Supported aten operators

Note

This table and page are auto generated by docs/contributing/generate_support_page.py and reflect the PyTorch reference docs at the time of generation. Targeted torch version: 2.11. Generated on 14 May 2026.

Warning

Take these results with a grain of salt: many of the listed operators never appear in the torch IR graph that torch_to_nnef traces (they get remapped to more generic ops upstream), and some uncommon operators are rare in real models so support may be lacking even when marked unsupported. SONOS maintains operators on a per-need basis, and contributions are always welcome see how.

'is core' column refers to this PyTorch IR documentation page (the page for torch 2.11 was emptied upstream; falling back to torch 2.9 which is the last version that still enumerates the core ATen IR).

We filter out 'backward' and 'sym' operators from the listing since they are unwanted in an inference engine. In-place operations are merged with their non-inplace counterparts since that distinction is an inference implementation detail.

We also exclude a long tail of identifiers that the aten::* source-grep picks up but that can never surface in an inference JIT trace (see the full list at the bottom of this page). This trims the unsupported column to the names where a torch_to_nnef emitter (or a deliberate no-op map) would actually be meaningful.

TractNNEFONNX

Total matched operators in torch_to_nnef compared to:

core PyTorch opset:

127/138

and support from full aten:::

323/530

(total operators listed as supported by torch_to_nnef being 376)

translated	aten name	aliases	can in-place	is core
✅	abs	absolute	✅	✅
✅	acos	arccos	✅	✅
✅	acosh	arccosh	✅	✅
✅	adaptive_avg_pool1d		❌	✅
✅	add		✅	✅
✅	addmm		✅	✅
✅	alias		❌	✅
✅	amax		❌	✅
✅	amin		❌	✅
✅	any		❌	✅
✅	arange		❌	✅
✅	argmax		❌	✅
✅	argmin		❌	✅
❌	as_strided		✅	✅
✅	asin	arcsin	✅	✅
✅	asinh	arcsinh	✅	✅
✅	atan	arctan	✅	✅
✅	atan2	arctan2	✅	✅
✅	atanh	arctanh	✅	✅
✅	avg_pool1d		❌	✅
✅	avg_pool2d		❌	✅
✅	avg_pool3d		❌	✅
✅	bitwise_and		✅	✅
✅	bitwise_not		✅	✅
✅	bitwise_or		✅	✅
✅	bitwise_xor		✅	✅
✅	bmm		❌	✅
✅	cat	concat, concatenate	❌	✅
✅	ceil		✅	✅
✅	clamp	clip	✅	✅
✅	clone		❌	✅
✅	col2im		❌	✅
✅	constant_pad_nd		❌	✅
✅	convolution		❌	✅
✅	copy		✅	✅
✅	cos		✅	✅
✅	cosh		✅	✅
✅	cumsum		✅	✅
✅	diagonal		❌	✅
✅	div	divide, true_divide	✅	✅
✅	elu		✅	✅
✅	embedding		❌	✅
❌	empty		❌	✅
❌	empty_strided		❌	✅
✅	eq		✅	✅
✅	erf	special_erf	✅	✅
✅	exp		✅	✅
✅	expand		❌	✅
✅	expm1	special_expm1	✅	✅
✅	fill		✅	✅
✅	flip		❌	✅
✅	floor		✅	✅
✅	fmod		✅	✅
✅	full		❌	✅
✅	full_like		❌	✅
✅	gather		❌	✅
✅	ge	greater_equal	✅	✅
✅	gelu		✅	✅
❌	grid_sampler_2d		❌	✅
✅	gt	greater	✅	✅
✅	hardtanh		✅	✅
✅	index		❌	✅
✅	index_put		✅	✅
✅	index_select		❌	✅
✅	isinf		❌	✅
✅	isnan		❌	✅
✅	le	less_equal	✅	✅
✅	leaky_relu		✅	✅
✅	log		✅	✅
✅	log10		✅	✅
✅	log1p	special_log1p	✅	✅
✅	log2		✅	✅
✅	logical_and		✅	✅
✅	logical_not		✅	✅
✅	logical_or		✅	✅
✅	logical_xor		✅	✅
✅	lt	less	✅	✅
❌	masked_scatter		✅	✅
✅	max		❌	✅
✅	max_pool2d_with_indices		❌	✅
✅	max_pool3d_with_indices		❌	✅
✅	maximum		❌	✅
✅	mean		❌	✅
✅	min		❌	✅
✅	minimum		❌	✅
✅	mm		❌	✅
✅	mul	multiply	✅	✅
✅	native_dropout		❌	✅
✅	native_group_norm		❌	✅
✅	native_layer_norm		❌	✅
✅	ne	not_equal	✅	✅
✅	neg	negative	✅	✅
❌	nonzero		❌	✅
✅	permute		❌	✅
✅	pow		✅	✅
✅	prod		❌	✅
❌	rand		❌	✅
❌	randn		❌	✅
❌	randperm		❌	✅
✅	reciprocal		✅	✅
✅	reflection_pad1d		❌	✅
✅	reflection_pad2d		❌	✅
✅	reflection_pad3d		❌	✅
✅	relu		✅	✅
✅	remainder		✅	✅
✅	repeat		❌	✅
✅	replication_pad2d		❌	✅
✅	replication_pad3d		❌	✅
❌	resize_		❌	✅
✅	round	special_round	✅	✅
✅	rsqrt		✅	✅
✅	scalar_tensor		❌	✅
✅	scatter		✅	✅
✅	scatter_add		✅	✅
✅	scatter_reduce		✅	✅
✅	select		❌	✅
✅	sigmoid	special_expit	✅	✅
✅	sign		✅	✅
✅	sin		✅	✅
✅	sinh		✅	✅
✅	slice		❌	✅
✅	slice_scatter		❌	✅
✅	sort		❌	✅
✅	split_with_sizes		❌	✅
✅	sqrt		✅	✅
✅	squeeze		✅	✅
✅	sub	subtract	✅	✅
✅	sum		❌	✅
✅	tan		✅	✅
✅	tanh		✅	✅
✅	topk		❌	✅
✅	trunc	fix	✅	✅
✅	unsqueeze		✅	✅
❌	upsample_bilinear2d		❌	✅
✅	upsample_nearest2d		❌	✅
✅	var		❌	✅
✅	view		❌	✅
✅	where		❌	✅
✅	adaptive_avg_pool2d		❌	-
✅	adaptive_avg_pool3d		❌	-
✅	adaptive_max_pool1d		❌	-
✅	adaptive_max_pool2d		❌	-
✅	adaptive_max_pool3d		❌	-
❌	addbmm		✅	-
❌	addcdiv		❌	-
✅	addcmul		❌	-
❌	addmv		✅	-
❌	addr		✅	-
✅	affine_grid_generator		❌	-
✅	all		❌	-
❌	allclose		❌	-
✅	alpha_dropout		✅	-
✅	aminmax		❌	-
✅	angle		❌	-
✅	argsort		❌	-
❌	argwhere		❌	-
❌	as_tensor		❌	-
✅	atleast_1d		❌	-
✅	atleast_2d		❌	-
✅	atleast_3d		❌	-
✅	baddbmm		✅	-
❌	bartlett_window		❌	-
✅	batch_norm		❌	-
❌	bernoulli		✅	-
❌	bias_addmm		❌	-
❌	bilinear		❌	-
❌	binary_cross_entropy		❌	-
✅	binary_cross_entropy_with_logits		❌	-
❌	bincount		❌	-
❌	binomial		❌	-
✅	bitwise_left_shift		✅	-
✅	bitwise_right_shift		✅	-
✅	blackman_window		❌	-
✅	broadcast_tensors		❌	-
✅	broadcast_to		❌	-
✅	bucketize		❌	-
❌	cauchy		✅	-
✅	cdist		❌	-
✅	celu		❌	-
❌	chain_matmul		❌	-
❌	chalf		❌	-
✅	channel_shuffle		❌	-
❌	cholesky		❌	-
❌	cholesky_inverse		❌	-
❌	cholesky_solve		❌	-
✅	chunk		❌	-
✅	clamp_max		✅	-
✅	clamp_min		✅	-
❌	column_stack		❌	-
✅	complex		❌	-
✅	conj		❌	-
✅	conj_physical		✅	-
✅	contiguous		❌	-
❌	conv		❌	-
✅	conv1d		❌	-
✅	conv2d		❌	-
✅	conv3d		❌	-
✅	conv_tbc		❌	-
✅	conv_transpose1d		❌	-
✅	conv_transpose2d		❌	-
✅	conv_transpose3d		❌	-
✅	convolution_overrideable		❌	-
❌	copy_sparse_to_sparse_		❌	-
❌	copy_to		❌	-
✅	copysign		✅	-
❌	cosine_embedding_loss		❌	-
✅	cosine_similarity		❌	-
✅	cross		❌	-
✅	cross_entropy_loss		❌	-
❌	ctc_loss		❌	-
✅	cummax		❌	-
✅	cummin		❌	-
✅	cumprod		✅	-
✅	dequantize		❌	-
✅	detach		✅	-
❌	diag		❌	-
❌	diagflat		❌	-
❌	digamma	special_digamma, special_psi	✅	-
❌	dist		❌	-
✅	dot		❌	-
✅	dropout		✅	-
✅	dstack		❌	-
✅	einsum		❌	-
✅	embedding_bag		❌	-
❌	embedding_renorm_		❌	-
✅	empty_like		❌	-
❌	empty_permuted		❌	-
❌	empty_quantized		❌	-
❌	erfc	special_erfc	✅	-
❌	erfinv	special_erfinv	✅	-
✅	exp2	special_exp2	✅	-
✅	expand_as		❌	-
❌	exponential		✅	-
✅	eye		❌	-
✅	feature_alpha_dropout		✅	-
✅	feature_dropout		✅	-
❌	fft_fftfreq		❌	-
❌	fft_ihfft2		❌	-
❌	fft_ihfftn		❌	-
❌	fft_rfftfreq		❌	-
✅	flatten		❌	-
❌	flatten_dense_tensors		❌	-
✅	floor_divide		✅	-
✅	floordiv		❌	-
✅	fmax		❌	-
✅	fmin		❌	-
❌	frac		❌	-
❌	fractional_max_pool2d		❌	-
❌	fractional_max_pool3d		❌	-
❌	frexp		❌	-
✅	frobenius_norm		❌	-
❌	gamma		❌	-
❌	gcd		✅	-
❌	geometric		✅	-
❌	geqrf		❌	-
✅	glu		❌	-
❌	grid_sampler		❌	-
❌	grid_sampler_3d		❌	-
✅	group_norm		❌	-
✅	gru		❌	-
❌	gru_cell		❌	-
✅	hamming_window		❌	-
✅	hann_window		❌	-
✅	hardshrink		❌	-
✅	hardsigmoid		✅	-
✅	hardswish		✅	-
❌	hash_tensor		❌	-
✅	heaviside		❌	-
❌	hinge_embedding_loss		❌	-
❌	histc		❌	-
❌	histogram		❌	-
❌	histogramdd		❌	-
✅	hstack		❌	-
✅	hypot		✅	-
❌	i0	special_i0	✅	-
❌	igamma	special_gammainc	✅	-
❌	igammac	special_gammaincc	✅	-
✅	im2col		❌	-
✅	index_add		❌	-
✅	index_copy		❌	-
✅	index_fill		❌	-
❌	index_reduce		✅	-
✅	instance_norm		❌	-
✅	isclose		❌	-
✅	isfinite		❌	-
❌	istft		❌	-
✅	kaiser_window		❌	-
✅	kl_div		❌	-
❌	kthvalue		❌	-
❌	l1_loss		❌	-
✅	layer_norm		❌	-
❌	lcm		✅	-
❌	ldexp		✅	-
✅	lerp		❌	-
❌	lgamma		✅	-
❌	linalg__powsum		❌	-
❌	linalg_cond		❌	-
✅	linalg_cross		❌	-
❌	linalg_det	det	❌	-
❌	linalg_eig		❌	-
❌	linalg_eigh		❌	-
❌	linalg_eigvals		❌	-
❌	linalg_eigvalsh		❌	-
❌	linalg_householder_product	orgqr	❌	-
❌	linalg_inv	inverse	❌	-
❌	linalg_ldl_solve		❌	-
❌	linalg_lstsq		❌	-
❌	linalg_lu		❌	-
❌	linalg_lu_solve		❌	-
❌	linalg_matrix_exp	matrix_exp	❌	-
✅	linalg_matrix_norm		❌	-
❌	linalg_matrix_power	matrix_power	❌	-
❌	linalg_matrix_rank		❌	-
✅	linalg_norm		❌	-
❌	linalg_pinv		❌	-
❌	linalg_qr		❌	-
❌	linalg_slogdet		❌	-
❌	linalg_solve		❌	-
❌	linalg_solve_triangular		❌	-
❌	linalg_svd		❌	-
❌	linalg_tensorsolve		❌	-
✅	linalg_vector_norm		❌	-
✅	linear		❌	-
✅	linspace		❌	-
❌	log_normal		✅	-
✅	log_sigmoid		❌	-
✅	log_softmax	special_log_softmax	❌	-
❌	logcumsumexp		❌	-
❌	logdet		❌	-
✅	logit	special_logit	✅	-
❌	logspace		❌	-
✅	logsumexp	special_logsumexp	❌	-
✅	lstm		❌	-
✅	lstm_cell		❌	-
❌	lu_solve		❌	-
❌	lu_unpack		❌	-
✅	mH	adjoint	❌	-
✅	mT		❌	-
❌	margin_ranking_loss		❌	-
✅	masked_fill		✅	-
❌	masked_select		❌	-
✅	matmul	linalg_matmul	❌	-
✅	matrix_H		❌	-
✅	max_pool1d		❌	-
✅	max_pool1d_with_indices		❌	-
✅	max_pool2d		❌	-
✅	max_pool3d		❌	-
❌	median		❌	-
✅	meshgrid		❌	-
✅	mish		❌	-
❌	mode		❌	-
❌	modf		❌	-
✅	movedim	moveaxis	❌	-
✅	mse_loss		❌	-
❌	multi_margin_loss		❌	-
❌	multilabel_margin_loss		❌	-
❌	multinomial		❌	-
✅	mv		❌	-
❌	mvlgamma	special_multigammaln	❌	-
✅	nan_to_num		✅	-
❌	nanmedian		❌	-
❌	nanquantile		❌	-
✅	narrow		❌	-
✅	native_channel_shuffle		❌	-
❌	native_multi_head_self_attention		❌	-
✅	new_empty		❌	-
❌	new_empty_strided		❌	-
✅	new_full		❌	-
✅	new_ones		❌	-
✅	new_zeros		❌	-
❌	nextafter		✅	-
✅	nll_loss		❌	-
✅	nll_loss2d		❌	-
✅	nll_loss_nd		❌	-
❌	nonzero_numpy		❌	-
❌	nonzero_static		❌	-
✅	norm		❌	-
❌	normal		✅	-
❌	nuclear_norm		❌	-
✅	numel		❌	-
✅	numpy_T		❌	-
✅	one_hot		❌	-
✅	ones		❌	-
✅	ones_like		❌	-
❌	ormqr		❌	-
✅	outer	ger	❌	-
✅	pad		❌	-
❌	pad_sequence		❌	-
✅	pairwise_distance		❌	-
❌	pdist		❌	-
✅	pixel_shuffle		❌	-
✅	pixel_unshuffle		❌	-
❌	poisson		❌	-
✅	polar		❌	-
❌	polygamma	special_polygamma	✅	-
✅	prelu		❌	-
❌	put		✅	-
❌	qr		❌	-
❌	quantile		❌	-
❌	quantize		❌	-
❌	quantize_per_channel		❌	-
✅	quantize_per_tensor		❌	-
❌	quantize_per_tensor_dynamic		❌	-
❌	quantized_batch_norm		❌	-
❌	quantized_gru		❌	-
❌	quantized_lstm		❌	-
❌	quantized_max_pool1d		❌	-
❌	quantized_max_pool2d		❌	-
❌	quantized_max_pool3d		❌	-
❌	rand_like		❌	-
❌	randint		❌	-
❌	randint_like		❌	-
❌	randn_like		❌	-
❌	random		✅	-
❌	range		❌	-
✅	relu6		✅	-
❌	renorm		❌	-
✅	repeat_interleave		❌	-
✅	replication_pad1d		❌	-
✅	reshape		❌	-
✅	reshape_as		❌	-
❌	resize_as_sparse_		❌	-
✅	resolve_conj		❌	-
✅	resolve_neg		❌	-
✅	rnn_relu		❌	-
❌	rnn_relu_cell		❌	-
✅	rnn_tanh		❌	-
❌	rnn_tanh_cell		❌	-
✅	roll		❌	-
✅	rot90		❌	-
❌	rrelu		❌	-
❌	rrelu_with_noise		✅	-
✅	rsub		❌	-
✅	scaled_dot_product_attention		❌	-
✅	searchsorted		❌	-
❌	segment_reduce		❌	-
✅	selu		✅	-
❌	set_		❌	-
❌	signbit		❌	-
✅	silu		✅	-
✅	sinc	special_sinc	❌	-
✅	size		❌	-
✅	smooth_l1_loss		❌	-
❌	soft_margin_loss		❌	-
✅	softmax	special_softmax	❌	-
✅	softplus		❌	-
✅	softshrink		❌	-
❌	special_airy_ai		❌	-
❌	special_bessel_j0		❌	-
❌	special_bessel_j1		❌	-
❌	special_bessel_y0		❌	-
❌	special_bessel_y1		❌	-
❌	special_chebyshev_polynomial_t		❌	-
❌	special_chebyshev_polynomial_u		❌	-
❌	special_chebyshev_polynomial_v		❌	-
❌	special_chebyshev_polynomial_w		❌	-
❌	special_erfcx		❌	-
❌	special_hermite_polynomial_h		❌	-
❌	special_hermite_polynomial_he		❌	-
❌	special_i0e		❌	-
❌	special_i1		❌	-
❌	special_i1e		❌	-
❌	special_laguerre_polynomial_l		❌	-
❌	special_legendre_polynomial_p		❌	-
❌	special_log_ndtr		❌	-
❌	special_modified_bessel_i0		❌	-
❌	special_modified_bessel_i1		❌	-
❌	special_modified_bessel_k0		❌	-
❌	special_modified_bessel_k1		❌	-
❌	special_ndtri		❌	-
❌	special_scaled_modified_bessel_k0		❌	-
❌	special_scaled_modified_bessel_k1		❌	-
❌	special_shifted_chebyshev_polynomial_t		❌	-
❌	special_shifted_chebyshev_polynomial_u		❌	-
❌	special_shifted_chebyshev_polynomial_v		❌	-
❌	special_shifted_chebyshev_polynomial_w		❌	-
❌	special_spherical_bessel_j0		❌	-
❌	special_zeta		❌	-
✅	split		❌	-
✅	square		✅	-
✅	stack		❌	-
✅	std		❌	-
✅	std_mean		❌	-
✅	stft		❌	-
✅	t		✅	-
✅	take		❌	-
❌	tanhshrink		❌	-
✅	tensor_split		❌	-
✅	tensordot		❌	-
✅	threshold		❌	-
✅	tile		❌	-
✅	to		❌	-
❌	trace		❌	-
✅	transpose	swapaxes, swapdims	✅	-
❌	triangular_solve		❌	-
✅	tril		❌	-
❌	tril_indices		❌	-
❌	triplet_margin_loss		❌	-
✅	triu		❌	-
❌	triu_indices		❌	-
✅	type_as		❌	-
✅	unbind		❌	-
✅	unflatten		❌	-
❌	unflatten_dense_tensors		❌	-
✅	unfold		❌	-
❌	uniform		✅	-
❌	unique_consecutive		❌	-
❌	unique_dim		❌	-
❌	unique_dim_consecutive		❌	-
✅	unsafe_chunk		❌	-
✅	unsafe_split		❌	-
✅	unsafe_split_with_sizes		❌	-
❌	upsample		❌	-
❌	upsample_bicubic2d		❌	-
❌	upsample_linear1d		❌	-
✅	upsample_nearest1d		❌	-
✅	upsample_nearest3d		❌	-
❌	upsample_trilinear3d		❌	-
❌	vander		❌	-
✅	var_mean		❌	-
✅	view_as		❌	-
✅	view_as_complex		❌	-
✅	view_as_real		❌	-
✅	vstack	row_stack	❌	-
❌	wrapped_linear_prepack		❌	-
❌	wrapped_quantized_linear_prepacked		❌	-
✅	xlogy	special_xlogy	❌	-
✅	zero		✅	-
✅	zeros		❌	-
✅	zeros_like		❌	-

Total matched operators in builtin PyTorch ONNX support based on this page (the page for torch 2.11 was removed upstream in torch 2.9+; falling back to torch 2.8 which is the last version that still ships the TorchScript ONNX support listing) compared to:

core PyTorch opset:

123/138

and support from full aten:::

306/530

(total operators listed as supported by PyTorch's TorchScript ONNX exporter being 348)

translated	aten name	aliases	can in-place	is core
✅	abs	absolute	✅	✅
✅	acos	arccos	✅	✅
❌	acosh	arccosh	✅	✅
✅	adaptive_avg_pool1d		❌	✅
✅	add		✅	✅
✅	addmm		✅	✅
✅	alias		❌	✅
✅	amax		❌	✅
✅	amin		❌	✅
✅	any		❌	✅
✅	arange		❌	✅
✅	argmax		❌	✅
✅	argmin		❌	✅
✅	as_strided		✅	✅
✅	asin	arcsin	✅	✅
❌	asinh	arcsinh	✅	✅
✅	atan	arctan	✅	✅
✅	atan2	arctan2	✅	✅
❌	atanh	arctanh	✅	✅
✅	avg_pool1d		❌	✅
✅	avg_pool2d		❌	✅
✅	avg_pool3d		❌	✅
✅	bitwise_and		✅	✅
✅	bitwise_not		✅	✅
✅	bitwise_or		✅	✅
❌	bitwise_xor		✅	✅
✅	bmm		❌	✅
✅	cat	concat, concatenate	❌	✅
✅	ceil		✅	✅
✅	clamp	clip	✅	✅
✅	clone		❌	✅
✅	col2im		❌	✅
✅	constant_pad_nd		❌	✅
✅	convolution		❌	✅
❌	copy		✅	✅
✅	cos		✅	✅
❌	cosh		✅	✅
✅	cumsum		✅	✅
✅	diagonal		❌	✅
✅	div	divide, true_divide	✅	✅
✅	elu		✅	✅
✅	embedding		❌	✅
✅	empty		❌	✅
❌	empty_strided		❌	✅
✅	eq		✅	✅
✅	erf	special_erf	✅	✅
✅	exp		✅	✅
✅	expand		❌	✅
❌	expm1	special_expm1	✅	✅
✅	fill		✅	✅
✅	flip		❌	✅
✅	floor		✅	✅
✅	fmod		✅	✅
✅	full		❌	✅
✅	full_like		❌	✅
✅	gather		❌	✅
✅	ge	greater_equal	✅	✅
✅	gelu		✅	✅
❌	grid_sampler_2d		❌	✅
✅	gt	greater	✅	✅
✅	hardtanh		✅	✅
✅	index		❌	✅
✅	index_put		✅	✅
✅	index_select		❌	✅
✅	isinf		❌	✅
✅	isnan		❌	✅
✅	le	less_equal	✅	✅
✅	leaky_relu		✅	✅
✅	log		✅	✅
✅	log10		✅	✅
✅	log1p	special_log1p	✅	✅
✅	log2		✅	✅
✅	logical_and		✅	✅
✅	logical_not		✅	✅
✅	logical_or		✅	✅
✅	logical_xor		✅	✅
✅	lt	less	✅	✅
✅	masked_scatter		✅	✅
✅	max		❌	✅
✅	max_pool2d_with_indices		❌	✅
✅	max_pool3d_with_indices		❌	✅
✅	maximum		❌	✅
✅	mean		❌	✅
✅	min		❌	✅
✅	minimum		❌	✅
✅	mm		❌	✅
✅	mul	multiply	✅	✅
✅	native_dropout		❌	✅
❌	native_group_norm		❌	✅
✅	native_layer_norm		❌	✅
✅	ne	not_equal	✅	✅
✅	neg	negative	✅	✅
✅	nonzero		❌	✅
✅	permute		❌	✅
✅	pow		✅	✅
✅	prod		❌	✅
✅	rand		❌	✅
✅	randn		❌	✅
❌	randperm		❌	✅
✅	reciprocal		✅	✅
✅	reflection_pad1d		❌	✅
✅	reflection_pad2d		❌	✅
✅	reflection_pad3d		❌	✅
✅	relu		✅	✅
✅	remainder		✅	✅
✅	repeat		❌	✅
✅	replication_pad2d		❌	✅
✅	replication_pad3d		❌	✅
❌	resize_		❌	✅
✅	round	special_round	✅	✅
✅	rsqrt		✅	✅
✅	scalar_tensor		❌	✅
✅	scatter		✅	✅
✅	scatter_add		✅	✅
✅	scatter_reduce		✅	✅
✅	select		❌	✅
✅	sigmoid	special_expit	✅	✅
✅	sign		✅	✅
✅	sin		✅	✅
❌	sinh		✅	✅
✅	slice		❌	✅
❌	slice_scatter		❌	✅
✅	sort		❌	✅
✅	split_with_sizes		❌	✅
✅	sqrt		✅	✅
✅	squeeze		✅	✅
✅	sub	subtract	✅	✅
✅	sum		❌	✅
✅	tan		✅	✅
✅	tanh		✅	✅
✅	topk		❌	✅
❌	trunc	fix	✅	✅
✅	unsqueeze		✅	✅
✅	upsample_bilinear2d		❌	✅
✅	upsample_nearest2d		❌	✅
✅	var		❌	✅
✅	view		❌	✅
✅	where		❌	✅
✅	adaptive_avg_pool2d		❌	-
✅	adaptive_avg_pool3d		❌	-
✅	adaptive_max_pool1d		❌	-
✅	adaptive_max_pool2d		❌	-
✅	adaptive_max_pool3d		❌	-
❌	addbmm		✅	-
❌	addcdiv		❌	-
✅	addcmul		❌	-
❌	addmv		✅	-
❌	addr		✅	-
✅	affine_grid_generator		❌	-
✅	all		❌	-
❌	allclose		❌	-
✅	alpha_dropout		✅	-
✅	aminmax		❌	-
❌	angle		❌	-
✅	argsort		❌	-
❌	argwhere		❌	-
✅	as_tensor		❌	-
✅	atleast_1d		❌	-
✅	atleast_2d		❌	-
✅	atleast_3d		❌	-
✅	baddbmm		✅	-
❌	bartlett_window		❌	-
✅	batch_norm		❌	-
✅	bernoulli		✅	-
❌	bias_addmm		❌	-
❌	bilinear		❌	-
❌	binary_cross_entropy		❌	-
✅	binary_cross_entropy_with_logits		❌	-
❌	bincount		❌	-
❌	binomial		❌	-
✅	bitwise_left_shift		✅	-
✅	bitwise_right_shift		✅	-
❌	blackman_window		❌	-
✅	broadcast_tensors		❌	-
✅	broadcast_to		❌	-
✅	bucketize		❌	-
❌	cauchy		✅	-
✅	cdist		❌	-
✅	celu		❌	-
❌	chain_matmul		❌	-
❌	chalf		❌	-
❌	channel_shuffle		❌	-
❌	cholesky		❌	-
❌	cholesky_inverse		❌	-
❌	cholesky_solve		❌	-
✅	chunk		❌	-
✅	clamp_max		✅	-
✅	clamp_min		✅	-
❌	column_stack		❌	-
❌	complex		❌	-
❌	conj		❌	-
✅	conj_physical		✅	-
✅	contiguous		❌	-
❌	conv		❌	-
✅	conv1d		❌	-
✅	conv2d		❌	-
✅	conv3d		❌	-
✅	conv_tbc		❌	-
✅	conv_transpose1d		❌	-
✅	conv_transpose2d		❌	-
✅	conv_transpose3d		❌	-
❌	convolution_overrideable		❌	-
❌	copy_sparse_to_sparse_		❌	-
❌	copy_to		❌	-
❌	copysign		✅	-
❌	cosine_embedding_loss		❌	-
✅	cosine_similarity		❌	-
✅	cross		❌	-
✅	cross_entropy_loss		❌	-
❌	ctc_loss		❌	-
❌	cummax		❌	-
❌	cummin		❌	-
❌	cumprod		✅	-
✅	dequantize		❌	-
✅	detach		✅	-
❌	diag		❌	-
❌	diagflat		❌	-
❌	digamma	special_digamma, special_psi	✅	-
❌	dist		❌	-
✅	dot		❌	-
✅	dropout		✅	-
❌	dstack		❌	-
✅	einsum		❌	-
✅	embedding_bag		❌	-
❌	embedding_renorm_		❌	-
✅	empty_like		❌	-
❌	empty_permuted		❌	-
❌	empty_quantized		❌	-
❌	erfc	special_erfc	✅	-
❌	erfinv	special_erfinv	✅	-
❌	exp2	special_exp2	✅	-
✅	expand_as		❌	-
❌	exponential		✅	-
✅	eye		❌	-
✅	feature_alpha_dropout		✅	-
✅	feature_dropout		✅	-
❌	fft_fftfreq		❌	-
❌	fft_ihfft2		❌	-
❌	fft_ihfftn		❌	-
❌	fft_rfftfreq		❌	-
✅	flatten		❌	-
❌	flatten_dense_tensors		❌	-
✅	floor_divide		✅	-
✅	floordiv		❌	-
❌	fmax		❌	-
❌	fmin		❌	-
❌	frac		❌	-
❌	fractional_max_pool2d		❌	-
❌	fractional_max_pool3d		❌	-
❌	frexp		❌	-
✅	frobenius_norm		❌	-
❌	gamma		❌	-
❌	gcd		✅	-
❌	geometric		✅	-
❌	geqrf		❌	-
✅	glu		❌	-
✅	grid_sampler		❌	-
❌	grid_sampler_3d		❌	-
✅	group_norm		❌	-
✅	gru		❌	-
❌	gru_cell		❌	-
❌	hamming_window		❌	-
✅	hann_window		❌	-
✅	hardshrink		❌	-
✅	hardsigmoid		✅	-
✅	hardswish		✅	-
❌	hash_tensor		❌	-
❌	heaviside		❌	-
❌	hinge_embedding_loss		❌	-
❌	histc		❌	-
❌	histogram		❌	-
❌	histogramdd		❌	-
✅	hstack		❌	-
❌	hypot		✅	-
❌	i0	special_i0	✅	-
❌	igamma	special_gammainc	✅	-
❌	igammac	special_gammaincc	✅	-
✅	im2col		❌	-
✅	index_add		❌	-
✅	index_copy		❌	-
✅	index_fill		❌	-
❌	index_reduce		✅	-
✅	instance_norm		❌	-
❌	isclose		❌	-
✅	isfinite		❌	-
❌	istft		❌	-
❌	kaiser_window		❌	-
✅	kl_div		❌	-
❌	kthvalue		❌	-
❌	l1_loss		❌	-
✅	layer_norm		❌	-
❌	lcm		✅	-
❌	ldexp		✅	-
✅	lerp		❌	-
❌	lgamma		✅	-
❌	linalg__powsum		❌	-
❌	linalg_cond		❌	-
✅	linalg_cross		❌	-
✅	linalg_det	det	❌	-
❌	linalg_eig		❌	-
❌	linalg_eigh		❌	-
❌	linalg_eigvals		❌	-
❌	linalg_eigvalsh		❌	-
❌	linalg_householder_product	orgqr	❌	-
❌	linalg_inv	inverse	❌	-
❌	linalg_ldl_solve		❌	-
❌	linalg_lstsq		❌	-
❌	linalg_lu		❌	-
❌	linalg_lu_solve		❌	-
❌	linalg_matrix_exp	matrix_exp	❌	-
✅	linalg_matrix_norm		❌	-
❌	linalg_matrix_power	matrix_power	❌	-
❌	linalg_matrix_rank		❌	-
✅	linalg_norm		❌	-
❌	linalg_pinv		❌	-
❌	linalg_qr		❌	-
❌	linalg_slogdet		❌	-
❌	linalg_solve		❌	-
❌	linalg_solve_triangular		❌	-
❌	linalg_svd		❌	-
❌	linalg_tensorsolve		❌	-
✅	linalg_vector_norm		❌	-
✅	linear		❌	-
✅	linspace		❌	-
❌	log_normal		✅	-
✅	log_sigmoid		❌	-
✅	log_softmax	special_log_softmax	❌	-
❌	logcumsumexp		❌	-
✅	logdet		❌	-
✅	logit	special_logit	✅	-
❌	logspace		❌	-
✅	logsumexp	special_logsumexp	❌	-
✅	lstm		❌	-
✅	lstm_cell		❌	-
❌	lu_solve		❌	-
❌	lu_unpack		❌	-
❌	mH	adjoint	❌	-
❌	mT		❌	-
❌	margin_ranking_loss		❌	-
✅	masked_fill		✅	-
✅	masked_select		❌	-
✅	matmul	linalg_matmul	❌	-
❌	matrix_H		❌	-
✅	max_pool1d		❌	-
✅	max_pool1d_with_indices		❌	-
✅	max_pool2d		❌	-
✅	max_pool3d		❌	-
❌	median		❌	-
✅	meshgrid		❌	-
✅	mish		❌	-
❌	mode		❌	-
❌	modf		❌	-
✅	movedim	moveaxis	❌	-
✅	mse_loss		❌	-
❌	multi_margin_loss		❌	-
❌	multilabel_margin_loss		❌	-
✅	multinomial		❌	-
✅	mv		❌	-
❌	mvlgamma	special_multigammaln	❌	-
✅	nan_to_num		✅	-
❌	nanmedian		❌	-
❌	nanquantile		❌	-
✅	narrow		❌	-
❌	native_channel_shuffle		❌	-
❌	native_multi_head_self_attention		❌	-
✅	new_empty		❌	-
❌	new_empty_strided		❌	-
✅	new_full		❌	-
✅	new_ones		❌	-
✅	new_zeros		❌	-
❌	nextafter		✅	-
✅	nll_loss		❌	-
✅	nll_loss2d		❌	-
✅	nll_loss_nd		❌	-
✅	nonzero_numpy		❌	-
❌	nonzero_static		❌	-
✅	norm		❌	-
✅	normal		✅	-
❌	nuclear_norm		❌	-
✅	numel		❌	-
✅	numpy_T		❌	-
✅	one_hot		❌	-
✅	ones		❌	-
✅	ones_like		❌	-
❌	ormqr		❌	-
✅	outer	ger	❌	-
✅	pad		❌	-
❌	pad_sequence		❌	-
✅	pairwise_distance		❌	-
❌	pdist		❌	-
✅	pixel_shuffle		❌	-
✅	pixel_unshuffle		❌	-
❌	poisson		❌	-
❌	polar		❌	-
❌	polygamma	special_polygamma	✅	-
✅	prelu		❌	-
❌	put		✅	-
❌	qr		❌	-
❌	quantile		❌	-
❌	quantize		❌	-
❌	quantize_per_channel		❌	-
✅	quantize_per_tensor		❌	-
❌	quantize_per_tensor_dynamic		❌	-
❌	quantized_batch_norm		❌	-
❌	quantized_gru		❌	-
❌	quantized_lstm		❌	-
❌	quantized_max_pool1d		❌	-
❌	quantized_max_pool2d		❌	-
❌	quantized_max_pool3d		❌	-
✅	rand_like		❌	-
✅	randint		❌	-
✅	randint_like		❌	-
✅	randn_like		❌	-
❌	random		✅	-
❌	range		❌	-
✅	relu6		✅	-
❌	renorm		❌	-
✅	repeat_interleave		❌	-
✅	replication_pad1d		❌	-
✅	reshape		❌	-
✅	reshape_as		❌	-
❌	resize_as_sparse_		❌	-
✅	resolve_conj		❌	-
✅	resolve_neg		❌	-
✅	rnn_relu		❌	-
❌	rnn_relu_cell		❌	-
✅	rnn_tanh		❌	-
❌	rnn_tanh_cell		❌	-
✅	roll		❌	-
❌	rot90		❌	-
✅	rrelu		❌	-
❌	rrelu_with_noise		✅	-
✅	rsub		❌	-
✅	scaled_dot_product_attention		❌	-
❌	searchsorted		❌	-
❌	segment_reduce		❌	-
✅	selu		✅	-
❌	set_		❌	-
❌	signbit		❌	-
✅	silu		✅	-
❌	sinc	special_sinc	❌	-
✅	size		❌	-
❌	smooth_l1_loss		❌	-
❌	soft_margin_loss		❌	-
✅	softmax	special_softmax	❌	-
✅	softplus		❌	-
✅	softshrink		❌	-
❌	special_airy_ai		❌	-
❌	special_bessel_j0		❌	-
❌	special_bessel_j1		❌	-
❌	special_bessel_y0		❌	-
❌	special_bessel_y1		❌	-
❌	special_chebyshev_polynomial_t		❌	-
❌	special_chebyshev_polynomial_u		❌	-
❌	special_chebyshev_polynomial_v		❌	-
❌	special_chebyshev_polynomial_w		❌	-
❌	special_erfcx		❌	-
❌	special_hermite_polynomial_h		❌	-
❌	special_hermite_polynomial_he		❌	-
❌	special_i0e		❌	-
❌	special_i1		❌	-
❌	special_i1e		❌	-
❌	special_laguerre_polynomial_l		❌	-
❌	special_legendre_polynomial_p		❌	-
❌	special_log_ndtr		❌	-
❌	special_modified_bessel_i0		❌	-
❌	special_modified_bessel_i1		❌	-
❌	special_modified_bessel_k0		❌	-
❌	special_modified_bessel_k1		❌	-
❌	special_ndtri		❌	-
❌	special_scaled_modified_bessel_k0		❌	-
❌	special_scaled_modified_bessel_k1		❌	-
❌	special_shifted_chebyshev_polynomial_t		❌	-
❌	special_shifted_chebyshev_polynomial_u		❌	-
❌	special_shifted_chebyshev_polynomial_v		❌	-
❌	special_shifted_chebyshev_polynomial_w		❌	-
❌	special_spherical_bessel_j0		❌	-
❌	special_zeta		❌	-
✅	split		❌	-
✅	square		✅	-
✅	stack		❌	-
✅	std		❌	-
✅	std_mean		❌	-
✅	stft		❌	-
✅	t		✅	-
✅	take		❌	-
❌	tanhshrink		❌	-
✅	tensor_split		❌	-
✅	tensordot		❌	-
✅	threshold		❌	-
✅	tile		❌	-
✅	to		❌	-
❌	trace		❌	-
✅	transpose	swapaxes, swapdims	✅	-
❌	triangular_solve		❌	-
✅	tril		❌	-
❌	tril_indices		❌	-
❌	triplet_margin_loss		❌	-
✅	triu		❌	-
❌	triu_indices		❌	-
✅	type_as		❌	-
✅	unbind		❌	-
✅	unflatten		❌	-
❌	unflatten_dense_tensors		❌	-
✅	unfold		❌	-
❌	uniform		✅	-
❌	unique_consecutive		❌	-
✅	unique_dim		❌	-
❌	unique_dim_consecutive		❌	-
✅	unsafe_chunk		❌	-
✅	unsafe_split		❌	-
✅	unsafe_split_with_sizes		❌	-
❌	upsample		❌	-
✅	upsample_bicubic2d		❌	-
✅	upsample_linear1d		❌	-
✅	upsample_nearest1d		❌	-
✅	upsample_nearest3d		❌	-
✅	upsample_trilinear3d		❌	-
❌	vander		❌	-
✅	var_mean		❌	-
✅	view_as		❌	-
❌	view_as_complex		❌	-
❌	view_as_real		❌	-
✅	vstack	row_stack	❌	-
❌	wrapped_linear_prepack		❌	-
❌	wrapped_quantized_linear_prepacked		❌	-
❌	xlogy	special_xlogy	❌	-
✅	zero		✅	-
✅	zeros		❌	-
✅	zeros_like		❌	-

Appendix: identifiers excluded from this page

These names are filtered out of the support tables above because they cannot surface in an inference JIT trace. Each group has a documented rationale; click any to expand the full member list.

Python builtin / scripting false positives (52 names)

capitalize, center, chr, clear, dict, endswith, expandtabs, extend, find, format, get, getelem, hash, hex, isalnum, isalpha, isdecimal, isdigit, isidentifier, islower, isnumeric, isprintable, isspace, istitle, isupper, items, join, keys, ljust, lower, lstrip, oct, ord, partition, popitem, rfind, rindex, rjust, rpartition, rsplit, rstrip, setdefault, sorted, splitlines, startswith, strip, swapcase, title, update, upper, values, zfill

Test harness / placeholder identifiers (13 names)

confirmed_by_owner, foo, mathremainder, percentFormat, pointwise_placeholder, symbolic_b, test, test_symbol, test_vartype, test_vartype2, unknown, view_expand_placeholder, your_op

Distributed / RPC primitives (14 names)

all_gather_into_tensor, all_reduce, fork, get_gradients, is_owner, local_value, owner, owner_name, reduce_scatter_tensor, to_here, wait, wait_tensor, warn, warns

Backend-specific dispatch shims (cudnn / miopen / mkldnn / mps) (26 names)

cudnn_affine_grid_generator, cudnn_batch_norm, cudnn_convolution, cudnn_convolution_add_relu, cudnn_convolution_relu, cudnn_convolution_transpose, cudnn_grid_sampler, cudnn_is_acceptable, miopen_batch_norm, miopen_convolution, miopen_convolution_add_relu, miopen_convolution_relu, miopen_convolution_transpose, miopen_ctc_loss, miopen_depthwise_convolution, miopen_rnn, mkldnn_adaptive_avg_pool2d, mkldnn_convolution, mkldnn_linear, mkldnn_max_pool2d, mkldnn_max_pool3d, mkldnn_reorder_conv2d_weight, mkldnn_reorder_conv3d_weight, mkldnn_rnn_layer, mps_linear, to_mkldnn

Tensor metadata accessors (return a Python value, not a tensor) (60 names)

can_cast, data, dense_dim, device, dtype, element_size, enable_grad, get_autocast_dtype, get_device, get_pool_ceil_padding, grad, has_torch_function, iinfo, initial_seed, int_repr, is_autocast_cpu_enabled, is_autocast_enabled, is_coalesced, is_complex, is_conj, is_contiguous, is_cuda, is_grad_enabled, is_leaf, is_non_overlapping_and_dense, is_nonzero, is_pinned, is_same_size, is_scripting, is_set_to, is_signed, is_strides_like_format, manual_seed, node, op, op_name, output_nr, pin_memory, promote_types, q_per_channel_axis, q_per_channel_scales, q_per_channel_zero_points, q_scale, q_zero_point, qscheme, record_stream, refine_names, rename, requires_grad_, result_type, retain_grad, retains_grad, save, seed, set_data, set_grad_enabled, set_source_Tensor_storage_offset, sparse_dim, storage_offset, stride

Named-tensor API (names stripped before JIT trace) (3 names)

align_as, align_tensors, align_to

Sparse-tensor machinery (NNEF / tract are dense-only) (25 names)

ccol_indices, ccol_indices_copy, coalesce, col_indices, col_indices_copy, copy_sparse_to_sparse, crow_indices, crow_indices_copy, hspmm, indices, indices_copy, nested_to_padded_tensor, row_indices, row_indices_copy, smm, sparse_compressed_tensor, sparse_coo_tensor, sparse_mask, sparse_resize, sparse_resize_and_clear, sparse_sampled_addmm, sspaddmm, to_dense, to_padded_tensor, values_copy

Functionalization *_copy / *_scatter variants (20 names)

alias_copy, as_strided_copy, as_strided_scatter, detach_copy, diagonal_copy, diagonal_scatter, lift, lift_fresh, lift_fresh_copy, narrow_copy, permute_copy, select_copy, slice_copy, slice_inverse, split_copy, unbind_copy, unfold_copy, view_as_complex_copy, view_as_real_copy, view_copy

Batch-norm training / backward-only internals (8 names)

batch_norm_elemt, batch_norm_gather_stats, batch_norm_gather_stats_with_counts, batch_norm_stats, batch_norm_update_stats, native_batch_norm, native_norm, norm_except_dim

linalg_*_ex paired-output variants + deprecated linalg wrappers (12 names)

eig, linalg_cholesky_ex, linalg_inv_ex, linalg_ldl_factor_ex, linalg_lu_factor_ex, lstsq, matrix_rank, pinv, pinverse, solve, svd, symeig

QAT fake_quantize_* training-only ops (6 names)

choose_qparams_optimized, fake_quantize_per_channel_affine, fake_quantize_per_channel_affine_cachemask, fake_quantize_per_tensor_affine, fake_quantize_per_tensor_affine_cachemask, fused_moving_avg_obs_fake_quant

slow_conv* / thnn_conv* dispatcher-fallback kernels (8 names)

conv_depthwise3d, slow_conv3d, slow_conv3d_forward, slow_conv_dilated2d, slow_conv_dilated3d, slow_conv_transpose2d, slow_conv_transpose3d, thnn_conv2d

Python / TorchScript scalar builtins (extra) (25 names)

append, bin, count, cpu, cuda, degrees, dim, divmod, equal, fabs, factorial, insert, is_floating_point, item, len, list, list_with_default, neq, pop, radians, remove, replace, reverse, str, tensor

Inplace storage / metadata mutators stripped by JIT (9 names)

fill_diagonal_, float_power_, rename_, resize, resize_as_, resize_as_sparse, set, sparse_resize_, sparse_resize_and_clear_

Backward / dynamo-autograd internals (10 names)

embedding_renorm, from_file, glu_jvp, log_sigmoid_forward, multilabel_margin_loss_forward, nll_loss_forward, normal_functional, rowwise_prune, rrelu_with_noise_functional, sum_to