T — Input tensor whose elements to be clipped
min (optional, non-differentiable) : T — Minimum value, under which element is replaced by min. It must be a scalar(tensor of empty shape).
max (optional, non-differentiable) : T — Maximum value, above which element is replaced by max. It must be a scalar(tensor of empty shape).

output (differentiable) : T — Output tensor with clipped input elements

None.

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input and output types to all numeric tensors.

test_clip_example, test_clip, test_clip_inbounds, test_clip_outbounds, test_clip_splitbounds, test_clip_min_greater_than_max, test_clip_default_min, test_clip_default_max, test_clip_default_inbounds, test_clip_default_int8_min, test_clip_default_int8_max, test_clip_default_int8_inbounds

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

18

input (differentiable) : T — Input data tensor to be rearranged from column blocks back into an image. This is a 3-dimensional tensor containing [N, C * n-ary-product(block_shape), L], where N is batch dimension, C is image channel dimension and L is number of blocks.The blocks are enumerated in increasing lexicographic-order of their indices.For example, with an image-size 1020 and block-size 918, there would be 2*3 blocks, enumerated in the order block(0, 0), block(0, 1), block(0, 2), block(1, 0), block(1, 1), block(1, 2).
image_shape (non-differentiable) : tensor(int64) — The shape of the spatial dimensions of the image after rearranging the column blocks.This is a 1-dimensional tensor with size of at least 2, containing the value [H_img, W_img] for a 2-D image or [dim_i1, dim_i2, …​, dim_iN] for a N-D image.
block_shape (non-differentiable) : tensor(int64) — The shape of the block to apply on the input.This is a 1-dimensional tensor of size of at least 2, containing the value [H_block, W_block] for a 2-D image or [dim_b1, dim_b2, …​, dim_bN] for a N-D block.This is the block-shape before dilation is applied to it.

output (differentiable) : T — Output tensor produced by rearranging blocks into an image.

dilations : list of ints — 1-dimensional tensor with dilation value along each spatial axis of the image. If not present, the dilation defaults to 1 along each spatial axis of the image.
pads : list of ints — 1-dimensional tensor with padding value for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…​x1_end, x2_end,…​], where xi_begin is the number of pixels added at the beginning of axis i and xi_end is the number of pixels added at the end of axis i. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides : list of ints — 1-dimensional tensor with stride value along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128) — Constrain input and output types to all numeric tensor types.

test_col2im, test_col2im_5d, test_col2im_dilations, test_col2im_pads, test_col2im_strides

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

28

9, 11

input (differentiable) : T — Tensor of rank r >= 1.
condition (non-differentiable) : T1 — Rank 1 tensor of booleans to indicate which slices or data elements to be selected. Its length can be less than the input length along the axis or the flattened input size if axis is not specified. In such cases data slices or elements exceeding the condition length are discarded.

output (differentiable) : T — Tensor of rank r if axis is specified. Otherwise output is a Tensor of rank 1.

axis : int — (Optional) Axis along which to take slices. If not specified, input is flattened before elements being selected. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128) — Constrain input and output types to all tensor types.
T1 : tensor(bool) — Constrain to boolean tensors.

test_compress_0, test_compress_1, test_compress_bfloat16, test_compress_default_axis, test_compress_negative_axis

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

1, 4, 11

inputs (variadic, differentiable) : T — List of tensors for concatenation

concat_result (differentiable) : T — Concatenated tensor

axis : int (required) — Which axis to concat on. A negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(inputs)..

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128) — Constrain output types to any tensor type.

test_concat_

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

11

input_sequence : S — Sequence of tensors for concatenation

concat_result : T — Concatenated tensor

axis : int (required) — Which axis to concat on. Accepted range in [-r, r - 1], where r is the rank of input tensors. When new_axis is 1, accepted range is [-r - 1, r].
new_axis : int (default is 0) — Insert and concatenate on a new axis or not, default 0 means do not insert new axis.

S : seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)) — Constrain input types to any tensor type.
T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128) — Constrain output types to any tensor type.

None published for this operator.

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

25

1, 9, 11, 12, 13, 19, 21, 23, 24

None.

output : T — Output tensor containing the same value of the provided tensor.

sparse_value : sparse_tensor — The value for the elements of the output tensor in sparse format.
value : tensor — The value for the elements of the output tensor.
value_float : float — The value for the sole element for the scalar, float32, output tensor.
value_floats : list of floats — The values for the elements for the 1D, float32, output tensor.
value_int : int — The value for the sole element for the scalar, int64, output tensor.
value_ints : list of ints — The values for the elements for the 1D, int64, output tensor.
value_string : string — The value for the sole element for the scalar, UTF-8 string, output tensor.
value_strings : list of strings — The values for the elements for the 1D, UTF-8 string, output tensor.

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2) — Constrain input and output types to all tensor types.

test_constant

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

25

9, 20, 21, 23, 24

input : T1 — 1D tensor. The shape of the expected output tensor. If empty tensor is given, the output would be a scalar. All values must be >= 0.

output : T2 — Output tensor of shape specified by ‘input’.If attribute ‘value’ is specified, the value and datatype of the output tensor is taken from ‘value’.If attribute ‘value’ is not specified, the value in the output defaults to 0, and the datatype defaults to float32.

value : tensor — (Optional) The value of the output elements.Should be a one-element tensor. If not specified, it defaults to a tensor of value 0 and datatype float32

T1 : tensor(int64) — Constrain input types.
T2 : tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint4), tensor(int4), tensor(bool), tensor(bfloat16), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2) — Constrain output types to be numerics or boolean.

test_constantofshape_float_ones, test_constantofshape_int_shape_zero, test_constantofshape_int_zeros

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

1, 11

X (differentiable) : T — Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 …​ x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …​].
W (differentiable) : T — The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x …​ x kn), where (k1 x k2 x …​ kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL …​]. Assuming zero based indices for the shape array, X.shape[1] == (W.shape[1] * group) == C and W.shape[0] mod G == 0. Or in other words FILTER_IN_CHANNEL multiplied by the number of groups should be equal to DATA_CHANNEL and the number of feature maps M should be a multiple of the number of groups G.
B (optional, differentiable) : T — Optional 1D bias to be added to the convolution, has size of M.

Y (differentiable) : T — Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.

auto_pad : string (default is NOTSET) — auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
dilations : list of ints — dilation value along each spatial axis of the filter. If not present, the dilation defaults is 1 along each spatial axis.
group : int (default is 1) — number of groups input channels and output channels are divided into.
kernel_shape : list of ints — The shape of the convolution kernel. If not present, should be inferred from input W.
pads : list of ints — Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…​x1_end, x2_end,…​], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides : list of ints — Stride along each spatial axis. If not present, the stride defaults is 1 along each spatial axis.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_basic_conv_with_padding, test_basic_conv_without_padding, test_conv_with_autopad_same, test_conv_with_strides_padding, test_conv_with_strides_no_padding, test_conv_with_strides_and_asymmetric_padding

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

10

x : T1 — Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 …​ x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …​].
w : T2 — The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x …​ x kn), where (k1 x k2 x …​ kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL …​]. X.shape[1] == (W.shape[1] * group) == C (assuming zero based indices for the shape array). Or in other words FILTER_IN_CHANNEL should be equal to DATA_CHANNEL.
x_zero_point (optional) : T1 — Zero point tensor for input ‘x’. It’s optional and default value is 0. It’s a scalar, which means a per-tensor/layer quantization.
w_zero_point (optional) : T2 — Zero point tensor for input ‘w’. It’s optional and default value is 0. It could be a scalar or a 1-D tensor, which means a per-tensor/layer or per output channel quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of output channels (M)

y : T3 — Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.

auto_pad : string (default is NOTSET) — auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
dilations : list of ints — dilation value along each spatial axis of the filter. If not present, the dilation defaults to 1 along each axis.
group : int (default is 1) — number of groups input channels and output channels are divided into. default is 1.
kernel_shape : list of ints — The shape of the convolution kernel. If not present, should be inferred from input ‘w’.
pads : list of ints — Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0.The value represent the number of pixels added to the beginning and end part of the corresponding axis.pads format should be as follow [x1_begin, x2_begin…​x1_end, x2_end,…​], where xi_begin the number ofpixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i.This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaultsto 0 along start and end of each spatial axis.
strides : list of ints — Stride along each spatial axis. If not present, the stride defaults to 1 along each axis.

T1 : tensor(int8), tensor(uint8) — Constrain input x and its zero point data type to 8-bit integer tensor.
T2 : tensor(int8), tensor(uint8) — Constrain input w and its zero point data type to 8-bit integer tensor.
T3 : tensor(int32) — Constrain output y data type to 32-bit integer tensor.

test_convinteger_with_padding, test_convinteger_without_padding

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

1, 11

X (differentiable) : T — Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 …​ x Dn)
W (differentiable) : T — The weight tensor that will be used in the convolutions; has size (C x M/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the weight shape will be (C x M/group x k1 x k2 x …​ x kn), where (k1 x k2 x …​ x kn) is the dimension of the kernel. The number of channels in the output should be equal to W.shape[1] * group (assuming zero based indices of the shape array)
B (optional, differentiable) : T — Optional 1D bias to be added to the convolution, has size of M.

Y (differentiable) : T — Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, pad lengths and group count. The number of channels in the output should be equal to W.shape[1] * group (assuming zero based indices of the shape array)

auto_pad : string (default is NOTSET) — auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = input_shape[i] * strides[i] for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
dilations : list of ints — dilation value along each spatial axis of the filter. If not present, the dilation defaults to 1 along each spatial axis.
group : int (default is 1) — number of groups input channels and output channels are divided into.
kernel_shape : list of ints — The shape of the convolution kernel. If not present, should be inferred from input W.
output_padding : list of ints — Additional elements added to the side with higher coordinate indices in the output. Each padding value in “output_padding” must be less than the corresponding stride/dilation dimension. By default, this attribute is a zero vector. Note that this attribute doesn’t directly affect the computed output values. It only controls the selection of the computed values, so changing this attribute only adds or removes output elements. If “output_shape” is explicitly provided, “output_padding” does not contribute additional size to “output_shape” but participates in the computation of the needed padding amount. This is also called adjs or adjustment in some frameworks.
output_shape : list of ints — The shape of the output can be explicitly set which will cause pads values to be auto generated. If output_shape is specified pads values are ignored. See doc for details for equations to generate pads. Note that the output_shape attribute value should not include dimensions for batch size and channels, which are automatically inferred.
pads : list of ints — Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…​x1_end, x2_end,…​], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides : list of ints — Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_convtranspose, test_convtranspose_1d, test_convtranspose_3d, test_convtranspose_output_shape, test_convtranspose_pad, test_convtranspose_kernel_shape, test_convtranspose_autopad_same, test_convtranspose_dilations, test_convtranspose_group_2, test_convtranspose_group_2_image_3, test_convtranspose_pads

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

7

input (differentiable) : T — Input tensor

output (differentiable) : T — The cosine of the input tensor computed element-wise

None.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_cos_example, test_cos

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

9

input (differentiable) : T — Input tensor

output (differentiable) : T — The hyperbolic cosine values of the input tensor computed element-wise

None.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_cosh_example, test_cosh

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

26

x (differentiable) : T — An input tensor that is to be processed.
axis (non-differentiable) : T2 — A 0-D tensor. Must be in the range [-rank(x), rank(x)-1]. Negative value means counting dimensions from the back.

y (differentiable) : T — Output tensor of the same type as ‘x’ with cumulative products of the x’s elements

exclusive : int (default is 0) — If set to 1 will return exclusive product in which the top element is not included. In other terms, if set to 1, the j-th output element would be the product of the first (j-1) elements. Otherwise, it would be the product of the first j elements.
reverse : int (default is 0) — If set to 1 will perform the products in reverse direction.

T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input and output types to numeric tensors.
T2 : tensor(int32), tensor(int64) — axis tensor can be int32 or int64 only

test_cumprod_1d, test_cumprod_1d_exclusive, test_cumprod_1d_int32_exclusive, test_cumprod_1d_reverse, test_cumprod_1d_reverse_exclusive, test_cumprod_2d_axis_0, test_cumprod_2d_axis_1, test_cumprod_2d_int32, test_cumprod_2d_negative_axis

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

14

11

x (differentiable) : T — An input tensor that is to be processed.
axis (non-differentiable) : T2 — A 0-D tensor. Must be in the range [-rank(x), rank(x)-1]. Negative value means counting dimensions from the back.

y (differentiable) : T — Output tensor of the same type as ‘x’ with cumulative sums of the x’s elements

exclusive : int (default is 0) — If set to 1 will return exclusive sum in which the top element is not included. In other terms, if set to 1, the j-th output element would be the sum of the first (j-1) elements. Otherwise, it would be the sum of the first j elements.
reverse : int (default is 0) — If set to 1 will perform the sums in reverse direction.

T : tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input and output types to numeric tensors.
T2 : tensor(int32), tensor(int64) — axis tensor can be int32 or int64 only

test_cumsum_1d, test_cumsum_1d_exclusive, test_cumsum_1d_int32_exclusive, test_cumsum_1d_reverse, test_cumsum_1d_reverse_exclusive, test_cumsum_2d_axis_0, test_cumsum_2d_axis_1, test_cumsum_2d_int32, test_cumsum_2d_negative_axis

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

20

17

input (non-differentiable) : T1 — For real input, the following shape is expected: [signal_dim0][signal_dim1][signal_dim2]...[signal_dimN][1]. For complex input, the following shape is expected: [signal_dim0][signal_dim1][signal_dim2]...[signal_dimN][2]. The final dimension represents the real and imaginary parts of the value in that order.
dft_length (optional, non-differentiable) : T2 — The length of the signal as a scalar. If greater than the axis dimension, the signal will be zero-padded up to dft_length. If less than the axis dimension, only the first dft_length values will be used as the signal. If not provided, the default dft_length = signal_dim_axis, except for the IRFFT case (onesided=1, inverse=1), in which case the default dft_length is 2 * (signal_dim_axis - 1).
axis (optional, non-differentiable) : tensor(int64) — The axis as a scalar on which to perform the DFT. Default is -2 (last signal axis). Negative value means counting dimensions from the back. Accepted range is $[-r, -2] \cup [0, r-2]$ where r = rank(input). The last dimension is for representing complex numbers and thus is an invalid axis.

output : T1 — The Fourier Transform of the input vector. For standard DFT (onesided=0), the output shape is: [signal_dim0][signal_dim1][signal_dim2]...[signal_dimN][2] (complex), with signal_dim_axis = dft_length. For RFFT (onesided=1, inverse=0), the output shape is: [signal_dim0][signal_dim1][signal_dim2]...[signal_dimN][2] (one-sided complex), with signal_dim_axis = floor(dft_length/2) + 1. For IRFFT (onesided=1, inverse=1), the output shape is: [signal_dim0][signal_dim1][signal_dim2]...[signal_dimN][1] (real), where signal_dim_axis = dft_length.

inverse : int (default is 0) — Whether to perform the inverse discrete Fourier Transform. Default is 0, which corresponds to false.
onesided : int (default is 0) — If onesided is 1, only values for k in [0, 1, 2, ..., floor(n_fft/2) + 1] are used or returned because the real-to-complex Fourier transform satisfies the conjugate symmetry, i.e., X[m, k] = X[m, n_fft-k]*, where m denotes “all other dimensions” DFT was not applied on. When onesided=1 and inverse=0 (forward DFT), only real input is supported and a one-sided complex spectrum is returned (RFFT). When onesided=1 and inverse=1 (inverse DFT), only complex input is supported and a full real signal is returned (IRFFT). Value can be 0 or 1. Default is 0.

T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.
T2 : tensor(int32), tensor(int64) — Constrain scalar length types to integers.

test_dft, test_dft_axis, test_dft_inverse, test_dft_rfft, test_dft_irfft, test_dft_opset19, test_dft_axis_opset19, test_dft_inverse_opset19, test_dft_rfft_opset19, test_dft_irfft_opset19

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

19

X : T — Input data tensor. For 2D image data, it has shape (N, C, H, W) where N is the batch size, C is the number of input channels, and H and W are the height and width. In general, the shape is (N, C, D1, D2, …​ , Dn) for n-dimensional data, where D1 to Dn are the spatial dimension sizes. Most common use cases have n = 2 or 3.
W : T — Weight tensor that will be used in the convolutions. It has shape (oC, C/group, kH, kW), where oC is the number of output channels and kH and kW are the kernel height and width. For more than 2 dimensions, it has shape (oC, C/group, k1, k2, …​ , kn).
offset : T — Offset tensor denoting the offset for the sampling locations in the convolution kernel. It has shape (N, offset_group * kH * kW * 2, oH, oW) for 2D data or (N, offset_group * k1 * k2 * …​ * kn * n, o1, o2, …​ , on) for nD data. Use linear interpolationfor fractional offset values. Sampling locations outside of the padded input tensor gives zero.
B (optional) : T — Optional 1D bias of length oC to be added to the convolution. Default is a tensor of zeros.
mask (optional) : T — The mask tensor to be applied to each position in the convolution kernel. It has shape (N, offset_group * kH * kW, oH, oW) for 2D data or (N, offset_group * k1 * k2 * …​ * kn * n, o1, o2, …​ , on) for nD data. Default is a tensor of ones.

Y : T — Output data tensor that contains the result of convolution. It has shape (N, oC, oH, oW) for 2D data or (N, oC, o1, o2, …​, on) for nD data

dilations : list of ints — Dilation value along each spatial axis of the kernel. Default is 1 along each axis.
group : int (default is 1) — Number of groups the input and output channels, C and oC, are divided into. C and oC must both be divisible by group. Default is 1.
kernel_shape : list of ints — Shape of the convolution kernel. If not present, it is inferred from the shape of input W.
offset_group : int (default is 1) — Number of groups of offset. C must be divisible by offset_group. Default is 1.
pads : list of ints — Padding for the beginning and end along each spatial axis. The values represent the number of pixels added to the beginning and end of the corresponding axis and can take any nonnegative value. The format should be as follows: [x1_begin, x2_begin, …​, x1_end, x2_end, …​], where xi_begin is the number of pixels added at the beginning of axis i and xi_end is the number of pixels added at the end of axis i. Default is 0 along each axis.
strides : list of ints — Stride along each spatial axis. Default is 1 along each axis.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_basic_deform_conv_with_padding, test_basic_deform_conv_without_padding, test_deform_conv_with_mask_bias, test_deform_conv_with_multiple_offset_groups

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

28

1, 11, 13

input (differentiable) : T — Input tensor of [N,C,H,W], where N is the batch axis, C is the channel or depth, H is the height and W is the width.

output (differentiable) : T — Output tensor of [N, C/(blocksize * blocksize), H * blocksize, W * blocksize].

blocksize : int (required) — Blocks of [blocksize, blocksize] are moved.
mode : string (default is DCR) — DCR (default) for depth-column-row order re-arrangement. Use CRD for column-row-depth order.

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128) — Constrain input and output types to all tensor types.

test_depthtospace_crd_mode_example, test_depthtospace_example

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

28

10, 13, 19, 21, 23, 24, 25

x : T1 — N-D quantized input tensor to be de-quantized.
x_scale : T2 — Scale for input x. For per-tensor/layer dequantization the scale is a scalar, for per per-axis dequantization it is a 1-D Tensor and for blocked dequantization it has the same shape as the input, except for one dimension in which blocking is performed.
x_zero_point (optional) : T1 — Zero point for input x. Shape must match x_scale. It’s optional. Zero point is 0 when it’s not specified.

y : T3 — N-D full precision output tensor. It has the same shape as input x. The data type is specified by the output_dtype attribute or, in its absence, the type of x_scale.

axis : int (default is 1) — (Optional) The axis of the dequantizing dimension of the input tensor. Used for per-axis and blocked quantization. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
block_size : int (default is 0) — (Optional) The size of the quantization block (number of times every scale is replicated). Used only for blocked quantization. The block size is a positive integer. Given x shape (D0, ..., Di, ..., Dn), y_scale shape (S0, ... Si, ...Sn) and axis=i, the accepted range is [ceil(Di/Si), ceil(Di/(Si-1))-1]
output_dtype : int (default is 0) — (Optional) The output data type. If not supplied, the output data type is inferred from x_scale data type (T2)

T1 : tensor(int8), tensor(uint8), tensor(int16), tensor(uint16), tensor(int32), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(uint2), tensor(int2), tensor(float6e2m3), tensor(float6e3m2) — The type of the inputs ‘x_zero_point’ and ‘x’.
T2 : tensor(float), tensor(float16), tensor(bfloat16), tensor(float8e8m0) — The type of the input ‘x_scale’.
T3 : tensor(float), tensor(float16), tensor(bfloat16) — The type of the output ‘y’.

test_dequantizelinear_axis, test_dequantizelinear_blocked, test_dequantizelinear, test_dequantizelinear_e4m3fn, test_dequantizelinear_e4m3fn_float16, test_dequantizelinear_e4m3fn_zero_point, test_dequantizelinear_e5m2, test_dequantizelinear_float4e2m1, test_dequantizelinear_int16, test_dequantizelinear_int2, test_dequantizelinear_int4, test_dequantizelinear_uint16, test_dequantizelinear_uint2, test_dequantizelinear_uint4

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

11

X (differentiable) : T — Input tensor

Y (differentiable) : T — Output tensor

None.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to floating-point tensors.

test_det_2d, test_det_nd

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

14

1, 6, 7, 13

A (differentiable) : T — First operand.
B (differentiable) : T — Second operand.

C (differentiable) : T — Result, has same element type as two inputs

None.

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input and output types to all numeric tensors.

test_div_example, test_div, test_div_int8, test_div_int16, test_div_int32_trunc, test_div_uint8, test_div_uint16, test_div_uint32, test_div_uint64, test_div_bcast

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

1, 6, 7, 10, 12, 13

data (differentiable) : T — The input data as Tensor.
ratio (optional, non-differentiable) : T1 — The ratio of random dropout, with value in [0, 1). If set to 0, the output would be a simple copy of the input. If it’s non-zero, output will be a random dropout of the scaled input, which is typically the case during training. It is an optional value, if not specified it will default to 0.5.
training_mode (optional, non-differentiable) : T2 — If set to true then it indicates dropout is being used for training. It is an optional value hence unless specified explicitly, it is false. If it is false, ratio is ignored and the operation mimics inference mode where nothing will be dropped from the input data and if mask is requested as output it will contain all ones.

output (differentiable) : T — The output.
mask (optional, non-differentiable) : T2 — The output mask.

seed : int — (Optional) Seed to the random generator, if not specified we will auto generate one.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz) — Constrain input and output types to float tensors.
T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz) — Constrain input ‘ratio’ types to float tensors.
T2 : tensor(bool) — Constrain output ‘mask’ types to boolean tensors.

test_dropout_default, test_dropout_default_mask, test_dropout_default_mask_ratio, test_dropout_default_old, test_dropout_default_ratio, test_dropout_random_old, test_training_dropout, test_training_dropout_default, test_training_dropout_default_mask, test_training_dropout_zero_ratio, test_training_dropout_zero_ratio_mask, test_training_dropout_mask

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

28

12

Inputs (variadic, differentiable) : T — Operands

Output (differentiable) : T — Output tensor

equation : string (required) — Einsum expression string.

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input and output types to all numerical tensor types.

test_einsum_batch_diagonal, test_einsum_batch_matmul, test_einsum_batch_matmul_bfloat16, test_einsum_inner_prod, test_einsum_scalar, test_einsum_sum, test_einsum_sum_bfloat16, test_einsum_transpose, test_einsum_transpose_bfloat16

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

1, 6

X (differentiable) : T — Input tensor

Y (differentiable) : T — Output tensor

alpha : float (default is 1.0) — Coefficient of ELU.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_elu_example, test_elu, test_elu_default

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

19

1, 7, 11, 13

A (non-differentiable) : T — First input operand for the logical operator.
B (non-differentiable) : T — Second input operand for the logical operator.

C (non-differentiable) : T1 — Result tensor.

None.

T : tensor(bool), tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16), tensor(string) — Constrain input types to all (non-complex) tensors.
T1 : tensor(bool) — Constrain output to boolean tensor.

test_equal, test_equal_int8, test_equal_int16, test_equal_uint8, test_equal_uint16, test_equal_uint32, test_equal_uint64, test_equal_bcast, test_equal_string, test_equal_string_broadcast

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

9

input (differentiable) : T — Input tensor

output (differentiable) : T — The error function of the input tensor computed element-wise. It has the same shape and type of the input.

None.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_erf

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

1, 6

input (differentiable) : T — Input tensor

output (differentiable) : T — The exponential of the input tensor computed element-wise

None.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_exp_example, test_exp

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

8

input (differentiable) : T — Input tensor
shape (non-differentiable) : tensor(int64) — A 1-D tensor indicates the shape you want to expand to, following the broadcast rule

output (differentiable) : T — Output tensor

None.

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128) — Constrain input and output types to all tensors.

test_expand_dim_changed, test_expand_dim_unchanged

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

9

input : T1 — 2D input tensor to copy shape, and optionally, type information from.

output : T2 — Output tensor, same shape as input tensor T1.

dtype : int — (Optional) The data type for the elements of the output tensor. If not specified, the data type of the input tensor T1 is used.
k : int (default is 0) — (Optional) Index of the diagonal to be populated with ones. Default is 0. If T2 is the output, this op sets T2[i, i+k] = 1. k = 0 populates the main diagonal, k > 0 populates an upper diagonal, and k < 0 populates a lower diagonal.

T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(bool) — Constrain input types. Strings and complex are not supported.
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(bool) — Constrain output types. Strings and complex are not supported.

test_eyelike_populate_off_main_diagonal, test_eyelike_with_dtype, test_eyelike_without_dtype

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

25

1, 9, 11, 13, 21, 23, 24

input (differentiable) : T — A tensor of rank >= axis.

output (differentiable) : T — A 2D tensor with the contents of the input tensor, with input dimensions up to axis flattened to the outer dimension of the output and remaining input dimensions flattened into the inner dimension of the output.

axis : int (default is 1) — Indicate up to which input dimensions (exclusive) should be flattened to the outer dimension of the output. The value for axis must be in the range [-r, r], where r is the rank of the input tensor. Negative value means counting dimensions from the back. When axis = 0, the shape of the output tensor is (1, (d_0 X d_1 …​ d_n), where the shape of the input tensor is (d_0, d_1, …​ d_n).

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2) — Constrain input and output to all tensor types up to IRv13.

test_flatten_axis, test_flatten_negative_axis, test_flatten_default_axis

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

1, 6

X (non-differentiable) : T — Input tensor

Y (non-differentiable) : T — Output tensor

None.

T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input and output types to float tensors.

test_floor_example, test_floor

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

1, 11

data (differentiable) : T — Tensor of rank r >= 1.
indices (non-differentiable) : Tind — Tensor of int32/int64 indices, of any rank q. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.

output (differentiable) : T — Tensor of rank q + (r — 1).

axis : int (default is 0) — Which axis to gather on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128) — Constrain input and output types to any tensor type.
Tind : tensor(int32), tensor(int64) — Constrain indices to integer types

test_gather_0, test_gather_1, test_gather_2d_indices, test_gather_negative_indices

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

11

data (differentiable) : T — Tensor of rank r >= 1.
indices (non-differentiable) : Tind — Tensor of int32/int64 indices, with the same rank r as the input. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.

output (differentiable) : T — Tensor of the same shape as indices.

axis : int (default is 0) — Which axis to gather on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128) — Constrain input and output types to any tensor type.
Tind : tensor(int32), tensor(int64) — Constrain indices to integer types

test_gather_elements_0, test_gather_elements_1, test_gather_elements_negative_indices

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

11, 12

data (differentiable) : T — Tensor of rank r >= 1.
indices (non-differentiable) : tensor(int64) — Tensor of rank q >= 1. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.

output (differentiable) : T — Tensor of rank q + r — indices_shape[-1] — 1.

batch_dims : int (default is 0) — The number of batch dimensions. The gather of indexing starts from dimension of data[batch_dims:]

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128) — Constrain input and output types to any tensor type.

test_gathernd_example_float32, test_gathernd_example_int32, test_gathernd_example_int32_batch_dim1

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

20

X (differentiable) : T — Input tensor

Y (differentiable) : T — Output tensor

approximate : string (default is none) — Gelu approximation algorithm: "tanh", "none"(default)."none": do not use approximation."tanh": use tanh approximation.

T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input and output types to float tensors.

test_gelu_default_1, test_gelu_default_2, test_gelu_tanh_1, test_gelu_tanh_2

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

1, 6, 7, 9, 11

A (differentiable) : T — Input tensor A. The shape of A should be (M, K) if transA is 0, or (K, M) if transA is non-zero.
B (differentiable) : T — Input tensor B. The shape of B should be (K, N) if transB is 0, or (N, K) if transB is non-zero.
C (optional, differentiable) : T — Optional input tensor C. If not specified, the computation is done as if C is a scalar 0. The shape of C should be unidirectional broadcastable to (M, N).

Y (differentiable) : T — Output tensor of shape (M, N).

alpha : float (default is 1.0) — Scalar multiplier for the product of input tensors A * B.
beta : float (default is 1.0) — Scalar multiplier for input tensor C.
transA : int (default is 0) — Whether A should be transposed
transB : int (default is 0) — Whether B should be transposed

T : tensor(float16), tensor(float), tensor(double), tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(bfloat16) — Constrain input and output types to float/int tensors.

test_gemm_all_attributes, test_gemm_alpha, test_gemm_beta, test_gemm_default_matrix_bias, test_gemm_default_no_bias, test_gemm_default_scalar_bias, test_gemm_default_single_elem_vector_bias, test_gemm_default_vector_bias, test_gemm_default_zero_bias, test_gemm_transposeA, test_gemm_transposeB

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

1

X (differentiable) : T — Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 …​ Dn), where N is the batch size.

Y (differentiable) : T — Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. The first two dimensions of output shape are the same as the input (N x C), while the other dimensions are all 1.

None.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_globalaveragepool, test_globalaveragepool_precomputed

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

1, 2

X (differentiable) : T — Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 …​ Dn), where N is the batch size.

Y (differentiable) : T — Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. The first two dimensions of output shape are the same as the input (N x C), while the other dimensions are all 1.

p : int (default is 2) — p value of the Lp norm used to pool over the input data.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

None published for this operator.

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

1

X (differentiable) : T — Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 …​ Dn), where N is the batch size.

Y (differentiable) : T — Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. The first two dimensions of output shape are the same as the input (N x C), while the other dimensions are all 1.

None.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_globalmaxpool, test_globalmaxpool_precomputed

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

1, 7, 9

A (non-differentiable) : T — First input operand for the logical operator.
B (non-differentiable) : T — Second input operand for the logical operator.

C (non-differentiable) : T1 — Result tensor.

None.

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input types to all numeric tensors.
T1 : tensor(bool) — Constrain output to boolean tensor.

test_greater, test_greater_int8, test_greater_int16, test_greater_uint8, test_greater_uint16, test_greater_uint32, test_greater_uint64, test_greater_bcast

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

16

12

A (non-differentiable) : T — First input operand for the logical operator.
B (non-differentiable) : T — Second input operand for the logical operator.

C (non-differentiable) : T1 — Result tensor.

None.

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input types to all numeric tensors.
T1 : tensor(bool) — Constrain output to boolean tensor.

test_greater_equal_bcast, test_greater_equal, test_greater_equal_int8, test_greater_equal_int16, test_greater_equal_uint8, test_greater_equal_uint16, test_greater_equal_uint32, test_greater_equal_uint64

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

16, 20

X (differentiable) : T1 — Input tensor of rank r+2 that has shape (N, C, D1, D2, …​, Dr), where N is the batch size, C is the number of channels, D1, D2, …​, Dr are the spatial dimensions.
grid (non-differentiable) : T2 — Input offset of shape (N, D1_out, D2_out, …​, Dr_out, r), where D1_out, D2_out, …​, Dr_out are the spatial dimensions of the grid and output, and r is the number of spatial dimensions. Grid specifies the sampling locations normalized by the input spatial dimensions. Therefore, it should have most values in the range of [-1, 1]. If the grid has values outside the range of [-1, 1], the corresponding outputs will be handled as defined by padding_mode. Following computer vision convention, the coordinates in the length-r location vector are listed from the innermost tensor dimension to the outermost, the opposite of regular tensor indexing.

Y (differentiable) : T1 — Output tensor of rank r+2 that has shape (N, C, D1_out, D2_out, …​, Dr_out) of the sampled values. For integer input types, intermediate values are computed as floating point and cast to integer at the end.

align_corners : int (default is 0) — If align_corners=1, the extrema (-1 and 1) are considered as referring to the center points of the input’s corner pixels (voxels, etc.). If align_corners=0, they are instead considered as referring to the corner points of the input’s corner pixels (voxels, etc.), making the sampling more resolution agnostic.
mode : string (default is linear) — Three interpolation modes: linear (default), nearest and cubic. The “linear” mode includes linear and N-linear interpolation modes depending on the number of spatial dimensions of the input tensor (i.e. linear for 1 spatial dimension, bilinear for 2 spatial dimensions, etc.). The “cubic” mode also includes N-cubic interpolation modes following the same rules. The “nearest” mode rounds to the nearest even index when the sampling point falls halfway between two indices.
padding_mode : string (default is zeros) — Support padding modes for outside grid values: zeros(default), border, reflection. zeros: use 0 for out-of-bound grid locations, border: use border values for out-of-bound grid locations, reflection: use values at locations reflected by the border for out-of-bound grid locations. If index 0 represents the margin pixel, the reflected value at index -1 will be the same as the value at index 1. For location far away from the border, it will keep being reflected until becoming in bound. If pixel location x = -3.5 reflects by border -1 and becomes x’ = 1.5, then reflects by border 1 and becomes x’’ = 0.5.

T1 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128) — Constrain input X and output Y types to all tensor types.
T2 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain grid types to float tensors.

test_gridsample, test_gridsample_bilinear, test_gridsample_aligncorners_true, test_gridsample_nearest, test_gridsample_bicubic, test_gridsample_nearest_align_corners_0_additional_1, test_gridsample_nearest_align_corners_1_additional_1, test_gridsample_bilinear_align_corners_0_additional_1, test_gridsample_bilinear_align_corners_1_additional_1, test_gridsample_bicubic_align_corners_0_additional_1, test_gridsample_bicubic_align_corners_1_additional_1, test_gridsample_zeros_padding, test_gridsample_border_padding, test_gridsample_reflection_padding, test_gridsample_volumetric_nearest_align_corners_0, test_gridsample_volumetric_nearest_align_corners_1, test_gridsample_volumetric_bilinear_align_corners_0, test_gridsample_volumetric_bilinear_align_corners_1

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

21

18

X (differentiable) : T — Input data tensor. Dimensions for image cases are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width of the data. Statistics are computed for every group of channels over C, H, and W. For non-image cases, the dimensions are in the form of (N x C x D1 x D2 ... Dn).
scale (differentiable) : T — Scale tensor of shape (C).
bias (differentiable) : T — Bias tensor of shape (C).

Y (differentiable) : T — The output tensor of the same shape as X.

epsilon : float (default is 1e-05) — The epsilon value to use to avoid division by zero.
num_groups : int (required) — The number of groups of channels. It should be a divisor of the number of channels C.
stash_type : int (default is 1) — The floating-point precision used in stage one of the computation.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_group_normalization_epsilon, test_group_normalization_example

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

17

size (non-differentiable) : T1 — A scalar value indicating the length of the window.

output (non-differentiable) : T2 — A Hamming window with length: size. The output has the shape: [size].

output_datatype : int (default is 1) — The data type of the output tensor. Strictly must be one of the values from DataType enum in TensorProto whose values correspond to T2. The default value is 1 = FLOAT.
periodic : int (default is 1) — If 1, returns a window to be used as periodic function. If 0, return a symmetric window. When ‘periodic’ is specified, hann computes a window of length size + 1 and returns the first size points. The default value is 1.

T1 : tensor(int32), tensor(int64) — Constrain the input size to int32_t or int64_t.
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain output types to numeric tensors.

test_hammingwindow, test_hammingwindow_symmetric

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

17

size (non-differentiable) : T1 — A scalar value indicating the length of the window.

output (non-differentiable) : T2 — A Hann window with length: size. The output has the shape: [size].

output_datatype : int (default is 1) — The data type of the output tensor. Strictly must be one of the values from DataType enum in TensorProto whose values correspond to T2. The default value is 1 = FLOAT.
periodic : int (default is 1) — If 1, returns a window to be used as periodic function. If 0, return a symmetric window. When ‘periodic’ is specified, hann computes a window of length size + 1 and returns the first size points. The default value is 1.

T1 : tensor(int32), tensor(int64) — Constrain the input size to int32_t or int64_t.
T2 : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain output types to numeric tensors.

test_hannwindow, test_hannwindow_symmetric

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

1, 6

X (differentiable) : T — Input tensor

Y (differentiable) : T — Output tensor

alpha : float (default is 0.2) — Value of alpha.
beta : float (default is 0.5) — Value of beta.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_hardsigmoid_example, test_hardsigmoid, test_hardsigmoid_default

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

14

X (differentiable) : T — Input tensor

Y (differentiable) : T — Output tensor

None.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_hardswish

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

1, 11

input (differentiable) : T — The input tensor of rank >= axis.

output (differentiable) : T — The output values with the same shape as the input tensor.

axis : int (default is -1) — Describes the dimension Hardmax will be performed on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).

T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input and output types to float tensors.

test_hardmax_example, test_hardmax_one_hot, test_hardmax_axis_0, test_hardmax_axis_1, test_hardmax_axis_2, test_hardmax_negative_axis, test_hardmax_default_axis

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

25

1, 13, 14, 16, 19, 21, 23, 24

input (differentiable) : V — Input tensor

output (differentiable) : V — Tensor to copy input into.

None.

V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128)) — Constrain input and output types to all tensor, sequence, and optional types.

test_identity, test_identity_opt, test_identity_sequence

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

25

1, 11, 13, 16, 19, 21, 23, 24

cond : B — Condition for the if. The tensor must contain a single element.

outputs (variadic, heterogeneous) : V — Values that are live-out to the enclosing scope. The return values in the then_branch and else_branch must be of the same data type. The then_branch and else_branch may produce tensors with the same element type and different shapes. If corresponding outputs from the then-branch and the else-branch have static shapes S1 and S2, then the shape of the corresponding output variable of the if-node (if present) must be compatible with both S1 and S2 as it represents the union of both possible shapes.For example, if in a model file, the first output of then_branch is typed float tensor with shape [2] and the first output of else_branch is another float tensor with shape [3], If’s first output should have (a) no shape set, or (b) a shape of rank 1 with neither dim_value nor dim_param set, or © a shape of rank 1 with a unique dim_param. In contrast, the first output cannot have the shape [2] since [2] and [3] are not compatible.

else_branch : graph (required) — Graph to run if condition is false. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the then_branch.
then_branch : graph (required) — Graph to run if condition is true. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the else_branch.

V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(bfloat16)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), seq(tensor(float8e4m3fn)), seq(tensor(float8e4m3fnuz)), seq(tensor(float8e5m2)), seq(tensor(float8e5m2fnuz)), seq(tensor(uint4)), seq(tensor(int4)), seq(tensor(float4e2m1)), seq(tensor(float8e8m0)), seq(tensor(uint2)), seq(tensor(int2)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(bfloat16))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(bfloat16)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128)), optional(tensor(float8e4m3fn)), optional(tensor(float8e4m3fnuz)), optional(tensor(float8e5m2)), optional(tensor(float8e5m2fnuz)), optional(tensor(uint4)), optional(tensor(int4)), optional(tensor(float4e2m1)), optional(tensor(float8e8m0)), optional(tensor(uint2)), optional(tensor(int2)) — All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types up to IRv13.
B : tensor(bool) — Only bool

test_if, test_if_opt, test_if_seq

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

1, 6

input (differentiable) : T — Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 …​ Dn), where N is the batch size.
scale (differentiable) : T — The input 1-dimensional scale tensor of size C.
B (differentiable) : T — The input 1-dimensional bias tensor of size C.

output (differentiable) : T — The output tensor of the same shape as input.

epsilon : float (default is 1e-05) — The epsilon value to use to avoid division by zero.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_instancenorm_example, test_instancenorm_epsilon

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

20

10

X (non-differentiable) : T1 — input

Y (non-differentiable) : T2 — output

detect_negative : int (default is 1) — (Optional) Whether map negative infinity to true. Default to 1 so that negative infinity induces true. Set this attribute to 0 if negative infinity should be mapped to false.
detect_positive : int (default is 1) — (Optional) Whether map positive infinity to true. Default to 1 so that positive infinity induces true. Set this attribute to 0 if positive infinity should be mapped to false.

T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz) — Constrain input types to float tensors.
T2 : tensor(bool) — Constrain output types to boolean tensors.

test_isinf, test_isinf_float16, test_isinf_negative, test_isinf_positive

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

20

9, 13

X (non-differentiable) : T1 — input

Y (non-differentiable) : T2 — output

None.

T1 : tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz) — Constrain input types to float tensors.
T2 : tensor(bool) — Constrain output types to boolean tensors.

test_isnan_float16, test_isnan

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

1

X (differentiable) : T — Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 …​ Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …​].

Y (differentiable) : T — Output tensor, which has the shape and type as input tensor

alpha : float (default is 0.0001) — Scaling parameter.
beta : float (default is 0.75) — The exponent.
bias : float (default is 1.0)
size : int (required) — The number of channels to sum over

T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input and output types to float tensors.

test_lrn_default, test_lrn

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

17

X : T — Tensor to be normalized.
Scale : T — Scale tensor.
B (optional) : T — Bias tensor.

Y : T — Normalized tensor.
Mean (optional) : U — Saved mean used during training to speed up gradient computation
InvStdDev (optional) : U — Saved inverse standard deviation used during training to speed up gradient computation.

axis : int (default is -1) — The first normalization dimension. If rank(X) is r, axis’ allowed range is [-r, r). Negative value means counting dimensions from the back.
epsilon : float (default is 1e-05) — The epsilon value to use to avoid division by zero.
stash_type : int (default is 1) — Type of Mean and InvStdDev. This also specifies stage one’s computation precision.

T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input types and output Y type to float tensors.
U : tensor(float), tensor(bfloat16) — Type of Mean and InvStdDev tensors.

test_layer_normalization_default_axis

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

16

1, 6

X (differentiable) : T — Input tensor

Y (differentiable) : T — Output tensor

alpha : float (default is 0.01) — Coefficient of leakage.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_leakyrelu_example, test_leakyrelu, test_leakyrelu_default

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

1, 7, 9

A (non-differentiable) : T — First input operand for the logical operator.
B (non-differentiable) : T — Second input operand for the logical operator.

C (non-differentiable) : T1 — Result tensor.

None.

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input types to all numeric tensors.
T1 : tensor(bool) — Constrain output to boolean tensor.

test_less, test_less_int8, test_less_int16, test_less_uint8, test_less_uint16, test_less_uint32, test_less_uint64, test_less_bcast

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

16

12

A (non-differentiable) : T — First input operand for the logical operator.
B (non-differentiable) : T — Second input operand for the logical operator.

C (non-differentiable) : T1 — Result tensor.

None.

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input types to all numeric tensors.
T1 : tensor(bool) — Constrain output to boolean tensor.

test_less_equal_bcast, test_less_equal, test_less_equal_int8, test_less_equal_int16, test_less_equal_uint8, test_less_equal_uint16, test_less_equal_uint32, test_less_equal_uint64

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

27

query (differentiable) : T — Query vectors with 3D packed shape (B, T, H_q * d_k). Heads are packed into the last dimension.
key (differentiable) : T — Key vectors with 3D packed shape (B, T, H_kv * d_k). Should be L2-normalized for delta/gated_delta modes.
value (differentiable) : T — Value vectors with 3D packed shape (B, T, H_kv * d_v).
past_state (optional, non-differentiable) : S — Recurrent state from previous step with shape (B, H_kv, d_k, d_v). Always 4D. If not provided, defaults to zeros.
decay (optional, differentiable) : T — Exponential decay gate in log-space. 3D packed shape: (B, T, H_kv * d_k) for per-key-dimension decay (GLA/RWKV-6), or (B, T, H_kv) for per-head scalar decay (DeltaNet/RetNet). Required for ‘gated’ and ‘gated_delta’ modes.
beta (optional, differentiable) : T — Update rate (sigmoid output). 3D packed shape: (B, T, H_kv) or (B, T, 1). Required for ‘delta’ and ‘gated_delta’ modes.

output (differentiable) : T — Attention output with 3D packed shape (B, T, H_q * d_v).
present_state (non-differentiable) : S — Updated recurrent state with shape (B, H_kv, d_k, d_v). Always 4D.

chunk_size : int (default is 64) — Chunk size for the chunk-parallel WY decomposition during prefill (T>1). Tuning hint; does not affect output correctness.
kv_num_heads : int (required) — Number of key/value heads. Always required.
q_num_heads : int (required) — Number of query heads. Always required.
scale : float (default is 0.0) — Output scaling factor. When 0.0 (default), derives d_k = query.shape[-1] / q_num_heads and uses 1/sqrt(d_k). Set explicitly to override.
update_rule : string (default is gated_delta) — The update rule for the linear attention recurrence. One of: ‘linear’, ‘gated’, ‘delta’, ‘gated_delta’. Default is ‘gated_delta’.

T : tensor(float16), tensor(bfloat16), tensor(float) — Constrain activation input and output types to float16, bfloat16, or float32 tensors.
S : tensor(float16), tensor(bfloat16), tensor(float) — Constrain state types to float16, bfloat16, or float32 tensors. Should be float32 or the same as T for numerical stability on long sequences.

test_linear_attention_decode_step, test_linear_attention_delta, test_linear_attention_explicit_scale, test_linear_attention_fp16, test_linear_attention_gated, test_linear_attention_gated_delta, test_linear_attention_gated_delta_beta_scalar, test_linear_attention_gated_delta_gqa, test_linear_attention_gated_delta_mqa, test_linear_attention_gated_per_head_decay, test_linear_attention_linear, test_linear_attention_linear_t1_no_past, test_linear_attention_no_past_explicit_zeros, test_linear_attention_prefill_with_past

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

1, 6

input (differentiable) : T — Input tensor

output (differentiable) : T — The natural log of the input tensor computed element-wise

None.

T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input and output types to float tensors.

test_log_example, test_log

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

1, 11

input (differentiable) : T — The input tensor of rank >= axis.

output (differentiable) : T — The output values with the same shape as the input tensor.

axis : int (default is -1) — Describes the dimension LogSoftmax will be performed on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).

T : tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input and output types to float tensors.

test_logsoftmax_example_1, test_logsoftmax_large_number, test_logsoftmax_axis_0, test_logsoftmax_axis_1, test_logsoftmax_axis_2, test_logsoftmax_negative_axis, test_logsoftmax_default_axis

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

25

1, 11, 13, 16, 19, 21, 23, 24

M (optional) : I — A maximum trip-count for the loop specified at runtime. Optional. Pass empty string to skip.
cond (optional) : B — A boolean termination condition. Optional. Pass empty string to skip.
v_initial (variadic, heterogeneous) : V — The initial values of any loop-carried dependencies (values that change across loop iterations)

v_final_and_scan_outputs (variadic, heterogeneous) : V — Final N loop carried dependency values then K scan_outputs. Scan outputs must be Tensors.

body : graph (required) — The graph run each iteration. It has 2+N inputs: (iteration_num, condition, loop carried dependencies…​). It has 1+N+K outputs: (condition, loop carried dependencies…​, scan_outputs…​). Each scan_output is created by concatenating the value of the specified output value at the end of each iteration of the loop. It is an error if the dimensions or data type of these scan_outputs change across loop iterations.

V : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(string), tensor(bool), tensor(complex64), tensor(complex128), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(uint4), tensor(int4), tensor(float4e2m1), tensor(float8e8m0), tensor(uint2), tensor(int2), seq(tensor(uint8)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(int8)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(bfloat16)), seq(tensor(float16)), seq(tensor(float)), seq(tensor(double)), seq(tensor(string)), seq(tensor(bool)), seq(tensor(complex64)), seq(tensor(complex128)), seq(tensor(float8e4m3fn)), seq(tensor(float8e4m3fnuz)), seq(tensor(float8e5m2)), seq(tensor(float8e5m2fnuz)), seq(tensor(uint4)), seq(tensor(int4)), seq(tensor(float4e2m1)), seq(tensor(float8e8m0)), seq(tensor(uint2)), seq(tensor(int2)), optional(seq(tensor(uint8))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(int8))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(bfloat16))), optional(seq(tensor(float16))), optional(seq(tensor(float))), optional(seq(tensor(double))), optional(seq(tensor(string))), optional(seq(tensor(bool))), optional(seq(tensor(complex64))), optional(seq(tensor(complex128))), optional(tensor(uint8)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(int8)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(bfloat16)), optional(tensor(float16)), optional(tensor(float)), optional(tensor(double)), optional(tensor(string)), optional(tensor(bool)), optional(tensor(complex64)), optional(tensor(complex128)), optional(tensor(float8e4m3fn)), optional(tensor(float8e4m3fnuz)), optional(tensor(float8e5m2)), optional(tensor(float8e5m2fnuz)), optional(tensor(uint4)), optional(tensor(int4)), optional(tensor(float4e2m1)), optional(tensor(float8e8m0)), optional(tensor(uint2)), optional(tensor(int2)) — All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types up to IRv13.
I : tensor(int64) — tensor of int64, which should be a scalar.
B : tensor(bool) — tensor of bool, which should be a scalar.

test_loop11, test_loop13_seq, test_loop16_seq_none

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

1

input (differentiable) : T — Input matrix

output (differentiable) : T — Matrix after normalization

axis : int (default is -1) — The axis on which to apply normalization, -1 mean last axis.
p : int (default is 2) — The order of the normalization, only 1 or 2 are supported.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_lpnormalization_default, test_l1normalization_axis_0, test_l1normalization_axis_1, test_l1normalization_axis_last, test_l2normalization_axis_0, test_l2normalization_axis_1

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

1, 2, 11, 18

X (differentiable) : T — Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 …​ Dn), where N is the batch size.

Y (differentiable) : T — Output data tensor from Lp pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes.

auto_pad : string (default is NOTSET) — auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
ceil_mode : int (default is 0) — Whether to use ceil or floor (default) to compute the output shape.
dilations : list of ints — dilation value along each spatial axis of the filter. If not present, the dilation defaults is 1 along each spatial axis.
kernel_shape : list of ints (required) — The size of the kernel along each axis.
p : int (default is 2) — p value of the Lp norm used to pool over the input data.
pads : list of ints — Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…​x1_end, x2_end,…​], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides : list of ints — Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

test_lppool_1d_default, test_lppool_2d_default, test_lppool_2d_dilations, test_lppool_2d_pads, test_lppool_2d_same_lower, test_lppool_2d_same_upper, test_lppool_2d_strides, test_lppool_3d_default

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

1, 9

A (differentiable) : T — N-dimensional matrix A
B (differentiable) : T — N-dimensional matrix B

Y (differentiable) : T — Matrix multiply results from A * B

None.

T : tensor(float16), tensor(float), tensor(double), tensor(uint32), tensor(uint64), tensor(int32), tensor(int64), tensor(bfloat16) — Constrain input and output types to float/int tensors.

test_matmul_2d, test_matmul_3d, test_matmul_4d, test_matmul_bcast, test_matmul_1d_3d, test_matmul_4d_1d, test_matmul_1d_1d

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

10

A (non-differentiable) : T1 — N-dimensional matrix A
B (non-differentiable) : T2 — N-dimensional matrix B
a_zero_point (optional, non-differentiable) : T1 — Zero point tensor for input ‘A’. It’s optional and default value is 0. It could be a scalar or N-D tensor. Scalar refers to per tensor quantization whereas N-D refers to per row quantization. If the input is 2D of shape [M, K] then zero point tensor may be an M element vector [zp_1, zp_2, …​, zp_M]. If the input is N-D tensor with shape [D1, D2, M, K] then zero point tensor may have shape [D1, D2, M, 1].
b_zero_point (optional, non-differentiable) : T2 — Zero point tensor for input ‘B’. It’s optional and default value is 0. It could be a scalar or a N-D tensor, Scalar refers to per tensor quantization whereas N-D refers to per col quantization. If the input is 2D of shape [K, N] then zero point tensor may be an N element vector [zp_1, zp_2, …​, zp_N]. If the input is N-D tensor with shape [D1, D2, K, N] then zero point tensor may have shape [D1, D2, 1, N].

Y (non-differentiable) : T3 — Matrix multiply results from A * B

None.

T1 : tensor(int8), tensor(uint8) — Constrain input A data type to 8-bit integer tensor.
T2 : tensor(int8), tensor(uint8) — Constrain input B data type to 8-bit integer tensor.
T3 : tensor(int32) — Constrain output Y data type as 32-bit integer tensor.

test_matmulinteger

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

13

1, 6, 8, 12

data_0 (variadic, differentiable) : T — List of tensors for max.

max (differentiable) : T — Output tensor.

None.

T : tensor(uint8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(int8), tensor(int16), tensor(int32), tensor(int64), tensor(float16), tensor(float), tensor(double), tensor(bfloat16) — Constrain input and output types to numeric tensors.

test_max_example, test_max_one_input, test_max_two_inputs

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

1, 8, 10, 11, 12

X (differentiable) : T — Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 …​ Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …​].

Y (differentiable) : T — Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
Indices (optional, non-differentiable) : I — Indices tensor from max pooling across the input tensor. The dimensions of indices are the same as output tensor. The values in indices of are the indices of the selected values during pooling. The indices are computed as flatten 1-D tensor, and the indices do not consider padding. So the values in indices are in [0, N x C x D1 x …​ x Dn).

auto_pad : string (default is NOTSET) — auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER.
ceil_mode : int (default is 0) — Whether to use ceil or floor (default) to compute the output shape.
dilations : list of ints — Dilation value along each spatial axis of filter. If not present, the dilation defaults to 1 along each spatial axis.
kernel_shape : list of ints (required) — The size of the kernel along each axis.
pads : list of ints — Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…​x1_end, x2_end,…​], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
storage_order : int (default is 0) — The storage order of the tensor. 0 is row major, and 1 is column major. This attribute is used only to convert an n-tuple index value into a single integer value for producing the second output.
strides : list of ints — Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double), tensor(int8), tensor(uint8) — Constrain input and output types to float and 8 bit tensors.
I : tensor(int64) — Constrain index tensor to int64

test_maxpool_1d_default, test_maxpool_2d_ceil, test_maxpool_2d_ceil_output_size_reduce_by_one, test_maxpool_2d_default, test_maxpool_2d_dilations, test_maxpool_2d_pads, test_maxpool_2d_precomputed_pads, test_maxpool_2d_precomputed_same_upper, test_maxpool_2d_precomputed_strides, test_maxpool_2d_same_lower, test_maxpool_2d_same_upper, test_maxpool_2d_strides, test_maxpool_2d_uint8, test_maxpool_3d_default, test_maxpool_3d_dilations, test_maxpool_3d_dilations_use_ref_impl, test_maxpool_3d_dilations_use_ref_impl_large, test_maxpool_with_argmax_2d_precomputed_pads, test_maxpool_with_argmax_2d_precomputed_strides

Shape inference, determinism, errors. See the editorial note at the head of this clause.

ai.onnx

22

1

X (differentiable) : T — Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data.
rois (non-differentiable) : T — RoIs (Regions of Interest) to pool over. Should be a 2-D tensor of shape (num_rois, 5) given as [[batch_id, x1, y1, x2, y2], …​].

Y (differentiable) : T — RoI pooled output 4-D tensor of shape (num_rois, channels, pooled_shape[0], pooled_shape[1]).

pooled_shape : list of ints (required) — ROI pool output shape (height, width).
spatial_scale : float (default is 1.0) — Multiplicative spatial scale factor to translate ROI coordinates from their input scale to the scale used when pooling.

T : tensor(bfloat16), tensor(float16), tensor(float), tensor(double) — Constrain input and output types to float tensors.

None published for this operator.

Shape inference, determinism, errors. See the editorial note at the head of this clause.