torch-rb 0.15.0 → 0.17.0

data/codegen/native_functions.yaml

@@ -134,7 +134,7 @@
   autogen: _new_zeros_with_same_feature_meta.out
 
 # This function compares the storage numel of self with that of other, where
-# storage numel is cumputed as: `other.storage().nbytes() / other.itemsize()`.
+# storage numel is computed as: `other.storage().nbytes() / other.itemsize()`.
 # We create this function for composite compliance purposes. The batching rule
 # always returns true because vmapped as_strided does not support accessing
 # storage locations not indexable by the input tensor.
@@ -175,12 +175,24 @@
     CPU: _assert_async_msg_cpu
     CUDA: _assert_async_msg_cuda
 
+- func: _assert_scalar(Scalar self, str assert_msg) -> ()
+  dispatch:
+    CompositeExplicitAutograd: _assert_scalar
+
+- func: _functional_assert_scalar(Scalar self, str assert_msg, Tensor dep_token) -> Tensor
+  dispatch:
+    CompositeExplicitAutograd: _functional_assert_scalar
+
 - func: _functional_assert_async.msg(Tensor self, str assert_msg, Tensor dep_token) -> Tensor
   dispatch:
     CPU: _functional_assert_async_msg_cpu
 
 - func: _assert_tensor_metadata(Tensor a, SymInt[]? size=None, SymInt[]? stride=None, ScalarType? dtype=None) -> ()
 
+- func: _print(str s) -> ()
+  dispatch:
+    CompositeExplicitAutograd: _print
+
 - func: sym_constrain_range(Scalar size, *, int? min=None, int? max=None) -> ()
   dispatch:
     CompositeExplicitAutograd: sym_constrain_range
@@ -470,6 +482,7 @@
 - func: conj_physical.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
   dispatch:
     CPU, CUDA: conj_physical_out
+    MPS: conj_physical_out_mps
     SparseCPU, SparseCUDA: conj_physical_out_sparse
     SparseCsrCPU, SparseCsrCUDA: conj_physical_sparse_csr_out
   tags: pointwise
@@ -536,8 +549,8 @@
   structured_delegate: add.out
   variants: function, method
   dispatch:
-    SparseCPU, SparseCUDA: add_sparse
-    SparseCsrCPU, SparseCsrCUDA: add_sparse_csr
+    SparseCPU, SparseCUDA, SparseMeta: add_sparse
+    SparseCsrCPU, SparseCsrCUDA, SparseCsrMeta: add_sparse_csr
     MkldnnCPU: mkldnn_add
     ZeroTensor: add_zerotensor
     NestedTensorCPU, NestedTensorCUDA: NestedTensor_add_Tensor
@@ -548,8 +561,8 @@
   variants: method
   structured_delegate: add.out
   dispatch:
-    SparseCPU, SparseCUDA: add_sparse_
-    SparseCsrCPU, SparseCsrCUDA: add_sparse_csr_
+    SparseCPU, SparseCUDA, SparseMeta: add_sparse_
+    SparseCsrCPU, SparseCsrCUDA, SparseCsrMeta: add_sparse_csr_
     MkldnnCPU: mkldnn_add_
     NestedTensorCPU, NestedTensorCUDA: NestedTensor_add__Tensor
   tags: pointwise
@@ -562,10 +575,10 @@
     Generic: add (AllAndComplex, BFloat16, Half, ComplexHalf)
     ScalarOnly: add (Bool)
   dispatch:
-    SparseCPU: add_out_sparse_cpu
+    SparseCPU, SparseMeta: add_out_sparse_cpu
     SparseCUDA: add_out_sparse_cuda
-    SparseCsrCPU: add_out_sparse_csr_cpu
-    SparseCsrCUDA: add_out_sparse_csr_cuda
+    SparseCsrCPU, SparseCsrMeta: add_out_sparse_compressed_cpu
+    SparseCsrCUDA: add_out_sparse_compressed_cuda
     MkldnnCPU: mkldnn_add_out
     MPS: add_out_mps
   tags: pointwise
@@ -763,7 +776,7 @@
   dispatch:
     CompositeExplicitAutograd: arange
 
-# This operator should be named `aragne.start_out` if following the naming convention. However that
+# This operator should be named `arange.start_out` if following the naming convention. However that
 # name is already taken. Disabled because of CI job failures.
 # FIXME: enable this
 #- func: arange.start_out_(Scalar start, Scalar end, *, Tensor(a!) out) -> Tensor(a!)
@@ -1220,6 +1233,13 @@
     CompositeExplicitAutograd: copysign_out
   tags: pointwise
 
+- func: _lazy_clone(Tensor self) -> Tensor
+  # Like clone, but the copy takes place lazily, only if either the
+  # input or the output are written.
+  variants: function, method
+  dispatch:
+    CompositeExplicitAutograd: _lazy_clone
+
 - func: logical_not(Tensor self) -> Tensor
   device_check: NoCheck   # TensorIterator
   variants: function, method
@@ -1621,6 +1641,7 @@
 - func: complex.out(Tensor real, Tensor imag, *, Tensor(a!) out) -> Tensor(a!)
   dispatch:
     CPU, CUDA: complex_out
+    MPS: complex_out_mps
 
 - func: polar(Tensor abs, Tensor angle) -> Tensor
   variants: function
@@ -1729,6 +1750,7 @@
 - func: copy(Tensor self, Tensor src, bool non_blocking=False) -> Tensor
   variants: function
   dispatch:
+    Meta: copy_meta
     CompositeExplicitAutogradNonFunctional: copy
   tags: core
 
@@ -1847,7 +1869,10 @@
 - func: cudnn_convolution(Tensor self, Tensor weight, SymInt[] padding, SymInt[] stride, SymInt[] dilation, SymInt groups, bool benchmark, bool deterministic, bool allow_tf32) -> Tensor
   dispatch:
     CUDA: cudnn_convolution
-  autogen: cudnn_convolution.out
+
+- func: cudnn_convolution.out(Tensor self, Tensor weight, SymInt[] padding, SymInt[] stride, SymInt[] dilation, SymInt groups, bool benchmark, bool deterministic, bool allow_tf32, *, Tensor(a!) out) -> Tensor(a!)
+  dispatch:
+    CUDA: cudnn_convolution_out
 
 - func: cudnn_convolution_transpose(Tensor self, Tensor weight, SymInt[] padding, SymInt[] output_padding, SymInt[] stride, SymInt[] dilation, SymInt groups, bool benchmark, bool deterministic, bool allow_tf32) -> Tensor
   dispatch:
@@ -2346,7 +2371,7 @@
     Meta: empty_meta_symint
     MkldnnCPU: empty_mkldnn
     SparseCPU, SparseCUDA, SparseMeta: empty_sparse
-    SparseCsrCPU, SparseCsrCUDA: empty_sparse_compressed
+    SparseCsrCPU, SparseCsrCUDA, SparseCsrMeta: empty_sparse_compressed
     QuantizedCPU, QuantizedCUDA, QuantizedMeta: empty_unknown_quantized
   tags: core
 
@@ -2452,7 +2477,7 @@
     CompositeExplicitAutograd: empty_like
     QuantizedCPU, QuantizedCUDA: empty_like_quantized
     SparseCPU, SparseCUDA, SparseMeta: empty_like_sparse_coo
-    SparseCsrCPU, SparseCsrCUDA: empty_like_sparse_csr
+    SparseCsrCPU, SparseCsrCUDA, SparseCsrMeta: empty_like_sparse_csr
     NestedTensorCPU, NestedTensorCUDA: empty_like_nested
   autogen: empty_like.out
 
@@ -2954,12 +2979,14 @@
   dispatch:
     CPU: _fft_r2c_mkl
     CUDA: _fft_r2c_cufft
+    MPS: _fft_r2c_mps
 
 - func: _fft_r2c.out(Tensor self, int[] dim, int normalization, bool onesided, *, Tensor(a!) out) -> Tensor(a!)
   variants: function
   dispatch:
     CPU: _fft_r2c_mkl_out
     CUDA: _fft_r2c_cufft_out
+    MPS: _fft_r2c_mps_out
 
 # Complex to real inverse FFT
 - func: _fft_c2r(Tensor self, int[] dim, int normalization, SymInt last_dim_size) -> Tensor
@@ -2967,12 +2994,14 @@
   dispatch:
     CPU: _fft_c2r_mkl
     CUDA: _fft_c2r_cufft
+    MPS: _fft_c2r_mps
 
 - func: _fft_c2r.out(Tensor self, int[] dim, int normalization, SymInt last_dim_size, *, Tensor(a!) out) -> Tensor(a!)
   variants: function
   dispatch:
     CPU: _fft_c2r_mkl_out
     CUDA: _fft_c2r_cufft_out
+    MPS: _fft_c2r_mps_out
 
 # Standard complex to complex FFT (forward or backward)
 - func: _fft_c2c(Tensor self, SymInt[] dim, int normalization, bool forward) -> Tensor
@@ -2980,12 +3009,14 @@
   dispatch:
     CPU: _fft_c2c_mkl
     CUDA: _fft_c2c_cufft
+    MPS: _fft_c2c_mps
 
 - func: _fft_c2c.out(Tensor self, SymInt[] dim, int normalization, bool forward, *, Tensor(a!) out) -> Tensor(a!)
   variants: function
   dispatch:
     CPU: _fft_c2c_mkl_out
     CUDA: _fft_c2c_cufft_out
+    MPS: _fft_c2c_mps_out
 
 - func: _validate_compressed_sparse_indices(bool is_crow, Tensor compressed_idx, Tensor plain_idx, int cdim, int dim, int nnz) -> ()
   device_check: NoCheck
@@ -3097,6 +3128,7 @@
   structured: True
   dispatch:
     CPU, CUDA: isin_Tensor_Tensor_out
+    MPS: isin_Tensor_Tensor_out_mps
 
 - func: isin.Tensor_Tensor(Tensor elements, Tensor test_elements, *, bool assume_unique=False, bool invert=False) -> Tensor
   variants: function
@@ -3238,6 +3270,8 @@
   autogen: native_layer_norm_backward.out
   tags: core
 
+- func: rms_norm(Tensor input, int[] normalized_shape, Tensor? weight=None, float? eps=None) -> Tensor
+
 - func: nan_to_num(Tensor self, float? nan=None, float? posinf=None, float? neginf=None) -> Tensor
   variants: function, method
   dispatch:
@@ -3302,14 +3336,39 @@
   dispatch:
     CUDA: _cslt_compress
 
-- func: _cslt_sparse_mm(Tensor compressed_A, Tensor dense_B, Tensor? bias=None, Tensor? alpha=None, ScalarType? out_dtype=None, bool transpose_result=False) -> Tensor
+- func: _cslt_sparse_mm(Tensor compressed_A, Tensor dense_B, Tensor? bias=None, Tensor? alpha=None, ScalarType? out_dtype=None, bool transpose_result=False, int alg_id=0) -> Tensor
   dispatch:
     CUDA: _cslt_sparse_mm
 
-- func: _sparse_semi_structured_linear(Tensor input, Tensor weight, Tensor meta, *, Tensor? bias=None, str? activation=None) -> Tensor
+- func: _cslt_sparse_mm_search(Tensor compressed_A, Tensor dense_B, Tensor? bias=None, Tensor? alpha=None, ScalarType? out_dtype=None, bool transpose_result=False) -> int
+  dispatch:
+    CUDA: _cslt_sparse_mm_search
+
+- func: _sparse_semi_structured_tile(Tensor input, str algorithm="", bool use_cutlass=True) -> (Tensor, Tensor, Tensor, Tensor, Tensor)
+  dispatch:
+    CUDA: _sparse_semi_structured_tile
+
+- func: _sparse_semi_structured_apply(Tensor input, Tensor thread_masks) -> (Tensor, Tensor)
+  dispatch:
+    CUDA: _sparse_semi_structured_apply
+
+- func: _sparse_semi_structured_apply_dense(Tensor input, Tensor thread_masks) -> Tensor
+  dispatch:
+    CUDA: _sparse_semi_structured_apply_dense
+
+# DEPRECATED: Use torch.__sparse_semi_structured_mm/torch._sparse_semi_structured_addmm instead
+- func: _sparse_semi_structured_linear(Tensor input, Tensor weight, Tensor meta, *, Tensor? bias=None, str? activation=None, ScalarType? out_dtype=None) -> Tensor
   dispatch:
     CUDA: _sparse_semi_structured_linear
 
+- func: _sparse_semi_structured_mm(Tensor mat1, Tensor mat1_meta, Tensor mat2, *, ScalarType? out_dtype=None) -> Tensor
+  dispatch:
+    CUDA: _sparse_semi_structured_mm
+
+- func: _sparse_semi_structured_addmm(Tensor input, Tensor mat1, Tensor mat1_meta, Tensor mat2, *, Scalar alpha=1, Scalar beta=1, ScalarType? out_dtype=None) -> Tensor
+  dispatch:
+    CUDA: _sparse_semi_structured_addmm
+
 - func: _mixed_dtypes_linear(Tensor input, Tensor weight, Tensor scale, *, Tensor? bias=None, str? activation=None) -> Tensor
   dispatch:
     CUDA: _mixed_dtypes_linear
@@ -4050,20 +4109,30 @@
 
 - func: _int_mm(Tensor self, Tensor mat2) -> Tensor
   dispatch:
+    CPU: _int_mm_cpu
     CUDA: _int_mm_cuda
 
 - func: _int_mm.out(Tensor self, Tensor mat2, *, Tensor(a!) out) -> Tensor(a!)
   dispatch:
+    CPU: _int_mm_out_cpu
     CUDA: _int_mm_out_cuda
 
 - func: _convert_weight_to_int4pack(Tensor self, int innerKTiles) -> Tensor
   dispatch:
+    CPU: _convert_weight_to_int4pack_cpu
     CUDA: _convert_weight_to_int4pack_cuda
 
 - func: _weight_int4pack_mm(Tensor self, Tensor mat2, int qGroupSize, Tensor qScaleAndZeros) -> Tensor
   dispatch:
+    CPU: _weight_int4pack_mm_cpu
+    MPS: _weight_int4pack_mm_mps
     CUDA: _weight_int4pack_mm_cuda
 
+- func: _weight_int8pack_mm(Tensor self, Tensor mat2, Tensor scales) -> Tensor
+  dispatch:
+    CPU: _weight_int8pack_mm_cpu
+    MPS: _weight_int8pack_mm_mps
+
 - func: _sparse_mm(Tensor sparse, Tensor dense) -> Tensor
   python_module: sparse
 
@@ -4439,7 +4508,6 @@
     MPS: pixel_shuffle_mps
     CompositeExplicitAutogradNonFunctional: math_pixel_shuffle
   autogen: pixel_shuffle.out
-  tags: core
 
 - func: pixel_unshuffle(Tensor self, int downscale_factor) -> Tensor
   dispatch:
@@ -4810,7 +4878,7 @@
   device_guard: False
   dispatch:
     CompositeImplicitAutograd: reshape_symint
-    CompositeImplicitAutogradNestedTensor: reshape_nested
+    CompositeImplicitAutogradNestedTensor: reshape_nested_symint
 
 - func: _reshape_copy(Tensor self, SymInt[] size) -> Tensor
   variants: function
@@ -4969,6 +5037,7 @@
   device_check: NoCheck   # TensorIterator
   python_module: nn
   dispatch:
+    QuantizedCPU: gelu_quantized_cpu_
     NestedTensorCPU, NestedTensorCUDA: NestedTensor_gelu_
 
 - func: gelu(Tensor self, *, str approximate='none') -> Tensor
@@ -5356,6 +5425,21 @@
     CompositeExplicitAutograd: slice_backward
   autogen: slice_backward.out
 
+# NB: This op exists to back the implementation of reverse view_funcs for various views (chunk,
+# slice.Tensor, split_with_sizes, et al.). Currently, these are only used during fake-ification
+# of PT2 graph input subclass instances that are views. This means:
+# * This op shouldn't really show up in eager mode (so e.g. XLA shouldn't have to implement it)
+# * This op shouldn't show up in a PT2 graph (so a PT2 backend shouldn't have to implement it)
+# * A subclass will have to implement this to work in PT2 if a subclass view is used as a graph
+#   input AND the view utilizes this op in its inverse. The idea is that slice_inverse() is
+#   easier to implement for a subclass than as_strided()
+- func: slice_inverse(Tensor(a) self, Tensor src, int dim=0, SymInt? start=None, SymInt? end=None, SymInt step=1) -> Tensor(a)
+  variants: function, method
+  device_check: NoCheck
+  device_guard: False
+  dispatch:
+    CompositeExplicitAutograd: slice_inverse_symint
+
 - func: slice_scatter(Tensor self, Tensor src, int dim=0, SymInt? start=None, SymInt? end=None, SymInt step=1) -> Tensor
   variants: function, method
   device_check: NoCheck
@@ -5363,7 +5447,7 @@
   dispatch:
     CompositeExplicitAutogradNonFunctional: slice_scatter
   autogen: slice_scatter.out
-  tags: core
+  tags: [core, view_copy]
 
 - func: select_scatter(Tensor self, Tensor src, int dim, SymInt index) -> Tensor
   variants: function, method
@@ -5562,6 +5646,16 @@
     SparseCPU: _sspaddmm_out_cpu
     SparseCUDA: _sspaddmm_out_cuda
 
+- func: _chunk_cat(Tensor[] tensors, int dim, int num_chunks) -> Tensor
+  dispatch:
+    CompositeExplicitAutograd: _chunk_cat
+    CUDA: _chunk_cat_cuda
+
+- func: _chunk_cat.out(Tensor[] tensors, int dim, int num_chunks, *, Tensor(a!) out) -> Tensor(a!)
+  dispatch:
+    CompositeExplicitAutograd: _chunk_cat_out
+    CUDA: _chunk_cat_out_cuda
+
 - func: stack(Tensor[] tensors, int dim=0) -> Tensor
   dispatch:
     CompositeExplicitAutograd: stack
@@ -5626,8 +5720,8 @@
   variants: function, method
   dispatch:
     CompositeExplicitAutograd: sum
-    SparseCPU, SparseCUDA: sum_coo
-    SparseCsrCPU, SparseCsrCUDA: sum_csr
+    SparseCPU, SparseCUDA, SparseMeta: sum_coo
+    SparseCsrCPU, SparseCsrCUDA, SparseCsrMeta: sum_csr
   autogen: sum.out
 
 - func: sum.dim_IntList(Tensor self, int[1]? dim, bool keepdim=False, *, ScalarType? dtype=None) -> Tensor
@@ -5753,6 +5847,7 @@
   variants: function
   dispatch:
     CPU, CUDA: std_mean
+    MPS: std_mean_mps
   autogen: std_mean.correction_out
 
 - func: std_mean.names_dim(Tensor self, Dimname[1] dim, bool unbiased=True, bool keepdim=False) -> (Tensor, Tensor)
@@ -6008,7 +6103,6 @@
     CPU, MPS: roll
     CUDA: roll_cuda
   autogen: roll.out
-  tags: core
 
 # default int[] value [0,1] should not add space after comma, since codegen parser uses ', ' to split args
 
@@ -6091,6 +6185,58 @@
     CompositeExplicitAutogradNonFunctional: _nested_view_from_buffer_copy
   autogen: _nested_view_from_buffer_copy.out
 
+- func: _nested_view_from_jagged(Tensor(a) self, Tensor offsets, Tensor dummy, Tensor? lengths=None, int ragged_idx=1) -> Tensor(a)
+  variants: function
+  device_check: NoCheck
+  dispatch: {}
+
+- func: _nested_view_from_jagged_copy(Tensor self, Tensor offsets, Tensor dummy, Tensor? lengths=None, int ragged_idx=1) -> Tensor
+  variants: function
+  device_check: NoCheck
+  tags: view_copy
+  dispatch:
+    CompositeExplicitAutogradNonFunctional: _nested_view_from_jagged_copy
+  autogen: _nested_view_from_jagged_copy.out
+
+- func: _nested_get_values(Tensor(a) self) -> Tensor(a)
+  variants: function
+  device_check: NoCheck
+  dispatch: {}
+
+- func: _nested_get_values_copy(Tensor self) -> Tensor
+  variants: function
+  device_check: NoCheck
+  tags: view_copy
+  dispatch:
+    CompositeExplicitAutogradNonFunctional: _nested_get_values_copy
+  autogen: _nested_get_values_copy.out
+
+- func: _nested_get_offsets(Tensor self) -> Tensor
+  variants: function
+  device_check: NoCheck
+  dispatch: {}
+
+# returns undefined Tensor if no lengths present
+- func: _nested_get_lengths(Tensor self) -> Tensor
+  variants: function
+  device_check: NoCheck
+  dispatch: {}
+
+- func: _nested_get_ragged_idx(Tensor self) -> int
+  variants: function
+  device_check: NoCheck
+  dispatch: {}
+
+- func: _nested_get_jagged_dummy(Tensor any) -> Tensor
+  category_override: dummy
+  dispatch: {}
+
+- func: _nested_compute_contiguous_strides_offsets(Tensor nested_size) -> (Tensor, Tensor)
+  variants: function
+  device_check: NoCheck
+  dispatch:
+    CPU, CUDA: _nested_compute_contiguous_strides_offsets
+
 - func: _trilinear(Tensor i1, Tensor i2, Tensor i3, int[] expand1, int[] expand2, int[] expand3, int[] sumdim, int unroll_dim=1) -> Tensor
   dispatch:
     # calls unsqueeze
@@ -6275,6 +6421,7 @@
   variants: function
   dispatch:
     CPU, CUDA: var_mean
+    MPS: var_mean_mps
   autogen: var_mean.correction_out
 
 - func: var_mean.names_dim(Tensor self, Dimname[1] dim, bool unbiased=True, bool keepdim=False) -> (Tensor, Tensor)
@@ -6295,15 +6442,13 @@
   device_check: NoCheck   # TensorIterator
   variants: function, method
   dispatch:
-    CPU, CUDA: where
-    MPS: where_mps
+    CPU, CUDA, MPS: where
   tags: [core, pointwise]
 
 - func: where.self_out(Tensor condition, Tensor self, Tensor other, *, Tensor(a!) out) -> Tensor(a!)
   device_check: NoCheck   # TensorIterator
   dispatch:
-    CPU, CUDA: where_self_out
-    MPS: where_self_out_mps
+    CPU, CUDA, MPS: where_self_out
 
 - func: where.ScalarSelf(Tensor condition, Scalar self, Tensor other) -> Tensor
   variants: function
@@ -6357,7 +6502,7 @@
     CPU: _efficientzerotensor
     CUDA: _efficientzerotensor_cuda
     MPS: _efficientzerotensor_mps
-    Meta: _efficientzerotensor_meta
+    Meta: _efficientzerotensor_meta_symint
   autogen: _efficientzerotensor.out
 
 - func: zeros(SymInt[] size, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=None) -> Tensor
@@ -6434,6 +6579,32 @@
     SparseCPU, SparseCUDA: norm_sparse
   autogen: native_norm.ScalarOpt_dim_dtype_out
 
+- func: _batch_norm_with_update(Tensor input, Tensor? weight, Tensor? bias, Tensor(a!) running_mean, Tensor(b!) running_var, float momentum, float eps) -> (Tensor, Tensor, Tensor, Tensor)
+  dispatch:
+    CPU: _batch_norm_with_update_cpu
+    CUDA: _batch_norm_with_update_cuda
+    MPS: _batch_norm_with_update_mps
+    MkldnnCPU: _batch_norm_with_update_mkldnn
+  autogen: _batch_norm_with_update_functional
+
+- func: _batch_norm_with_update.out(Tensor input, Tensor? weight, Tensor? bias, Tensor(a!) running_mean, Tensor(b!) running_var, float momentum, float eps, *, Tensor(d!) out, Tensor(e!) save_mean, Tensor(f!) save_invstd, Tensor(g!) reserve) -> (Tensor(d!), Tensor(e!), Tensor(f!), Tensor(g!))
+  dispatch:
+    CPU: _batch_norm_with_update_cpu_out
+    CUDA: _batch_norm_with_update_cuda_out
+    MPS: _batch_norm_with_update_mps_out
+
+- func: _batch_norm_no_update(Tensor input, Tensor? weight, Tensor? bias, Tensor? running_mean, Tensor? running_var, float momentum, float eps) -> (Tensor, Tensor, Tensor, Tensor)
+  dispatch:
+    CompositeExplicitAutograd: _batch_norm_no_update
+  autogen: _batch_norm_no_update.out
+
+- func: batch_norm_backward(Tensor grad_out, Tensor input, Tensor weight, Tensor? running_mean, Tensor? running_var, Tensor? save_mean, Tensor? save_var, bool update, float eps, bool[3] output_mask, Tensor reserve) -> (Tensor, Tensor, Tensor)
+  dispatch:
+    CPU: _new_batch_norm_backward_cpu
+    CUDA: _new_batch_norm_backward_cuda
+    MPS: _new_batch_norm_backward_mps
+    MkldnnCPU: _new_batch_norm_backward_mkldnn
+
 # TODO: reduce signatures down to one when optional args is available
 - func: _sparse_sum(Tensor self) -> Tensor
 
@@ -6644,7 +6815,7 @@
     MPS: zero_mps_
     Meta: zero_meta_
     SparseCPU, SparseCUDA, SparseMeta: zero_sparse_
-    SparseCsrCPU, SparseCsrCUDA: zero_sparse_csr_
+    SparseCsrCPU, SparseCsrCUDA, SparseCsrMeta: zero_sparse_csr_
     MkldnnCPU: mkldnn_zero_
     NestedTensorCPU, NestedTensorCUDA: zero_nested_
   autogen: zero, zero.out
@@ -6934,7 +7105,11 @@
 # FIXME: would be nicer if TensorOptions was optional based; not adding default arguments for options given
 # the default would never make sense.
 
-- func: sparse_compressed_tensor.comp_plain_value_size(Tensor compressed_indices, Tensor plain_indices, Tensor values, int[] size, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=False) -> Tensor
+- func: _sparse_compressed_tensor_with_dims(int nnz, int dense_dim, int[] size, int[] blocksize, ScalarType index_dtype, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=False) -> Tensor
+  dispatch:
+    CompositeExplicitAutograd: sparse_compressed_tensor_with_dims
+
+- func: sparse_compressed_tensor.comp_plain_value_size(Tensor compressed_indices, Tensor plain_indices, Tensor values, SymInt[] size, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=False) -> Tensor
   dispatch:
     CompositeExplicitAutograd: sparse_compressed_tensor
 
@@ -6951,7 +7126,10 @@
 - func: sparse_bsr_tensor.crow_col_value(Tensor crow_indices, Tensor col_indices, Tensor values, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=False) -> Tensor
 - func: sparse_bsc_tensor.ccol_row_value(Tensor ccol_indices, Tensor row_indices, Tensor values, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=False) -> Tensor
 
-- func: _sparse_compressed_tensor_unsafe(Tensor compressed_indices, Tensor plain_indices, Tensor values, int[] size, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=None) -> Tensor
+- func: _sparse_compressed_tensor_unsafe(Tensor compressed_indices, Tensor plain_indices, Tensor values, SymInt[] size, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=None) -> Tensor
+  dispatch:
+    CompositeImplicitAutograd: _sparse_compressed_tensor_unsafe_symint
+
 - func: _sparse_csr_tensor_unsafe(Tensor crow_indices, Tensor col_indices, Tensor values, int[] size, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=None) -> Tensor
 - func: _sparse_csc_tensor_unsafe(Tensor ccol_indices, Tensor row_indices, Tensor values, int[] size, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=None) -> Tensor
 - func: _sparse_bsr_tensor_unsafe(Tensor crow_indices, Tensor col_indices, Tensor values, int[] size, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=None) -> Tensor
@@ -7035,9 +7213,9 @@
 - func: sparse_dim(Tensor self) -> int
   variants: method
   dispatch:
-    CPU, CUDA: sparse_dim_strided
     SparseCPU, SparseCUDA, SparseMeta: sparse_dim_sparse
-    SparseCsrCPU, SparseCsrCUDA: sparse_dim_sparse_csr
+    SparseCsrCPU, SparseCsrCUDA, SparseCsrMeta: sparse_dim_sparse_csr
+    CompositeExplicitAutograd: sparse_dim_default
   device_check: NoCheck
   device_guard: False
 
@@ -7052,9 +7230,9 @@
 - func: dense_dim(Tensor self) -> int
   variants: method
   dispatch:
-    CPU, CUDA: dense_dim_strided
     SparseCPU, SparseCUDA, SparseMeta: dense_dim_sparse
-    SparseCsrCPU, SparseCsrCUDA: dense_dim_sparse_csr
+    SparseCsrCPU, SparseCsrCUDA, SparseCsrMeta: dense_dim_sparse_csr
+    CompositeExplicitAutograd: dense_dim_default
   device_check: NoCheck
   device_guard: False
 
@@ -7070,7 +7248,7 @@
   variants: method
   dispatch:
     SparseCPU, SparseCUDA, SparseMeta: _nnz_sparse
-    SparseCsrCPU, SparseCsrCUDA: _nnz_sparse_csr
+    SparseCsrCPU, SparseCsrCUDA, SparseCsrMeta: _nnz_sparse_csr
   device_check: NoCheck
   device_guard: False
 
@@ -7133,7 +7311,7 @@
   variants: method
   dispatch:
     SparseCPU, SparseCUDA, SparseMeta: values_sparse
-    SparseCsrCPU, SparseCsrCUDA: values_sparse_csr
+    SparseCsrCPU, SparseCsrCUDA, SparseCsrMeta: values_sparse_csr
     NestedTensorCPU, NestedTensorCUDA: values_nested
     CompositeExplicitAutograd: values_default
   device_check: NoCheck
@@ -7142,7 +7320,7 @@
 - func: crow_indices(Tensor(a) self) -> Tensor(a)
   variants: method
   dispatch:
-    SparseCsrCPU, SparseCsrCUDA: crow_indices_sparse_csr
+    SparseCsrCPU, SparseCsrCUDA, SparseCsrMeta: crow_indices_sparse_csr
     CompositeExplicitAutograd: crow_indices_default
   device_check: NoCheck
   device_guard: False
@@ -7150,7 +7328,7 @@
 - func: col_indices(Tensor(a) self) -> Tensor(a)
   variants: method
   dispatch:
-    SparseCsrCPU, SparseCsrCUDA: col_indices_sparse_csr
+    SparseCsrCPU, SparseCsrCUDA, SparseCsrMeta: col_indices_sparse_csr
     CompositeExplicitAutograd: col_indices_default
   device_check: NoCheck
   device_guard: False
@@ -7158,7 +7336,7 @@
 - func: ccol_indices(Tensor(a) self) -> Tensor(a)
   variants: method
   dispatch:
-    SparseCsrCPU, SparseCsrCUDA: ccol_indices_sparse_csr
+    SparseCsrCPU, SparseCsrCUDA, SparseCsrMeta: ccol_indices_sparse_csr
     CompositeExplicitAutograd: ccol_indices_default
   device_check: NoCheck
   device_guard: False
@@ -7166,7 +7344,7 @@
 - func: row_indices(Tensor(a) self) -> Tensor(a)
   variants: method
   dispatch:
-    SparseCsrCPU, SparseCsrCUDA: row_indices_sparse_csr
+    SparseCsrCPU, SparseCsrCUDA, SparseCsrMeta: row_indices_sparse_csr
     CompositeExplicitAutograd: row_indices_default
   device_check: NoCheck
   device_guard: False
@@ -7185,7 +7363,7 @@
   device_check: NoCheck  # Allows copy into different device
   variants: function
   dispatch:
-    SparseCPU, SparseCUDA: copy_sparse_
+    SparseCPU, SparseCUDA, SparseMeta: copy_sparse_
   autogen: copy_sparse_to_sparse, copy_sparse_to_sparse.out
 
 # By adding the AutogradNestedTensor this makes this function CompositeImplicit-like for nested tensors
@@ -7288,7 +7466,7 @@
     MkldnnCPU: mkldnn_reorder_conv2d_weight
   autogen: mkldnn_reorder_conv2d_weight.out
 
-- func: mkldnn_reorder_conv3d_weight(Tensor self, SymInt[3] padding=0, SymInt[3] stride=1, SymInt[3] dilation=1, SymInt groups=1) -> Tensor
+- func: mkldnn_reorder_conv3d_weight(Tensor self, SymInt[3] padding=0, SymInt[3] stride=1, SymInt[3] dilation=1, SymInt groups=1, SymInt[]? input_size=None) -> Tensor
   variants: function
   python_module: nn
   dispatch:
@@ -7536,7 +7714,7 @@
 
 - func: result_type.Scalar_Scalar(Scalar scalar1, Scalar scalar2) -> ScalarType
 
-- func: can_cast(ScalarType from, ScalarType to) -> bool
+- func: can_cast(ScalarType from_, ScalarType to) -> bool
   variants: function
 
 - func: promote_types(ScalarType type1, ScalarType type2) -> ScalarType
@@ -7675,6 +7853,7 @@
   dispatch:
     CPU, CUDA, Meta, MPS: set_
   autogen: set.source_Storage, set.source_Storage_out
+  tags: inplace_view
 
 - func: set_.source_Storage_storage_offset(Tensor(a!) self, Storage source, SymInt storage_offset, SymInt[] size, SymInt[] stride=[]) -> Tensor(a!)
   variants: method
@@ -7687,6 +7866,7 @@
     MPS: set_storage_mps_
     QuantizedCPU, QuantizedCUDA: set_storage_quantized_
   autogen: set.source_Storage_storage_offset, set.source_Storage_storage_offset_out
+  tags: inplace_view
 
 - func: set_.source_Tensor_storage_offset(Tensor(a!) self, Tensor source, SymInt storage_offset, SymInt[] size, SymInt[] stride=[]) -> Tensor(a!)
   variants: method
@@ -7694,6 +7874,7 @@
   device_guard: False
   dispatch:
     CompositeImplicitAutograd: set__symint
+  tags: inplace_view
 
 - func: set_.source_Tensor(Tensor(a!) self, Tensor source) -> Tensor(a!)
   variants: method
@@ -7702,6 +7883,7 @@
   dispatch:
     CPU, CUDA, Meta, MPS: set_tensor_
   autogen: set.source_Tensor, set.source_Tensor_out
+  tags: inplace_view
 
 - func: set_(Tensor(a!) self) -> Tensor(a!)
   variants: method
@@ -7711,6 +7893,7 @@
     Meta: set_meta_
     MPS: set_mps_
   autogen: set, set.out
+  tags: inplace_view
 
 # Not making it CompositeImplicitAutograd because lift
 # should be a primitive w.r.t. functorch
@@ -10106,18 +10289,21 @@
   variants: method, function
   dispatch:
     CompositeExplicitAutograd: alias
+    NestedTensorCPU, NestedTensorCUDA: alias_nested
   tags: core
 
 - func: _amp_foreach_non_finite_check_and_unscale_(Tensor(a!)[] self, Tensor(b!) found_inf, Tensor inv_scale) -> ()
   variants: function
   dispatch:
     CUDA: _amp_foreach_non_finite_check_and_unscale_cuda_
+    CPU: _amp_foreach_non_finite_check_and_unscale_cpu_
   autogen: _amp_foreach_non_finite_check_and_unscale, _amp_foreach_non_finite_check_and_unscale.out
 
 - func: _amp_update_scale_(Tensor(a!) self, Tensor(b!) growth_tracker, Tensor found_inf, float scale_growth_factor, float scale_backoff_factor, int growth_interval) -> Tensor(a!)
   variants: function
   dispatch:
     CUDA: _amp_update_scale_cuda_
+    CPU: _amp_update_scale_cpu_
   autogen: _amp_update_scale, _amp_update_scale.out
 
     #- func: _cat(Tensor[] tensors, int dim=0) -> Tensor
@@ -10137,14 +10323,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_add_scalar_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_add_scalar_kernel_slow
     CUDA: foreach_tensor_add_scalar_kernel_cuda
 
 - func: _foreach_add_.Scalar(Tensor(a!)[] self, Scalar scalar) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_add_scalar_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_add_scalar_kernel_slow_
     CUDA: foreach_tensor_add_scalar_kernel_cuda_
   autogen: _foreach_add.Scalar_out
 
@@ -10152,14 +10338,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_add_list_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_add_list_kernel_slow
     CUDA: foreach_tensor_add_list_kernel_cuda
 
 - func: _foreach_add_.List(Tensor(a!)[] self, Tensor[] other, *, Scalar alpha=1) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_add_list_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_add_list_kernel_slow_
     CUDA: foreach_tensor_add_list_kernel_cuda_
   autogen: _foreach_add.List_out
 
@@ -10167,14 +10353,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_add_scalarlist_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_add_scalarlist_kernel_slow
     CUDA: foreach_tensor_add_scalarlist_kernel_cuda
 
 - func: _foreach_add_.ScalarList(Tensor(a!)[] self, Scalar[] scalars) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_add_scalarlist_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_add_scalarlist_kernel_slow_
     CUDA: foreach_tensor_add_scalarlist_kernel_cuda_
   autogen: _foreach_add.ScalarList_out
 
@@ -10182,14 +10368,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_add_tensor_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_add_tensor_kernel_slow
     CUDA: foreach_tensor_add_tensor_kernel_cuda
 
 - func: _foreach_add_.Tensor(Tensor(a!)[] self, Tensor other, *, Scalar alpha=1) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_add_tensor_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_add_tensor_kernel_slow_
     CUDA: foreach_tensor_add_tensor_kernel_cuda_
   autogen: _foreach_add.Tensor_out
 
@@ -10197,14 +10383,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_sub_scalar_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_sub_scalar_kernel_slow
     CUDA: foreach_tensor_sub_scalar_kernel_cuda
 
 - func: _foreach_sub_.Scalar(Tensor(a!)[] self, Scalar scalar) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_sub_scalar_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_sub_scalar_kernel_slow_
     CUDA: foreach_tensor_sub_scalar_kernel_cuda_
   autogen: _foreach_sub.Scalar_out
 
@@ -10212,14 +10398,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_sub_list_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_sub_list_kernel_slow
     CUDA: foreach_tensor_sub_list_kernel_cuda
 
 - func: _foreach_sub_.List(Tensor(a!)[] self, Tensor[] other, *, Scalar alpha=1) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_sub_list_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_sub_list_kernel_slow_
     CUDA: foreach_tensor_sub_list_kernel_cuda_
   autogen: _foreach_sub.List_out
 
@@ -10227,14 +10413,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_sub_scalarlist_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_sub_scalarlist_kernel_slow
     CUDA: foreach_tensor_sub_scalarlist_kernel_cuda
 
 - func: _foreach_sub_.ScalarList(Tensor(a!)[] self, Scalar[] scalars) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_sub_scalarlist_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_sub_scalarlist_kernel_slow_
     CUDA: foreach_tensor_sub_scalarlist_kernel_cuda_
   autogen: _foreach_sub.ScalarList_out
 
@@ -10242,14 +10428,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_mul_scalar_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_mul_scalar_kernel_slow
     CUDA: foreach_tensor_mul_scalar_kernel_cuda
 
 - func: _foreach_mul_.Scalar(Tensor(a!)[] self, Scalar scalar) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_mul_scalar_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_mul_scalar_kernel_slow_
     CUDA: foreach_tensor_mul_scalar_kernel_cuda_
   autogen: _foreach_mul.Scalar_out
 
@@ -10257,14 +10443,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_mul_list_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_mul_list_kernel_slow
     CUDA: foreach_tensor_mul_list_kernel_cuda
 
 - func: _foreach_mul_.List(Tensor(a!)[] self, Tensor[] other) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_mul_list_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_mul_list_kernel_slow_
     CUDA: foreach_tensor_mul_list_kernel_cuda_
   autogen: _foreach_mul.List_out
 
@@ -10272,14 +10458,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_mul_scalarlist_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_mul_scalarlist_kernel_slow
     CUDA: foreach_tensor_mul_scalarlist_kernel_cuda
 
 - func: _foreach_mul_.ScalarList(Tensor(a!)[] self, Scalar[] scalars) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_mul_scalarlist_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_mul_scalarlist_kernel_slow_
     CUDA: foreach_tensor_mul_scalarlist_kernel_cuda_
   autogen: _foreach_mul.ScalarList_out
 
@@ -10287,14 +10473,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_mul_tensor_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_mul_tensor_kernel_slow
     CUDA: foreach_tensor_mul_tensor_kernel_cuda
 
 - func: _foreach_mul_.Tensor(Tensor(a!)[] self, Tensor other) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_mul_tensor_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_mul_tensor_kernel_slow_
     CUDA: foreach_tensor_mul_tensor_kernel_cuda_
   autogen: _foreach_mul.Tensor_out
 
@@ -10302,14 +10488,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_div_scalar_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_div_scalar_kernel_slow
     CUDA: foreach_tensor_div_scalar_kernel_cuda
 
 - func: _foreach_div_.Scalar(Tensor(a!)[] self, Scalar scalar) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_div_scalar_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_div_scalar_kernel_slow_
     CUDA: foreach_tensor_div_scalar_kernel_cuda_
   autogen: _foreach_div.Scalar_out
 
@@ -10317,14 +10503,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_div_list_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_div_list_kernel_slow
     CUDA: foreach_tensor_div_list_kernel_cuda
 
 - func: _foreach_div_.List(Tensor(a!)[] self, Tensor[] other) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_div_list_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_div_list_kernel_slow_
     CUDA: foreach_tensor_div_list_kernel_cuda_
   autogen: _foreach_div.List_out
 
@@ -10332,14 +10518,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_div_scalarlist_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_div_scalarlist_kernel_slow
     CUDA: foreach_tensor_div_scalarlist_kernel_cuda
 
 - func: _foreach_div_.ScalarList(Tensor(a!)[] self, Scalar[] scalars) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_div_scalarlist_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_div_scalarlist_kernel_slow_
     CUDA: foreach_tensor_div_scalarlist_kernel_cuda_
   autogen: _foreach_div.ScalarList_out
 
@@ -10347,14 +10533,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_div_tensor_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_div_tensor_kernel_slow
     CUDA: foreach_tensor_div_tensor_kernel_cuda
 
 - func: _foreach_div_.Tensor(Tensor(a!)[] self, Tensor other) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_div_tensor_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_div_tensor_kernel_slow_
     CUDA: foreach_tensor_div_tensor_kernel_cuda_
   autogen: _foreach_div.Tensor_out
 
@@ -10362,14 +10548,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_max_scalar_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_clamp_max_scalar_kernel_slow
     CUDA: foreach_tensor_clamp_max_scalar_kernel_cuda
 
 - func: _foreach_clamp_max_.Scalar(Tensor(a!)[] self, Scalar scalar) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_max_scalar_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_clamp_max_scalar_kernel_slow_
     CUDA: foreach_tensor_clamp_max_scalar_kernel_cuda_
   autogen: _foreach_clamp_max.Scalar_out
 
@@ -10377,14 +10563,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_max_list_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_clamp_max_list_kernel_slow
     CUDA: foreach_tensor_clamp_max_list_kernel_cuda
 
 - func: _foreach_clamp_max_.List(Tensor(a!)[] self, Tensor[] other) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_max_list_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_clamp_max_list_kernel_slow_
     CUDA: foreach_tensor_clamp_max_list_kernel_cuda_
   autogen: _foreach_clamp_max.List_out
 
@@ -10392,14 +10578,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_max_scalarlist_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_clamp_max_scalarlist_kernel_slow
     CUDA: foreach_tensor_clamp_max_scalarlist_kernel_cuda
 
 - func: _foreach_clamp_max_.ScalarList(Tensor(a!)[] self, Scalar[] scalars) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_max_scalarlist_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_clamp_max_scalarlist_kernel_slow_
     CUDA: foreach_tensor_clamp_max_scalarlist_kernel_cuda_
   autogen: _foreach_clamp_max.ScalarList_out
 
@@ -10407,14 +10593,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_min_scalar_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_clamp_min_scalar_kernel_slow
     CUDA: foreach_tensor_clamp_min_scalar_kernel_cuda
 
 - func: _foreach_clamp_min_.Scalar(Tensor(a!)[] self, Scalar scalar) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_min_scalar_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_clamp_min_scalar_kernel_slow_
     CUDA: foreach_tensor_clamp_min_scalar_kernel_cuda_
   autogen: _foreach_clamp_min.Scalar_out
 
@@ -10422,14 +10608,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_min_list_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_clamp_min_list_kernel_slow
     CUDA: foreach_tensor_clamp_min_list_kernel_cuda
 
 - func: _foreach_clamp_min_.List(Tensor(a!)[] self, Tensor[] other) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_min_list_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_clamp_min_list_kernel_slow_
     CUDA: foreach_tensor_clamp_min_list_kernel_cuda_
   autogen: _foreach_clamp_min.List_out
 
@@ -10437,14 +10623,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_min_scalarlist_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_clamp_min_scalarlist_kernel_slow
     CUDA: foreach_tensor_clamp_min_scalarlist_kernel_cuda
 
 - func: _foreach_clamp_min_.ScalarList(Tensor(a!)[] self, Scalar[] scalars) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_min_scalarlist_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_clamp_min_scalarlist_kernel_slow_
     CUDA: foreach_tensor_clamp_min_scalarlist_kernel_cuda_
   autogen: _foreach_clamp_min.ScalarList_out
 
@@ -10453,14 +10639,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_min_scalar_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_clamp_min_scalar_kernel_slow
     CUDA: foreach_tensor_clamp_min_scalar_kernel_cuda
 
 - func: _foreach_maximum_.Scalar(Tensor(a!)[] self, Scalar scalar) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_min_scalar_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_clamp_min_scalar_kernel_slow_
     CUDA: foreach_tensor_clamp_min_scalar_kernel_cuda_
   autogen: _foreach_maximum.Scalar_out
 
@@ -10469,14 +10655,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_min_list_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_clamp_min_list_kernel_slow
     CUDA: foreach_tensor_clamp_min_list_kernel_cuda
 
 - func: _foreach_maximum_.List(Tensor(a!)[] self, Tensor[] other) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_min_list_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_clamp_min_list_kernel_slow_
     CUDA: foreach_tensor_clamp_min_list_kernel_cuda_
   autogen: _foreach_maximum.List_out
 
@@ -10485,14 +10671,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_min_scalarlist_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_clamp_min_scalarlist_kernel_slow
     CUDA: foreach_tensor_clamp_min_scalarlist_kernel_cuda
 
 - func: _foreach_maximum_.ScalarList(Tensor(a!)[] self, Scalar[] scalars) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_min_scalarlist_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_clamp_min_scalarlist_kernel_slow_
     CUDA: foreach_tensor_clamp_min_scalarlist_kernel_cuda_
   autogen: _foreach_maximum.ScalarList_out
 
@@ -10500,14 +10686,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_max_scalar_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_clamp_max_scalar_kernel_slow
     CUDA: foreach_tensor_clamp_max_scalar_kernel_cuda
 
 - func: _foreach_minimum_.Scalar(Tensor(a!)[] self, Scalar scalar) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_max_scalar_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_clamp_max_scalar_kernel_slow_
     CUDA: foreach_tensor_clamp_max_scalar_kernel_cuda_
   autogen: _foreach_minimum.Scalar_out
 
@@ -10515,14 +10701,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_max_list_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_clamp_max_list_kernel_slow
     CUDA: foreach_tensor_clamp_max_list_kernel_cuda
 
 - func: _foreach_minimum_.List(Tensor(a!)[] self, Tensor[] other) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_max_list_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_clamp_max_list_kernel_slow_
     CUDA: foreach_tensor_clamp_max_list_kernel_cuda_
   autogen: _foreach_minimum.List_out
 
@@ -10530,14 +10716,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_max_scalarlist_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_clamp_max_scalarlist_kernel_slow
     CUDA: foreach_tensor_clamp_max_scalarlist_kernel_cuda
 
 - func: _foreach_minimum_.ScalarList(Tensor(a!)[] self, Scalar[] scalars) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_clamp_max_scalarlist_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_clamp_max_scalarlist_kernel_slow_
     CUDA: foreach_tensor_clamp_max_scalarlist_kernel_cuda_
   autogen: _foreach_minimum.ScalarList_out
 
@@ -10545,28 +10731,28 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_addcdiv_scalar_slow
+    CompositeExplicitAutograd: foreach_tensor_addcdiv_scalar_slow
     CUDA: foreach_tensor_addcdiv_scalar_cuda
 
 - func: _foreach_addcdiv.ScalarList(Tensor[] self, Tensor[] tensor1, Tensor[] tensor2, Scalar[] scalars) -> Tensor[]
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_addcdiv_scalarlist_slow
+    CompositeExplicitAutograd: foreach_tensor_addcdiv_scalarlist_slow
     CUDA: foreach_tensor_addcdiv_scalarlist_cuda
 
 - func: _foreach_addcdiv.Tensor(Tensor[] self, Tensor[] tensor1, Tensor[] tensor2, Tensor scalars) -> Tensor[]
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_addcdiv_tensor_slow
+    CompositeExplicitAutograd: foreach_tensor_addcdiv_tensor_slow
     CUDA: foreach_tensor_addcdiv_tensor_cuda
 
 - func: _foreach_addcdiv_.Scalar(Tensor(a!)[] self, Tensor[] tensor1, Tensor[] tensor2, Scalar value=1) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_addcdiv_scalar_slow_
+    CompositeExplicitAutograd: foreach_tensor_addcdiv_scalar_slow_
     CUDA: foreach_tensor_addcdiv_scalar_cuda_
   autogen: _foreach_addcdiv.Scalar_out
 
@@ -10574,7 +10760,7 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_addcdiv_scalarlist_slow_
+    CompositeExplicitAutograd: foreach_tensor_addcdiv_scalarlist_slow_
     CUDA: foreach_tensor_addcdiv_scalarlist_cuda_
   autogen: _foreach_addcdiv.ScalarList_out
 
@@ -10582,7 +10768,7 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_addcdiv_tensor_slow_
+    CompositeExplicitAutograd: foreach_tensor_addcdiv_tensor_slow_
     CUDA: foreach_tensor_addcdiv_tensor_cuda_
   autogen: _foreach_addcdiv.Tensor_out
 
@@ -10590,28 +10776,28 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_addcmul_scalar_slow
+    CompositeExplicitAutograd: foreach_tensor_addcmul_scalar_slow
     CUDA: foreach_tensor_addcmul_scalar_cuda
 
 - func: _foreach_addcmul.ScalarList(Tensor[] self, Tensor[] tensor1, Tensor[] tensor2, Scalar[] scalars) -> Tensor[]
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_addcmul_scalarlist_slow
+    CompositeExplicitAutograd: foreach_tensor_addcmul_scalarlist_slow
     CUDA: foreach_tensor_addcmul_scalarlist_cuda
 
 - func: _foreach_addcmul.Tensor(Tensor[] self, Tensor[] tensor1, Tensor[] tensor2, Tensor scalars) -> Tensor[]
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_addcmul_tensor_slow
+    CompositeExplicitAutograd: foreach_tensor_addcmul_tensor_slow
     CUDA: foreach_tensor_addcmul_tensor_cuda
 
 - func: _foreach_addcmul_.Scalar(Tensor(a!)[] self, Tensor[] tensor1, Tensor[] tensor2, Scalar value=1) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_addcmul_scalar_slow_
+    CompositeExplicitAutograd: foreach_tensor_addcmul_scalar_slow_
     CUDA: foreach_tensor_addcmul_scalar_cuda_
   autogen: _foreach_addcmul.Scalar_out
 
@@ -10619,7 +10805,7 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_addcmul_scalarlist_slow_
+    CompositeExplicitAutograd: foreach_tensor_addcmul_scalarlist_slow_
     CUDA: foreach_tensor_addcmul_scalarlist_cuda_
   autogen: _foreach_addcmul.ScalarList_out
 
@@ -10627,7 +10813,7 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_addcmul_tensor_slow_
+    CompositeExplicitAutograd: foreach_tensor_addcmul_tensor_slow_
     CUDA: foreach_tensor_addcmul_tensor_cuda_
   autogen: _foreach_addcmul.Tensor_out
 
@@ -10635,14 +10821,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_abs_slow
+    CompositeExplicitAutograd: foreach_tensor_abs_slow
     CUDA: foreach_tensor_abs_cuda
 
 - func: _foreach_abs_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_abs_slow_
+    CompositeExplicitAutograd: foreach_tensor_abs_slow_
     CUDA: foreach_tensor_abs_cuda_
   autogen: _foreach_abs.out
 
@@ -10650,14 +10836,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_acos_slow
+    CompositeExplicitAutograd: foreach_tensor_acos_slow
     CUDA: foreach_tensor_acos_cuda
 
 - func: _foreach_acos_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_acos_slow_
+    CompositeExplicitAutograd: foreach_tensor_acos_slow_
     CUDA: foreach_tensor_acos_cuda_
   autogen: _foreach_acos.out
 
@@ -10665,14 +10851,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_asin_slow
+    CompositeExplicitAutograd: foreach_tensor_asin_slow
     CUDA: foreach_tensor_asin_cuda
 
 - func: _foreach_asin_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_asin_slow_
+    CompositeExplicitAutograd: foreach_tensor_asin_slow_
     CUDA: foreach_tensor_asin_cuda_
   autogen: _foreach_asin.out
 
@@ -10680,14 +10866,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_atan_slow
+    CompositeExplicitAutograd: foreach_tensor_atan_slow
     CUDA: foreach_tensor_atan_cuda
 
 - func: _foreach_atan_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_atan_slow_
+    CompositeExplicitAutograd: foreach_tensor_atan_slow_
     CUDA: foreach_tensor_atan_cuda_
   autogen: _foreach_atan.out
 
@@ -10695,14 +10881,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_ceil_slow
+    CompositeExplicitAutograd: foreach_tensor_ceil_slow
     CUDA: foreach_tensor_ceil_cuda
 
 - func: _foreach_ceil_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_ceil_slow_
+    CompositeExplicitAutograd: foreach_tensor_ceil_slow_
     CUDA: foreach_tensor_ceil_cuda_
   autogen: _foreach_ceil.out
 
@@ -10710,14 +10896,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_cos_slow
+    CompositeExplicitAutograd: foreach_tensor_cos_slow
     CUDA: foreach_tensor_cos_cuda
 
 - func: _foreach_cos_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_cos_slow_
+    CompositeExplicitAutograd: foreach_tensor_cos_slow_
     CUDA: foreach_tensor_cos_cuda_
   autogen: _foreach_cos.out
 
@@ -10725,14 +10911,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_cosh_slow
+    CompositeExplicitAutograd: foreach_tensor_cosh_slow
     CUDA: foreach_tensor_cosh_cuda
 
 - func: _foreach_cosh_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_cosh_slow_
+    CompositeExplicitAutograd: foreach_tensor_cosh_slow_
     CUDA: foreach_tensor_cosh_cuda_
   autogen: _foreach_cosh.out
 
@@ -10740,14 +10926,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_erf_slow
+    CompositeExplicitAutograd: foreach_tensor_erf_slow
     CUDA: foreach_tensor_erf_cuda
 
 - func: _foreach_erf_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_erf_slow_
+    CompositeExplicitAutograd: foreach_tensor_erf_slow_
     CUDA: foreach_tensor_erf_cuda_
   autogen: _foreach_erf.out
 
@@ -10755,14 +10941,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_erfc_slow
+    CompositeExplicitAutograd: foreach_tensor_erfc_slow
     CUDA: foreach_tensor_erfc_cuda
 
 - func: _foreach_erfc_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_erfc_slow_
+    CompositeExplicitAutograd: foreach_tensor_erfc_slow_
     CUDA: foreach_tensor_erfc_cuda_
   autogen: _foreach_erfc.out
 
@@ -10770,14 +10956,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_exp_slow
+    CompositeExplicitAutograd: foreach_tensor_exp_slow
     CUDA: foreach_tensor_exp_cuda
 
 - func: _foreach_exp_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_exp_slow_
+    CompositeExplicitAutograd: foreach_tensor_exp_slow_
     CUDA: foreach_tensor_exp_cuda_
   autogen: _foreach_exp.out
 
@@ -10785,14 +10971,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_expm1_slow
+    CompositeExplicitAutograd: foreach_tensor_expm1_slow
     CUDA: foreach_tensor_expm1_cuda
 
 - func: _foreach_expm1_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_expm1_slow_
+    CompositeExplicitAutograd: foreach_tensor_expm1_slow_
     CUDA: foreach_tensor_expm1_cuda_
   autogen: _foreach_expm1.out
 
@@ -10800,14 +10986,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_floor_slow
+    CompositeExplicitAutograd: foreach_tensor_floor_slow
     CUDA: foreach_tensor_floor_cuda
 
 - func: _foreach_floor_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_floor_slow_
+    CompositeExplicitAutograd: foreach_tensor_floor_slow_
     CUDA: foreach_tensor_floor_cuda_
   autogen: _foreach_floor.out
 
@@ -10815,14 +11001,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_frac_slow
+    CompositeExplicitAutograd: foreach_tensor_frac_slow
     CUDA: foreach_tensor_frac_cuda
 
 - func: _foreach_frac_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_frac_slow_
+    CompositeExplicitAutograd: foreach_tensor_frac_slow_
     CUDA: foreach_tensor_frac_cuda_
   autogen: _foreach_frac.out
 
@@ -10830,7 +11016,7 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensors are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_ternary_lerp_slow
+    CompositeExplicitAutograd: foreach_tensor_ternary_lerp_slow
     CUDA: foreach_tensor_lerp_ternary_cuda
   autogen: _foreach_lerp.List_out
 
@@ -10838,7 +11024,7 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensors are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_ternary_lerp_slow_
+    CompositeExplicitAutograd: foreach_tensor_ternary_lerp_slow_
     CUDA: foreach_tensor_lerp_ternary_cuda_
   autogen: _foreach_lerp.List_out
 
@@ -10846,7 +11032,7 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensors are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_lerp_list_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_lerp_list_kernel_slow
     CUDA: foreach_tensor_lerp_list_cuda
   autogen: _foreach_lerp.Scalar_out
 
@@ -10854,7 +11040,7 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensors are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_lerp_list_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_lerp_list_kernel_slow_
     CUDA: foreach_tensor_lerp_list_cuda_
   autogen: _foreach_lerp.Scalar_out
 
@@ -10862,14 +11048,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_lgamma_slow
+    CompositeExplicitAutograd: foreach_tensor_lgamma_slow
     CUDA: foreach_tensor_lgamma_cuda
 
 - func: _foreach_lgamma_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_lgamma_slow_
+    CompositeExplicitAutograd: foreach_tensor_lgamma_slow_
     CUDA: foreach_tensor_lgamma_cuda_
   autogen: _foreach_lgamma.out
 
@@ -10877,14 +11063,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_log_slow
+    CompositeExplicitAutograd: foreach_tensor_log_slow
     CUDA: foreach_tensor_log_cuda
 
 - func: _foreach_log_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_log_slow_
+    CompositeExplicitAutograd: foreach_tensor_log_slow_
     CUDA: foreach_tensor_log_cuda_
   autogen: _foreach_log.out
 
@@ -10892,14 +11078,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_log10_slow
+    CompositeExplicitAutograd: foreach_tensor_log10_slow
     CUDA: foreach_tensor_log10_cuda
 
 - func: _foreach_log10_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_log10_slow_
+    CompositeExplicitAutograd: foreach_tensor_log10_slow_
     CUDA: foreach_tensor_log10_cuda_
   autogen: _foreach_log10.out
 
@@ -10907,14 +11093,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_log1p_slow
+    CompositeExplicitAutograd: foreach_tensor_log1p_slow
     CUDA: foreach_tensor_log1p_cuda
 
 - func: _foreach_log1p_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_log1p_slow_
+    CompositeExplicitAutograd: foreach_tensor_log1p_slow_
     CUDA: foreach_tensor_log1p_cuda_
   autogen: _foreach_log1p.out
 
@@ -10922,37 +11108,45 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_log2_slow
+    CompositeExplicitAutograd: foreach_tensor_log2_slow
     CUDA: foreach_tensor_log2_cuda
 
 - func: _foreach_log2_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_log2_slow_
+    CompositeExplicitAutograd: foreach_tensor_log2_slow_
     CUDA: foreach_tensor_log2_cuda_
   autogen: _foreach_log2.out
 
+- func: _foreach_max(Tensor[] self) -> Tensor[]
+  device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
+  variants: function
+  dispatch:
+    CompositeExplicitAutograd: foreach_tensor_max_slow
+    CUDA: foreach_tensor_max_cuda
+  autogen: _foreach_max.out
+
 - func: _foreach_neg(Tensor[] self) -> Tensor[]
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_neg_slow
+    CompositeExplicitAutograd: foreach_tensor_neg_slow
     CUDA: foreach_tensor_neg_cuda
 
 - func: _foreach_neg_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_neg_slow_
+    CompositeExplicitAutograd: foreach_tensor_neg_slow_
     CUDA: foreach_tensor_neg_cuda_
   autogen: _foreach_neg.out
 
-- func: _foreach_norm.Scalar(Tensor[] self, Scalar ord=2) -> Tensor[]
+- func: _foreach_norm.Scalar(Tensor[] self, Scalar ord=2, ScalarType? dtype=None) -> Tensor[]
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_norm_slow
+    CompositeExplicitAutograd: foreach_tensor_norm_slow
     CUDA: foreach_tensor_norm_cuda
   autogen: _foreach_norm.Scalar_out
 
@@ -10960,35 +11154,35 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_pow_list_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_pow_list_kernel_slow
     CUDA: foreach_tensor_pow_list_kernel_cuda
 
 - func: _foreach_pow.Scalar(Tensor[] self, Scalar exponent) -> Tensor[]
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_pow_scalar_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_pow_scalar_kernel_slow
     CUDA: foreach_tensor_pow_scalar_kernel_cuda
 
 - func: _foreach_pow.ScalarList(Tensor[] self, Scalar[] exponent) -> Tensor[]
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_pow_scalarlist_kernel_slow
+    CompositeExplicitAutograd: foreach_tensor_pow_scalarlist_kernel_slow
     CUDA: foreach_tensor_pow_scalarlist_kernel_cuda
 
 - func: _foreach_pow.ScalarAndTensor(Scalar self, Tensor[] exponent) -> Tensor[]
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_scalar_pow_list_kernel_slow
+    CompositeExplicitAutograd: foreach_scalar_pow_list_kernel_slow
     CUDA: foreach_scalar_pow_list_kernel_cuda
 
 - func: _foreach_pow_.List(Tensor(a!)[] self, Tensor[] exponent) -> ()
   device_check: NoCheck
   variants: function
   dispatch:
-    CPU: foreach_tensor_pow_list_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_pow_list_kernel_slow_
     CUDA: foreach_tensor_pow_list_kernel_cuda_
   autogen: _foreach_pow.List_out
 
@@ -10996,7 +11190,7 @@
   device_check: NoCheck
   variants: function
   dispatch:
-    CPU: foreach_tensor_pow_scalar_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_pow_scalar_kernel_slow_
     CUDA: foreach_tensor_pow_scalar_kernel_cuda_
   autogen: _foreach_pow.Scalar_out
 
@@ -11004,7 +11198,7 @@
   device_check: NoCheck
   variants: function
   dispatch:
-    CPU: foreach_tensor_pow_scalarlist_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_pow_scalarlist_kernel_slow_
     CUDA: foreach_tensor_pow_scalarlist_kernel_cuda_
   autogen: _foreach_pow.ScalarList_out
 
@@ -11012,14 +11206,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_reciprocal_slow
+    CompositeExplicitAutograd: foreach_tensor_reciprocal_slow
     CUDA: foreach_tensor_reciprocal_cuda
 
 - func: _foreach_reciprocal_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_reciprocal_slow_
+    CompositeExplicitAutograd: foreach_tensor_reciprocal_slow_
     CUDA: foreach_tensor_reciprocal_cuda_
   autogen: _foreach_reciprocal.out
 
@@ -11027,14 +11221,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_round_slow
+    CompositeExplicitAutograd: foreach_tensor_round_slow
     CUDA: foreach_tensor_round_cuda
 
 - func: _foreach_round_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_round_slow_
+    CompositeExplicitAutograd: foreach_tensor_round_slow_
     CUDA: foreach_tensor_round_cuda_
   autogen: _foreach_round.out
 
@@ -11042,14 +11236,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_sigmoid_slow
+    CompositeExplicitAutograd: foreach_tensor_sigmoid_slow
     CUDA: foreach_tensor_sigmoid_cuda
 
 - func: _foreach_sigmoid_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_sigmoid_slow_
+    CompositeExplicitAutograd: foreach_tensor_sigmoid_slow_
     CUDA: foreach_tensor_sigmoid_cuda_
   autogen: _foreach_sigmoid.out
 
@@ -11057,14 +11251,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_sign_slow
+    CompositeExplicitAutograd: foreach_tensor_sign_slow
     CUDA: foreach_tensor_sign_cuda
 
 - func: _foreach_sign_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_sign_slow_
+    CompositeExplicitAutograd: foreach_tensor_sign_slow_
     CUDA: foreach_tensor_sign_cuda_
   autogen: _foreach_sign.out
 
@@ -11072,14 +11266,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_sin_slow
+    CompositeExplicitAutograd: foreach_tensor_sin_slow
     CUDA: foreach_tensor_sin_cuda
 
 - func: _foreach_sin_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_sin_slow_
+    CompositeExplicitAutograd: foreach_tensor_sin_slow_
     CUDA: foreach_tensor_sin_cuda_
   autogen: _foreach_sin.out
 
@@ -11087,14 +11281,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_sinh_slow
+    CompositeExplicitAutograd: foreach_tensor_sinh_slow
     CUDA: foreach_tensor_sinh_cuda
 
 - func: _foreach_sinh_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_sinh_slow_
+    CompositeExplicitAutograd: foreach_tensor_sinh_slow_
     CUDA: foreach_tensor_sinh_cuda_
   autogen: _foreach_sinh.out
 
@@ -11102,14 +11296,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_sqrt_slow
+    CompositeExplicitAutograd: foreach_tensor_sqrt_slow
     CUDA: foreach_tensor_sqrt_cuda
 
 - func: _foreach_sqrt_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_sqrt_slow_
+    CompositeExplicitAutograd: foreach_tensor_sqrt_slow_
     CUDA: foreach_tensor_sqrt_cuda_
   autogen: _foreach_sqrt.out
 
@@ -11117,14 +11311,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_tan_slow
+    CompositeExplicitAutograd: foreach_tensor_tan_slow
     CUDA: foreach_tensor_tan_cuda
 
 - func: _foreach_tan_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_tan_slow_
+    CompositeExplicitAutograd: foreach_tensor_tan_slow_
     CUDA: foreach_tensor_tan_cuda_
   autogen: _foreach_tan.out
 
@@ -11132,14 +11326,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_tanh_slow
+    CompositeExplicitAutograd: foreach_tensor_tanh_slow
     CUDA: foreach_tensor_tanh_cuda
 
 - func: _foreach_tanh_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_tanh_slow_
+    CompositeExplicitAutograd: foreach_tensor_tanh_slow_
     CUDA: foreach_tensor_tanh_cuda_
   autogen: _foreach_tanh.out
 
@@ -11147,14 +11341,14 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_trunc_slow
+    CompositeExplicitAutograd: foreach_tensor_trunc_slow
     CUDA: foreach_tensor_trunc_cuda
 
 - func: _foreach_trunc_(Tensor(a!)[] self) -> ()
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_trunc_slow_
+    CompositeExplicitAutograd: foreach_tensor_trunc_slow_
     CUDA: foreach_tensor_trunc_cuda_
   autogen: _foreach_trunc.out
 
@@ -11162,7 +11356,7 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_zero_slow_
+    CompositeExplicitAutograd: foreach_tensor_zero_slow_
     CUDA: foreach_tensor_zero_cuda_
   autogen: _foreach_zero, _foreach_zero.out
 
@@ -11170,9 +11364,15 @@
   device_check: NoCheck   # foreach kernels fall back to slow path when tensor are on different devices
   variants: function
   dispatch:
-    CPU: foreach_tensor_copy_list_kernel_slow_
+    CompositeExplicitAutograd: foreach_tensor_copy_list_kernel_slow_
     CUDA: foreach_tensor_copy_list_kernel_cuda_
-  autogen: _foreach_copy, _foreach_copy.out
+  autogen: _foreach_copy.out
+
+- func: _foreach_copy(Tensor[] self, Tensor[] src, bool non_blocking=False) -> Tensor[] self_out
+  device_check: NoCheck
+  variants: function
+  dispatch:
+    CompositeExplicitAutograd: _foreach_copy
 
 - func: bucketize.Tensor(Tensor self, Tensor boundaries, *, bool out_int32=False, bool right=False) -> Tensor
   dispatch:
@@ -12341,6 +12541,7 @@
   dispatch:
     CPU: upsample_linear1d_out_cpu
     CUDA: upsample_linear1d_out_cuda
+    MPS: upsample_linear1d_out_mps
 
 - func: upsample_linear1d(Tensor self, SymInt[1] output_size, bool align_corners, float? scales=None) -> Tensor
   python_module: nn
@@ -12352,6 +12553,7 @@
   dispatch:
     CPU: upsample_linear1d_backward_out_cpu
     CUDA: upsample_linear1d_backward_out_cuda
+    MPS: upsample_linear1d_backward_out_mps
 
 - func: upsample_linear1d_backward(Tensor grad_output, SymInt[1] output_size, SymInt[3] input_size, bool align_corners, float? scales=None) -> Tensor
   python_module: nn
@@ -12824,7 +13026,7 @@
     SparseMeta: isinf_sparse_meta
     SparseCsrCPU, SparseCsrCUDA: isinf_sparse_csr
   autogen: isinf.out
-  tags: core
+  tags: [core, pointwise]
 
 - func: record_stream(Tensor(a!) self, Stream s) -> ()
   variants: method
@@ -13750,11 +13952,18 @@
   dispatch:
     CPU, CUDA: linalg_eig_out
 
+- func: _linalg_eigvals(Tensor self) -> Tensor
+  python_module: linalg
+  dispatch:
+    CPU, CUDA: _linalg_eigvals
+
 - func: linalg_eigvals(Tensor self) -> Tensor
   python_module: linalg
 
 - func: linalg_eigvals.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
   python_module: linalg
+  dispatch:
+    CPU, CUDA: linalg_eigvals_out
 
 # This function is exposes the `compute_v` flag, which is then used to implement `linalg.eigh` and
 # `linalg.eigvalsh` as composite functions that call this one
@@ -14058,6 +14267,12 @@
 # It is undocumented and should not be used outside of tests.
 - func: _test_serialization_subcmul(Tensor self, Tensor other, Scalar alpha=1) -> Tensor
 
+# Note: for testing COW materialization within `at::parallel_for` loop function
+- func: _test_parallel_materialize(Tensor self, int num_parallel, bool skip_first=False) -> Tensor
+  variants: function
+  dispatch:
+    CompositeExplicitAutograd: _test_parallel_materialize
+
 # Note: this function is only for testing.
 - func: _test_optional_intlist(Tensor values, int[]? addends) -> Tensor
   python_module: nn
@@ -14392,6 +14607,7 @@
   variants: function
   dispatch:
     CompositeExplicitAutograd: split_with_sizes_copy_out
+    CUDA: split_with_sizes_copy_out_cuda
 
 - func: view_copy(Tensor self, SymInt[] size) -> Tensor
   variants: function
@@ -14428,6 +14644,16 @@
     NestedTensorCUDA: NestedTensor_to_padded_tensor_cuda
   autogen: to_padded_tensor.out
 
+- func: _jagged_to_padded_dense_forward(Tensor values, Tensor[] offsets, SymInt[] max_lengths, float padding_value=0.0) -> Tensor
+  variants: function
+  dispatch:
+    CUDA: _fbgemm_jagged_to_padded_dense_forward
+
+- func: _padded_dense_to_jagged_forward(Tensor dense, Tensor[] offsets, SymInt? total_L=None) -> Tensor
+  variants: function
+  dispatch:
+    CUDA: _fbgemm_dense_to_jagged_forward_symint
+
 - func: _nested_tensor_softmax_with_shape(Tensor self, Tensor query) -> Tensor
   dispatch:
     NestedTensorCPU: NestedTensor_softmax_dropout
@@ -14468,19 +14694,28 @@
 
 - func: _scaled_dot_product_flash_attention(Tensor query, Tensor key, Tensor value, float dropout_p=0.0, bool is_causal=False, bool return_debug_mask=False, *, float? scale=None) -> (Tensor output, Tensor logsumexp, Tensor cum_seq_q, Tensor cum_seq_k, SymInt max_q, SymInt max_k, Tensor philox_seed, Tensor philox_offset, Tensor debug_attn_mask)
   dispatch:
-    CPU: _scaled_dot_product_flash_attention_cpu
     CUDA: _scaled_dot_product_flash_attention_cuda
     NestedTensorCUDA: _scaled_dot_product_flash_attention_nestedtensor_cuda
   tags: nondeterministic_seeded
 
+- func: _scaled_dot_product_flash_attention_for_cpu(Tensor query, Tensor key, Tensor value, float dropout_p=0.0, bool is_causal=False, *, Tensor? attn_mask=None, float? scale=None) -> (Tensor output, Tensor logsumexp)
+  dispatch:
+    CPU: _scaled_dot_product_flash_attention_cpu
+  tags: nondeterministic_seeded
+
 - func: _scaled_dot_product_flash_attention_backward(Tensor grad_out, Tensor query, Tensor key, Tensor value, Tensor out, Tensor logsumexp, Tensor cum_seq_q, Tensor cum_seq_k, SymInt max_q, SymInt max_k, float dropout_p, bool is_causal, Tensor philox_seed, Tensor philox_offset, *, float? scale=None) -> (Tensor grad_query, Tensor grad_key, Tensor grad_value)
   device_check: NoCheck
   variants: function
   dispatch:
-    CPU: _scaled_dot_product_flash_attention_backward_cpu
     CUDA: _scaled_dot_product_flash_attention_backward_cuda
     NestedTensorCUDA: _scaled_dot_product_flash_attention_backward_nested
 
+- func: _scaled_dot_product_flash_attention_for_cpu_backward(Tensor grad_out, Tensor query, Tensor key, Tensor value, Tensor out, Tensor logsumexp, float dropout_p, bool is_causal, *, Tensor? attn_mask=None, float? scale=None) -> (Tensor grad_query, Tensor grad_key, Tensor grad_value)
+  device_check: NoCheck
+  variants: function
+  dispatch:
+    CPU: _scaled_dot_product_flash_attention_cpu_backward
+
 - func: _scaled_dot_product_efficient_attention(Tensor query, Tensor key, Tensor value, Tensor? attn_bias, bool compute_log_sumexp, float dropout_p=0.0, bool is_causal=False, *, float? scale=None) -> (Tensor output, Tensor log_sumexp, Tensor philox_seed, Tensor philox_offset)
   dispatch:
     CUDA: _scaled_dot_product_efficient_attention_cuda
@@ -14493,26 +14728,36 @@
     CUDA: _scaled_dot_product_efficient_attention_backward_cuda
   tags: nondeterministic_seeded
 
-- func: _flash_attention_forward(Tensor query, Tensor key, Tensor value, Tensor? cum_seq_q, Tensor? cum_seq_k, SymInt max_q, SymInt max_k, float dropout_p, bool is_causal, bool return_debug_mask, *, float? scale=None) -> (Tensor output, Tensor softmax_logsumexp, Tensor philox_seed, Tensor philox_offset, Tensor debug_attn_mask)
+- func: _scaled_dot_product_cudnn_attention(Tensor query, Tensor key, Tensor value, float dropout_p=0.0, bool is_causal=False, bool return_debug_mask=False, *, float? scale=None) -> (Tensor output, Tensor logsumexp, Tensor cum_seq_q, Tensor cum_seq_k, SymInt max_q, SymInt max_k, Tensor philox_seed, Tensor philox_offset, Tensor debug_attn_mask)
+  dispatch:
+    CUDA: _scaled_dot_product_cudnn_attention_cuda
+  tags: nondeterministic_seeded
+
+- func: _scaled_dot_product_cudnn_attention_backward(Tensor grad_out, Tensor query, Tensor key, Tensor value, Tensor out, Tensor logsumexp, Tensor cum_seq_q, Tensor cum_seq_k, SymInt max_q, SymInt max_k, float dropout_p, bool is_causal, Tensor philox_seed, Tensor philox_offset, *, float? scale=None) -> (Tensor, Tensor, Tensor)
+  dispatch:
+    CUDA: _scaled_dot_product_cudnn_attention_backward_cuda
+  tags: nondeterministic_seeded
+
+- func: _flash_attention_forward(Tensor query, Tensor key, Tensor value, Tensor? cum_seq_q, Tensor? cum_seq_k, SymInt max_q, SymInt max_k, float dropout_p, bool is_causal, bool return_debug_mask, *, float? scale=None, SymInt? window_size_left=None, SymInt? window_size_right=None, Tensor? seqused_k=None, Tensor? alibi_slopes=None) -> (Tensor output, Tensor softmax_logsumexp, Tensor philox_seed, Tensor philox_offset, Tensor debug_attn_mask)
   variants: function
   dispatch:
     CUDA: _flash_attention_forward
   tags: nondeterministic_seeded
 
-- func: _flash_attention_backward(Tensor grad_out, Tensor query, Tensor key, Tensor value, Tensor out, Tensor logsumexp, Tensor cum_seq_q, Tensor cum_seq_k, SymInt max_q, SymInt max_k, float dropout_p, bool is_causal, Tensor philox_seed, Tensor philox_offset, *, float? scale=None) -> (Tensor, Tensor, Tensor)
+- func: _flash_attention_backward(Tensor grad_out, Tensor query, Tensor key, Tensor value, Tensor out, Tensor logsumexp, Tensor cum_seq_q, Tensor cum_seq_k, SymInt max_q, SymInt max_k, float dropout_p, bool is_causal, Tensor philox_seed, Tensor philox_offset, *, float? scale=None, SymInt? window_size_left=None, SymInt? window_size_right=None) -> (Tensor, Tensor, Tensor)
   device_check: NoCheck
   variants: function
   dispatch:
     CUDA: _flash_attention_backward
 
-# Returns ouput, logsumexp if compute_logsumexp
-- func: _efficient_attention_forward(Tensor query, Tensor key, Tensor value, Tensor? bias, Tensor? cu_seqlens_q, Tensor? cu_seqlens_k, int? max_seqlen_q, float dropout_p, int custom_mask_type, bool compute_log_sumexp=False, *, float? scale=None, Tensor? causal_diagonal=None, Tensor? seqlen_k=None) -> (Tensor output, Tensor logsumexp, Tensor philox_seed, Tensor philox_offset, SymInt max_seqlen_batch_q, SymInt max_seqlen_batch_k)
+# Returns output, logsumexp if compute_logsumexp
+- func: _efficient_attention_forward(Tensor query, Tensor key, Tensor value, Tensor? bias, Tensor? cu_seqlens_q, Tensor? cu_seqlens_k, SymInt? max_seqlen_q, SymInt? max_seqlen_k, float dropout_p, int custom_mask_type, bool compute_log_sumexp=False, *, float? scale=None, Tensor? seqlen_k=None, int? window_size=None) -> (Tensor output, Tensor logsumexp, Tensor philox_seed, Tensor philox_offset, SymInt max_seqlen_batch_q, SymInt max_seqlen_batch_k)
   variants: function
   dispatch:
     CUDA: _efficient_attention_forward
   tags: nondeterministic_seeded
 
-- func: _efficient_attention_backward(Tensor grad_out_, Tensor query, Tensor key, Tensor value, Tensor? bias, Tensor out, Tensor? cu_seqlens_q, Tensor? cu_seqlens_k, SymInt max_seqlen_q, SymInt max_seqlen_k, Tensor logsumexp, float dropout_p, Tensor philox_seed, Tensor philox_offset, int custom_mask_type, bool bias_requires_grad, *, float? scale=None, int? num_splits_key=None) -> (Tensor, Tensor, Tensor, Tensor)
+- func: _efficient_attention_backward(Tensor grad_out_, Tensor query, Tensor key, Tensor value, Tensor? bias, Tensor out, Tensor? cu_seqlens_q, Tensor? cu_seqlens_k, SymInt max_seqlen_q, SymInt max_seqlen_k, Tensor logsumexp, float dropout_p, Tensor philox_seed, Tensor philox_offset, int custom_mask_type, bool bias_requires_grad, *, float? scale=None, int? num_splits_key=None, int? window_size=None, bool shared_storage_dqdkdv=False) -> (Tensor, Tensor, Tensor, Tensor)
   device_check: NoCheck
   variants: function
   dispatch:
@@ -15312,11 +15557,11 @@
     CPU: foobar
   autogen: _foobar.out
 
-# Fused Optimizer CUDA kernels.
 - func: _fused_adam_(Tensor(a!)[] self, Tensor(b!)[] grads, Tensor(c!)[] exp_avgs, Tensor(d!)[] exp_avg_sqs, Tensor(e!)[] max_exp_avg_sqs, Tensor[] state_steps, *, float lr, float beta1, float beta2, float weight_decay, float eps, bool amsgrad, bool maximize, Tensor? grad_scale=None, Tensor? found_inf=None) -> ()
   # Unlike "foreach" functions, lists of tensors should be guaranteed to be on the same device (for now).
   variants: function
   dispatch:
+    CPU: _fused_adam_kernel_cpu_
     CUDA: _fused_adam_kernel_cuda_
   autogen: _fused_adam, _fused_adam.out
 
@@ -15326,6 +15571,7 @@
   device_check: NoCheck
   variants: function
   dispatch:
+    CPU: _fused_adam_kernel_cpu_
     CUDA: _fused_adam_kernel_cuda_
   autogen: _fused_adam.tensor_lr, _fused_adam.tensor_lr_out
 
@@ -15333,6 +15579,7 @@
   # Unlike "foreach" functions, lists of tensors should be guaranteed to be on the same device (for now).
   variants: function
   dispatch:
+    CPU: _fused_adamw_kernel_cpu_
     CUDA: _fused_adamw_kernel_cuda_
   autogen: _fused_adamw, _fused_adamw.out
 
@@ -15342,9 +15589,34 @@
   device_check: NoCheck
   variants: function
   dispatch:
+    CPU: _fused_adamw_kernel_cpu_
     CUDA: _fused_adamw_kernel_cuda_
   autogen: _fused_adamw.tensor_lr, _fused_adamw.tensor_lr_out
 
+- func: _fused_sgd_(Tensor(a!)[] self, Tensor(b!)[] grads, Tensor(c!)[] momentum_buffer_list, *, float weight_decay, float momentum, float lr, float dampening, bool nesterov, bool maximize, bool is_first_step, Tensor? grad_scale=None, Tensor? found_inf=None) -> ()
+  # Unlike "foreach" functions, lists of tensors should be guaranteed to be on the same device (for now).
+  variants: function
+  dispatch:
+    CPU: _fused_sgd_kernel_cpu_
+    CUDA: _fused_sgd_kernel_cuda_
+  autogen: _fused_sgd, _fused_sgd.out
+
+- func: _fused_sgd_.tensor_lr(Tensor(a!)[] self, Tensor(b!)[] grads, Tensor(c!)[] momentum_buffer_list, *, float weight_decay, float momentum, Tensor lr, float dampening, bool nesterov, bool maximize, bool is_first_step, Tensor? grad_scale=None, Tensor? found_inf=None) -> ()
+  # Unlike "foreach" functions, lists of tensors should be guaranteed to be on the same device (for now).
+  # but still skip the device check as the Tensor LR can be on CPU
+  device_check: NoCheck
+  variants: function
+  dispatch:
+    CPU: _fused_sgd_kernel_cpu_
+    CUDA: _fused_sgd_kernel_cuda_
+  autogen: _fused_sgd.tensor_lr, _fused_sgd.tensor_lr_out
+
+- func: _fused_adagrad_(Tensor(a!)[] self, Tensor(b!)[] grads, Tensor(c!)[] state_sums, Tensor(d!)[] state_steps, *, float lr, float lr_decay, float weight_decay, float eps, bool maximize, Tensor? grad_scale=None, Tensor? found_inf=None) -> ()
+  variants: function
+  dispatch:
+    CPU: _fused_adagrad_kernel_cpu_
+  autogen: _fused_adagrad, _fused_adagrad.out
+
 # This op is ONLY used by pytorch/XLA in functionalization, and should never show up in vanilla eager mode or in any pytorch tracing contexts.
 - func: _propagate_xla_data(Tensor input, Tensor output) -> ()
   variants: function