mindspore 2.7.0__cp310-cp310-manylinux1_x86_64.whl → 2.7.1__cp310-cp310-manylinux1_x86_64.whl

mindspore/ops/auto_generate/gen_ops_prim.py

@@ -41,7 +41,6 @@ from mindspore._c_expression import pyboost_inplace_mul
 from mindspore._c_expression import pyboost_mla
 from mindspore._c_expression import pyboost_upsample_bilinear2d_grad
 from mindspore._c_expression import pyboost_inplace_floor_divide
-from mindspore._c_expression import pyboost_dist_comm_isend
 from mindspore._c_expression import pyboost_arange
 from mindspore._c_expression import pyboost_tanh_grad
 from mindspore._c_expression import pyboost_lerp
@@ -57,7 +56,6 @@ from mindspore._c_expression import pyboost_split_tensor_view
 from mindspore._c_expression import pyboost_expand_dims
 from mindspore._c_expression import pyboost_empty
 from mindspore._c_expression import pyboost_paged_attention
-from mindspore._c_expression import pyboost_dist_comm_irecv
 from mindspore._c_expression import pyboost_avg_pool3d_grad_ext
 from mindspore._c_expression import pyboost_hsigmoid
 from mindspore._c_expression import pyboost_minimum
@@ -84,8 +82,8 @@ from mindspore._c_expression import pyboost_inplace_divmods
 from mindspore._c_expression import pyboost_chunk
 from mindspore._c_expression import pyboost_remainder_tensor_scalar
 from mindspore._c_expression import pyboost_smooth_l1_loss
-from mindspore._c_expression import pyboost_dist_comm_reduce_scatter_tensor
 from mindspore._c_expression import pyboost_hshrink_grad
+from mindspore._c_expression import pyboost_masked_fill_scalar
 from mindspore._c_expression import pyboost_type_as
 from mindspore._c_expression import pyboost_matmul_ext
 from mindspore._c_expression import pyboost_log2
@@ -99,7 +97,7 @@ from mindspore._c_expression import pyboost_masked_scatter
 from mindspore._c_expression import pyboost_randperm_ext
 from mindspore._c_expression import pyboost_binary_cross_entropy_with_logits_backward
 from mindspore._c_expression import pyboost_logsigmoid_grad
-from mindspore._c_expression import pyboost_dist_comm_gather_into_tensor
+from mindspore._c_expression import pyboost_inplace_masked_scatter
 from mindspore._c_expression import pyboost_rms_norm_grad
 from mindspore._c_expression import pyboost_adaptive_max_pool2d
 from mindspore._c_expression import pyboost_trunc
@@ -107,7 +105,6 @@ from mindspore._c_expression import pyboost_xlogy_scalar_self
 from mindspore._c_expression import pyboost_repeat_interleave_grad
 from mindspore._c_expression import pyboost_split_with_size_view
 from mindspore._c_expression import pyboost_batch_norm_elemt
-from mindspore._c_expression import pyboost_dist_comm_reduce_scatter_tensor_uneven
 from mindspore._c_expression import pyboost_matmul_reduce_scatter
 from mindspore._c_expression import pyboost_masked_select_grad
 from mindspore._c_expression import pyboost_reshape_and_cache
@@ -124,7 +121,6 @@ from mindspore._c_expression import pyboost_convolution_str
 from mindspore._c_expression import pyboost_adamw
 from mindspore._c_expression import pyboost_reflection_pad_1d_grad
 from mindspore._c_expression import pyboost_triangular_solve
-from mindspore._c_expression import pyboost_inner_comm_all_reduce
 from mindspore._c_expression import pyboost_median_dim
 from mindspore._c_expression import pyboost_inner_index
 from mindspore._c_expression import pyboost_atan_ext
@@ -132,23 +128,21 @@ from mindspore._c_expression import pyboost_isinf
 from mindspore._c_expression import pyboost_inplace_grouped_matmul_add
 from mindspore._c_expression import pyboost_linalg_qr
 from mindspore._c_expression import pyboost_equal
+from mindspore._c_expression import pyboost_inplace_sigmoid
 from mindspore._c_expression import pyboost_reflection_pad_2d_grad
 from mindspore._c_expression import pyboost_prompt_flash_attention
 from mindspore._c_expression import pyboost_constant_pad_nd
 from mindspore._c_expression import pyboost_expand_as
 from mindspore._c_expression import pyboost_equal_ext
-from mindspore._c_expression import pyboost_dist_comm_scatter_tensor
 from mindspore._c_expression import pyboost_floor
 from mindspore._c_expression import pyboost_round
 from mindspore._c_expression import pyboost_erf
 from mindspore._c_expression import pyboost_gcd
 from mindspore._c_expression import pyboost_isneginf
 from mindspore._c_expression import pyboost_inplace_div
-from mindspore._c_expression import pyboost_inner_comm_all_gather
 from mindspore._c_expression import pyboost_randn_like
 from mindspore._c_expression import pyboost_bitwise_or_scalar
 from mindspore._c_expression import pyboost_gather_d_grad_v2
-from mindspore._c_expression import pyboost_dist_comm_all_reduce
 from mindspore._c_expression import pyboost_inplace_fill_diagonal
 from mindspore._c_expression import pyboost_cast
 from mindspore._c_expression import pyboost_uniform_ext
@@ -159,7 +153,6 @@ from mindspore._c_expression import pyboost_group_norm_grad
 from mindspore._c_expression import pyboost_neg
 from mindspore._c_expression import pyboost_mse_loss_grad_ext
 from mindspore._c_expression import pyboost_lin_space_ext
-from mindspore._c_expression import pyboost_inner_comm_isend
 from mindspore._c_expression import pyboost_threshold_grad
 from mindspore._c_expression import pyboost_narrow
 from mindspore._c_expression import pyboost_logaddexp2
@@ -202,8 +195,6 @@ from mindspore._c_expression import pyboost_roll
 from mindspore._c_expression import pyboost_copy
 from mindspore._c_expression import pyboost_kthvalue
 from mindspore._c_expression import pyboost_rotary_position_embedding_grad
-from mindspore._c_expression import pyboost_dist_comm_reduce
-from mindspore._c_expression import pyboost_dist_comm_gather
 from mindspore._c_expression import pyboost_logical_or
 from mindspore._c_expression import pyboost_chunk_view
 from mindspore._c_expression import pyboost_add
@@ -228,7 +219,6 @@ from mindspore._c_expression import pyboost_rand_ext
 from mindspore._c_expression import pyboost_embedding
 from mindspore._c_expression import pyboost_clamp_scalar
 from mindspore._c_expression import pyboost_pow_scalar_tensor
-from mindspore._c_expression import pyboost_dist_comm_all_to_all_v
 from mindspore._c_expression import pyboost_softshrink
 from mindspore._c_expression import pyboost_cummax
 from mindspore._c_expression import pyboost_silu_grad
@@ -244,7 +234,6 @@ from mindspore._c_expression import pyboost_sin
 from mindspore._c_expression import pyboost_kl_div_grad
 from mindspore._c_expression import pyboost_broadcast_to_view
 from mindspore._c_expression import pyboost_inplace_bernoulli_scalar
-from mindspore._c_expression import pyboost_inner_comm_all_to_all_v
 from mindspore._c_expression import pyboost_silent_check_v2
 from mindspore._c_expression import pyboost_logsigmoid
 from mindspore._c_expression import pyboost_std
@@ -252,6 +241,7 @@ from mindspore._c_expression import pyboost_mv
 from mindspore._c_expression import pyboost_as_strided
 from mindspore._c_expression import pyboost_layer_norm_grad_ext
 from mindspore._c_expression import pyboost_masked_fill
+from mindspore._c_expression import pyboost_to_device
 from mindspore._c_expression import pyboost_relu_grad
 from mindspore._c_expression import pyboost_ones_like_ext
 from mindspore._c_expression import pyboost_cross_entropy_loss_grad
@@ -277,7 +267,6 @@ from mindspore._c_expression import pyboost_split_tensor
 from mindspore._c_expression import pyboost_inplace_sub_ext
 from mindspore._c_expression import pyboost_less_equal
 from mindspore._c_expression import pyboost_diag_ext
-from mindspore._c_expression import pyboost_inner_comm_irecv
 from mindspore._c_expression import pyboost_greater_equal
 from mindspore._c_expression import pyboost_batch_norm_reduce_grad
 from mindspore._c_expression import pyboost_log
@@ -287,6 +276,7 @@ from mindspore._c_expression import pyboost_gelu_grad_ext
 from mindspore._c_expression import pyboost_swiglu
 from mindspore._c_expression import pyboost_var_mean
 from mindspore._c_expression import pyboost_argmax_with_value
+from mindspore._c_expression import pyboost_real_view
 from mindspore._c_expression import pyboost_empty_like
 from mindspore._c_expression import pyboost_gelu_ext
 from mindspore._c_expression import pyboost_scatter_add_ext
@@ -324,7 +314,7 @@ from mindspore._c_expression import pyboost_topk_ext
 from mindspore._c_expression import pyboost_non_zero_ext
 from mindspore._c_expression import pyboost_avg_pool3d_ext
 from mindspore._c_expression import pyboost_reduce_min
-from mindspore._c_expression import pyboost_dist_comm_all_to_all_v_single
+from mindspore._c_expression import pyboost_inplace_index_fill_scalar
 from mindspore._c_expression import pyboost_conv3d_ext
 from mindspore._c_expression import pyboost_muls
 from mindspore._c_expression import pyboost_sub_scalar
@@ -356,8 +346,9 @@ from mindspore._c_expression import pyboost_sub_ext
 from mindspore._c_expression import pyboost_pow
 from mindspore._c_expression import pyboost_median_ext
 from mindspore._c_expression import pyboost_upsample_trilinear3d
+from mindspore._c_expression import pyboost_dequant_swiglu_quant
 from mindspore._c_expression import pyboost_upsample_linear1d
-from mindspore._c_expression import pyboost_dist_comm_all_gather_into_tensor
+from mindspore._c_expression import pyboost_inplace_sign
 from mindspore._c_expression import pyboost_group_norm
 from mindspore._c_expression import pyboost_moe_token_permute
 from mindspore._c_expression import pyboost_bernoulli_ext
@@ -379,14 +370,13 @@ from mindspore._c_expression import pyboost_scatter
 from mindspore._c_expression import pyboost_add_scalar
 from mindspore._c_expression import pyboost_layer_norm_ext
 from mindspore._c_expression import pyboost_reduce_max
-from mindspore._c_expression import pyboost_dist_comm_barrier
 from mindspore._c_expression import pyboost_inner_inplace_index_put
+from mindspore._c_expression import pyboost_imag_view
 from mindspore._c_expression import pyboost_log_softmax_grad
 from mindspore._c_expression import pyboost_upsample_trilinear3d_grad
 from mindspore._c_expression import pyboost_replication_pad_3d
 from mindspore._c_expression import pyboost_polar
 from mindspore._c_expression import pyboost_ffn_ext
-from mindspore._c_expression import pyboost_dist_comm_reduce_scatter
 from mindspore._c_expression import pyboost_bitwise_and_scalar
 from mindspore._c_expression import pyboost_rotary_position_embedding
 from mindspore._c_expression import pyboost_reverse_v2
@@ -395,6 +385,7 @@ from mindspore._c_expression import pyboost_batch_norm_stats
 from mindspore._c_expression import pyboost_ones
 from mindspore._c_expression import pyboost_add_layernorm_v2
 from mindspore._c_expression import pyboost_abs
+from mindspore._c_expression import pyboost_inplace_index_copy
 from mindspore._c_expression import pyboost_hardtanh_grad
 from mindspore._c_expression import pyboost_selu_grad
 from mindspore._c_expression import pyboost_new_ones
@@ -412,7 +403,6 @@ from mindspore._c_expression import pyboost_upsample_nearest2d_grad
 from mindspore._c_expression import pyboost_silu
 from mindspore._c_expression import pyboost_hardtanh
 from mindspore._c_expression import pyboost_transpose_view
-from mindspore._c_expression import pyboost_dist_comm_all_gather_into_tensor_uneven
 from mindspore._c_expression import pyboost_erfinv
 from mindspore._c_expression import pyboost_apply_rotary_pos_emb
 from mindspore._c_expression import pyboost_expm1
@@ -425,10 +415,10 @@ from mindspore._c_expression import pyboost_new_full
 from mindspore._c_expression import pyboost_floor_div_scalar
 from mindspore._c_expression import pyboost_soft_margin_loss
 from mindspore._c_expression import pyboost_acos_ext
-from mindspore._c_expression import pyboost_dist_comm_scatter
 from mindspore._c_expression import pyboost_tile
 from mindspore._c_expression import pyboost_index_select
 from mindspore._c_expression import pyboost_selu_ext
+from mindspore._c_expression import pyboost_inner_unique
 from mindspore._c_expression import pyboost_conv3d_padding
 from mindspore._c_expression import pyboost_prod_ext
 from mindspore._c_expression import pyboost_l1_loss_backward_ext
@@ -443,6 +433,7 @@ from mindspore._c_expression import pyboost_exp2
 from mindspore._c_expression import pyboost_slice
 from mindspore._c_expression import pyboost_non_zero
 from mindspore._c_expression import pyboost_nan_to_num
+from mindspore._c_expression import pyboost_inplace_index_fill_tensor
 from mindspore._c_expression import pyboost_softplus_ext
 from mindspore._c_expression import pyboost_hsigmoid_grad
 from mindspore._c_expression import pyboost_clone
@@ -451,7 +442,6 @@ from mindspore._c_expression import pyboost_reflection_pad_2d
 from mindspore._c_expression import pyboost_inplace_erfinv
 from mindspore._c_expression import pyboost_addmm
 from mindspore._c_expression import pyboost_adaptive_max_pool1d
-from mindspore._c_expression import pyboost_dist_comm_batch_isend_irecv
 from mindspore._c_expression import pyboost_mm_ext
 from mindspore._c_expression import pyboost_frac
 from mindspore._c_expression import pyboost_cosh
@@ -473,10 +463,8 @@ from mindspore._c_expression import pyboost_incre_flash_attention
 from mindspore._c_expression import pyboost_diagonal_view
 from mindspore._c_expression import pyboost_relu
 from mindspore._c_expression import pyboost_sigmoid
-from mindspore._c_expression import pyboost_dist_comm_broadcast
 from mindspore._c_expression import pyboost_inplace_log
 from mindspore._c_expression import pyboost_upsample_bicubic2d_grad
-from mindspore._c_expression import pyboost_inner_comm_reduce_scatter
 from mindspore._c_expression import pyboost_divs
 from mindspore._c_expression import pyboost_leaky_relu_grad_ext
 from mindspore._c_expression import pyboost_count_nonzero
@@ -504,6 +492,7 @@ from mindspore._c_expression import pyboost_unique2
 from mindspore._c_expression import pyboost_glu
 from mindspore._c_expression import pyboost_convolution
 from mindspore._c_expression import pyboost_dropout_grad_ext
+from mindspore._c_expression import pyboost_to_dtype
 from mindspore._c_expression import pyboost_moe_token_permute_grad
 from mindspore._c_expression import pyboost_inplace_elu
 from mindspore._c_expression import pyboost_max_pool_with_mask
@@ -522,11 +511,11 @@ from mindspore._c_expression import pyboost_greater_equal_scalar
 from mindspore._c_expression import pyboost_cummin_ext
 from mindspore._c_expression import pyboost_normal_tensor_float
 from mindspore._c_expression import pyboost_select_ext_view
+from mindspore._c_expression import pyboost_to_other
 from mindspore._c_expression import pyboost_speed_fusion_attention_grad
 from mindspore._c_expression import pyboost_batch_norm_grad_ext
 from mindspore._c_expression import pyboost_replication_pad_1d_grad
 from mindspore._c_expression import pyboost_inplace_random
-from mindspore._c_expression import pyboost_dist_comm_all_gather
 from mindspore._c_expression import pyboost_meshgrid
 from mindspore._c_expression import pyboost_randint_like
 from mindspore._c_expression import pyboost_add_layer_norm_grad
@@ -552,6 +541,32 @@ from mindspore._c_expression import pyboost_dynamic_quant_ext
 from mindspore._c_expression import pyboost_moe_compute_expert_tokens
 from mindspore._c_expression import pyboost_grouped_matmul
 from mindspore._c_expression import pyboost_moe_init_routing_v2
+from mindspore._c_expression import pyboost_dist_comm_isend
+from mindspore._c_expression import pyboost_dist_comm_irecv
+from mindspore._c_expression import pyboost_dist_comm_reduce_scatter_tensor
+from mindspore._c_expression import pyboost_dist_comm_gather_into_tensor
+from mindspore._c_expression import pyboost_dist_comm_reduce_scatter_tensor_uneven
+from mindspore._c_expression import pyboost_inner_comm_all_reduce
+from mindspore._c_expression import pyboost_dist_comm_scatter_tensor
+from mindspore._c_expression import pyboost_inner_comm_all_gather
+from mindspore._c_expression import pyboost_dist_comm_all_reduce
+from mindspore._c_expression import pyboost_inner_comm_isend
+from mindspore._c_expression import pyboost_dist_comm_reduce
+from mindspore._c_expression import pyboost_dist_comm_gather
+from mindspore._c_expression import pyboost_dist_comm_all_to_all_v
+from mindspore._c_expression import pyboost_inner_comm_all_to_all_v
+from mindspore._c_expression import pyboost_inner_comm_irecv
+from mindspore._c_expression import pyboost_dist_comm_all_to_all_v_single
+from mindspore._c_expression import pyboost_dist_comm_all_gather_into_tensor
+from mindspore._c_expression import pyboost_dist_comm_barrier
+from mindspore._c_expression import pyboost_dist_comm_reduce_scatter
+from mindspore._c_expression import pyboost_dist_comm_all_gather_into_tensor_uneven
+from mindspore._c_expression import pyboost_dist_comm_scatter
+from mindspore._c_expression import pyboost_dist_comm_batch_isend_irecv
+from mindspore._c_expression import pyboost_dist_comm_all_to_all_v_c
+from mindspore._c_expression import pyboost_dist_comm_broadcast
+from mindspore._c_expression import pyboost_inner_comm_reduce_scatter
+from mindspore._c_expression import pyboost_dist_comm_all_gather
 from mindspore._c_expression import pyboost_gmm_v2_backward
 from mindspore._c_expression import pyboost_gmm_v2
 from mindspore._c_expression import pyboost_gmm_v2_backward_fusion
@@ -563,6 +578,7 @@ from mindspore._c_expression import pyboost_func_max_pool2d
 from mindspore._c_expression import pyboost_moe_token_unpermute
 from mindspore._c_expression import pyboost_any
 from mindspore._c_expression import pyboost_gmm
+from mindspore._c_expression import pyboost_dropout2d_ext
 from mindspore._c_expression import pyboost_func_dropout_ext
 from mindspore._c_expression import pyboost_gmm_backward
 from mindspore._c_expression import pyboost_any_ext
@@ -1189,32 +1205,6 @@ class Mla(Primitive):
 mla_op=Mla()
 
 
-class EmbeddingFeatureMappingInsert(Primitive):
-    r"""
-    .. code-block::
-
-        prim = ops.EmbeddingFeatureMappingInsert()
-        out = prim(table_name, num, feature_id, offset_id)
-
-    is equivalent to
-
-    .. code-block::
-
-        ops.embedding_feature_mapping_insert(table_name, num, feature_id, offset_id)
-
-    Refer to :func:`mindspore.ops.embedding_feature_mapping_insert` for more details.
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, table_name, num, feature_id, offset_id):
-        return super().__call__(table_name, num, feature_id, offset_id)
-
-
-embedding_feature_mapping_insert_op=EmbeddingFeatureMappingInsert()
-
-
 class UpsampleBilinear2DGrad(Primitive):
     r"""
     Upsample the 2-D gradient data with bilinear interpolation algorithm.
@@ -1295,28 +1285,6 @@ class InplaceFloorDivide(Primitive):
 inplace_floor_divide_op=InplaceFloorDivide()
 
 
-class DistCommIsend(Primitive):
-    r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, input, dst, group, tag):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_dist_comm_isend(self, [input, dst, group, tag])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, input, dst, group, tag)
-        return res
-
-
-dist_comm_isend_op=DistCommIsend()
-
-
 class Arange(Primitive):
     r"""
     Creates a sequence of numbers that begins at `start` and extends by increments of
@@ -1900,20 +1868,21 @@ class Empty(Primitive):
         sig.make_sig('size'),
         sig.make_sig('dtype', default=None),
         sig.make_sig('device', default=None),
+        sig.make_sig('pin_memory', default=False),
     )
 
     @prim_arg_register
     def __init__(self):
         pass
 
-    def __call__(self, size, dtype=None, device=None):
+    def __call__(self, size, dtype=None, device=None, pin_memory=False):
         # Add for jit context.
         if jit_context() and jit_context().compiled:
             return jit_context().default_output()
-        res = pyboost_empty(self, [size, dtype if dtype is None else dtype_to_type_id('Empty', 'dtype', dtype), device if device is None else str_to_enum('Empty', 'device', device)])
+        res = pyboost_empty(self, [size, dtype if dtype is None else dtype_to_type_id('Empty', 'dtype', dtype), device if device is None else str_to_enum('Empty', 'device', device), pin_memory])
         # Add for jit context.
         if jit_context():
-            return jit_context().run_op(self, res, size, dtype if dtype is None else dtype_to_type_id('Empty', 'dtype', dtype), device if device is None else str_to_enum('Empty', 'device', device))
+            return jit_context().run_op(self, res, size, dtype if dtype is None else dtype_to_type_id('Empty', 'dtype', dtype), device if device is None else str_to_enum('Empty', 'device', device), pin_memory)
         return res
 
 
@@ -2013,28 +1982,6 @@ class PagedAttention(Primitive):
         return res
 
 
-class DistCommIrecv(Primitive):
-    r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, input, tag, src, group):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_dist_comm_irecv(self, [input, tag, src, group])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, input, tag, src, group)
-        return res
-
-
-dist_comm_irecv_op=DistCommIrecv()
-
-
 class AvgPool3DGradExt(Primitive):
     r"""
     
@@ -3315,28 +3262,6 @@ class MaximumGradGrad(Primitive):
         return super().__call__(x, y, dx, dy, self.grad_x, self.grad_y)
 
 
-class DistCommReduceScatterTensor(Primitive):
-    r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, other, input, rank_size, op_type, group):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_dist_comm_reduce_scatter_tensor(self, [other, input, rank_size, op_type, group])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, other, input, rank_size, op_type, group)
-        return res
-
-
-dist_comm_reduce_scatter_tensor_op=DistCommReduceScatterTensor()
-
-
 class HShrinkGrad(Primitive):
     r"""
     Computes gradients for HShrinkGrad operation.
@@ -3401,6 +3326,28 @@ class Conj(Primitive):
 conj_op=Conj()
 
 
+class MaskedFillScalar(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, input, mask, value):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_masked_fill_scalar(self, [input, mask, value])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input, mask, value)
+        return res
+
+
+masked_fill_scalar_op=MaskedFillScalar()
+
+
 class MatMulExt(Primitive):
     r"""
     .. code-block::
@@ -3999,51 +3946,6 @@ class LogSigmoidGrad(Primitive):
 logsigmoid_grad_op=LogSigmoidGrad()
 
 
-class EmbeddingApplyRmsprop(Primitive):
-    r"""
-    .. code-block::
-
-        prim = ops.EmbeddingApplyRmsprop()
-        out = prim(var_handle, lr, rho, momentum, epsilon, grad, keys, global_step, embedding_dim, mask_zero, padding_key, padding_key_mask, completion_key, completion_key_mask, _embedding_dim, _max_key_num)
-
-    is equivalent to
-
-    .. code-block::
-
-        ops.embedding_apply_rmsprop(var_handle, lr, rho, momentum, epsilon, grad, keys, global_step, embedding_dim, mask_zero, padding_key, padding_key_mask, completion_key, completion_key_mask, _embedding_dim, _max_key_num)
-
-    Refer to :func:`mindspore.ops.embedding_apply_rmsprop` for more details.
-    """
-    __mindspore_signature__ =  (
-        sig.make_sig('var_handle'),
-        sig.make_sig('lr'),
-        sig.make_sig('rho'),
-        sig.make_sig('momentum'),
-        sig.make_sig('epsilon'),
-        sig.make_sig('grad'),
-        sig.make_sig('keys'),
-        sig.make_sig('global_step'),
-        sig.make_sig('embedding_dim'),
-        sig.make_sig('mask_zero', default=(0,)),
-        sig.make_sig('padding_key', default=(0,)),
-        sig.make_sig('padding_key_mask', default=(1,)),
-        sig.make_sig('completion_key', default=(0,)),
-        sig.make_sig('completion_key_mask', default=(1,)),
-        sig.make_sig('_embedding_dim', default=1),
-        sig.make_sig('_max_key_num', default=1),
-    )
-
-    @prim_arg_register
-    def __init__(self):
-        self.add_prim_attr("_process_node_engine_id", 'PS')
-
-    def __call__(self, var_handle, lr, rho, momentum, epsilon, grad, keys, global_step, embedding_dim, mask_zero=(0,), padding_key=(0,), padding_key_mask=(1,), completion_key=(0,), completion_key_mask=(1,), _embedding_dim=1, _max_key_num=1):
-        return super().__call__(var_handle, lr, rho, momentum, epsilon, grad, keys, global_step, embedding_dim, mask_zero, padding_key, padding_key_mask, completion_key, completion_key_mask, _embedding_dim, _max_key_num)
-
-
-embedding_apply_rmsprop_op=EmbeddingApplyRmsprop()
-
-
 class GenerateEodMaskV2(Primitive):
     r"""
     
@@ -4074,26 +3976,58 @@ class GenerateEodMaskV2(Primitive):
 generate_eod_mask_v2_op=GenerateEodMaskV2()
 
 
-class DistCommGatherIntoTensor(Primitive):
+class InplaceMaskedScatter(Primitive):
     r"""
-    
+    Updates the value in the `input` with the `source` value according to the `mask`, and returns a Tensor.
+    The shape of `mask` and the `input` must be the same or `mask` is broadcastable.
+
+    Args:
+        input (Tensor): Input Tensor.
+        mask (Tensor[bool]): A bool tensor with a shape broadcastable to the `input`.
+        source (Tensor): A tensor with the same data type as the `input`. The number
+            of elements must be greater than or equal to the number of True's in `mask`.
+
+    Returns:
+        Tensor, with the same type and shape as the `input`.
+
+    Supported Platforms:
+        ``Ascend``
+
+    Examples:
+        >>> import numpy as np
+        >>> import mindspore
+        >>> from mindspore import Tensor, ops
+        >>> x = Tensor(np.array([1., 2., 3., 4.]), mindspore.float32)
+        >>> mask = Tensor(np.array([True, True, False, True]), mindspore.bool_)
+        >>> tensor = Tensor(np.array([5., 6., 7.]), mindspore.float32)
+        >>> output = ops.inplace_masked_scatter(x, mask, tensor)
+        >>> print(output)
+        [5. 6. 3. 7.]
+        >>> print(x)
+        [5. 6. 3. 7.]
     """
+    __mindspore_signature__ =  (
+        sig.make_sig('input', sig.sig_rw.RW_WRITE),
+        sig.make_sig('mask'),
+        sig.make_sig('source'),
+    )
+
     @prim_arg_register
     def __init__(self):
-        pass
+        self.add_prim_attr("side_effect_mem", True)
 
-    def __call__(self, other, input, rank_size, dst, rank_id, group):
+    def __call__(self, input, mask, source):
         # Add for jit context.
         if jit_context() and jit_context().compiled:
             return jit_context().default_output()
-        res = pyboost_dist_comm_gather_into_tensor(self, [other, input, rank_size, dst, rank_id, group])
+        res = pyboost_inplace_masked_scatter(self, [input, mask, source])
         # Add for jit context.
         if jit_context():
-            return jit_context().run_op(self, res, other, input, rank_size, dst, rank_id, group)
+            return jit_context().run_op(self, res, input, mask, source)
         return res
 
 
-dist_comm_gather_into_tensor_op=DistCommGatherIntoTensor()
+inplace_masked_scatter_op=InplaceMaskedScatter()
 
 
 class RmsNormGrad(Primitive):
@@ -4394,28 +4328,6 @@ class BatchNormElemt(Primitive):
 batch_norm_elemt_op=BatchNormElemt()
 
 
-class DistCommReduceScatterTensorUneven(Primitive):
-    r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, other, input, input_split_size, rank_size, op_type, group):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_dist_comm_reduce_scatter_tensor_uneven(self, [other, input, input_split_size, rank_size, op_type, group])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, other, input, input_split_size, rank_size, op_type, group)
-        return res
-
-
-dist_comm_reduce_scatter_tensor_uneven_op=DistCommReduceScatterTensorUneven()
-
-
 class GatherNd(Primitive):
     r"""
     .. code-block::
@@ -5291,28 +5203,6 @@ class TriangularSolve(Primitive):
 triangular_solve_op=TriangularSolve()
 
 
-class InnerCommAllReduce(Primitive):
-    r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, input, op_type, group):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_inner_comm_all_reduce(self, [input, op_type, group])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, input, op_type, group)
-        return res
-
-
-inner_comm_all_reduce_op=InnerCommAllReduce()
-
-
 class MedianDim(Primitive):
     r"""
     
@@ -5634,12 +5524,49 @@ class Equal(Primitive):
 equal_op=Equal()
 
 
-class ResizeNearestNeighborV2(Primitive):
+class InplaceSigmoid(Primitive):
     r"""
-    Resizes the input tensor to specific size by using the nearest neighbor algorithm.
+    .. code-block::
 
-    The nearest neighbor algorithm selects the value of the nearest point and does not consider the
-    values of neighboring points at all, yielding a piecewise-constant interpolant.
+        prim = ops.InplaceSigmoid()
+        out = prim(input)
+
+    is equivalent to
+
+    .. code-block::
+
+        ops.inplace_sigmoid(input)
+
+    Refer to :func:`mindspore.ops.inplace_sigmoid` for more details.
+    """
+    __mindspore_signature__ =  (
+        sig.make_sig('input', sig.sig_rw.RW_WRITE),
+    )
+
+    @prim_arg_register
+    def __init__(self):
+        self.add_prim_attr("side_effect_mem", True)
+
+    def __call__(self, input):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_inplace_sigmoid(self, [input])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input)
+        return res
+
+
+inplace_sigmoid_op=InplaceSigmoid()
+
+
+class ResizeNearestNeighborV2(Primitive):
+    r"""
+    Resizes the input tensor to specific size by using the nearest neighbor algorithm.
+
+    The nearest neighbor algorithm selects the value of the nearest point and does not consider the
+    values of neighboring points at all, yielding a piecewise-constant interpolant.
 
     Args:
         align_corners (bool, optional): If ``True`` , the centers of the 4 corner pixels of the input and output
@@ -6057,28 +5984,6 @@ class EqualExt(Primitive):
 equal_ext_op=EqualExt()
 
 
-class DistCommScatterTensor(Primitive):
-    r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, other, input, rank_size, src, rank_id, group):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_dist_comm_scatter_tensor(self, [other, input, rank_size, src, rank_id, group])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, other, input, rank_size, src, rank_id, group)
-        return res
-
-
-dist_comm_scatter_tensor_op=DistCommScatterTensor()
-
-
 class Floor(Primitive):
     r"""
     .. code-block::
@@ -6443,28 +6348,6 @@ class InplaceDiv(Primitive):
 inplace_div_op=InplaceDiv()
 
 
-class InnerCommAllGather(Primitive):
-    r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, input, rank_size, group):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_inner_comm_all_gather(self, [input, rank_size, group])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, input, rank_size, group)
-        return res
-
-
-inner_comm_all_gather_op=InnerCommAllGather()
-
-
 class EluGrad(Primitive):
     r"""
     Gradients of Elu operation.
@@ -6581,34 +6464,6 @@ class ACosGrad(Primitive):
 a_cos_grad_op=ACosGrad()
 
 
-class DistCommAllReduce(Primitive):
-    r"""
-    
-    """
-    __mindspore_signature__ =  (
-        sig.make_sig('input', sig.sig_rw.RW_WRITE),
-        sig.make_sig('op_type'),
-        sig.make_sig('group'),
-    )
-
-    @prim_arg_register
-    def __init__(self):
-        self.add_prim_attr("side_effect_mem", True)
-
-    def __call__(self, input, op_type, group):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_dist_comm_all_reduce(self, [input, op_type, group])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, input, op_type, group)
-        return res
-
-
-dist_comm_all_reduce_op=DistCommAllReduce()
-
-
 class InplaceFillDiagonal(Primitive):
     r"""
     .. code-block::
@@ -6848,48 +6703,6 @@ class Addmv(Primitive):
 addmv_op=Addmv()
 
 
-class EmbeddingApplyAdaGrad(Primitive):
-    r"""
-    .. code-block::
-
-        prim = ops.EmbeddingApplyAdaGrad()
-        out = prim(var_handle, lr, grad, keys, global_step, embedding_dim, mask_zero, padding_key, padding_key_mask, completion_key, completion_key_mask, _embedding_dim, _max_key_num)
-
-    is equivalent to
-
-    .. code-block::
-
-        ops.embedding_apply_ada_grad(var_handle, lr, grad, keys, global_step, embedding_dim, mask_zero, padding_key, padding_key_mask, completion_key, completion_key_mask, _embedding_dim, _max_key_num)
-
-    Refer to :func:`mindspore.ops.embedding_apply_ada_grad` for more details.
-    """
-    __mindspore_signature__ =  (
-        sig.make_sig('var_handle'),
-        sig.make_sig('lr'),
-        sig.make_sig('grad'),
-        sig.make_sig('keys'),
-        sig.make_sig('global_step'),
-        sig.make_sig('embedding_dim'),
-        sig.make_sig('mask_zero', default=(0,)),
-        sig.make_sig('padding_key', default=(0,)),
-        sig.make_sig('padding_key_mask', default=(1,)),
-        sig.make_sig('completion_key', default=(0,)),
-        sig.make_sig('completion_key_mask', default=(1,)),
-        sig.make_sig('_embedding_dim', default=1),
-        sig.make_sig('_max_key_num', default=1),
-    )
-
-    @prim_arg_register
-    def __init__(self):
-        self.add_prim_attr("_process_node_engine_id", 'PS')
-
-    def __call__(self, var_handle, lr, grad, keys, global_step, embedding_dim, mask_zero=(0,), padding_key=(0,), padding_key_mask=(1,), completion_key=(0,), completion_key_mask=(1,), _embedding_dim=1, _max_key_num=1):
-        return super().__call__(var_handle, lr, grad, keys, global_step, embedding_dim, mask_zero, padding_key, padding_key_mask, completion_key, completion_key_mask, _embedding_dim, _max_key_num)
-
-
-embedding_apply_ada_grad_op=EmbeddingApplyAdaGrad()
-
-
 class UpsampleBicubic2D(Primitive):
     r"""
     
@@ -7084,28 +6897,6 @@ class LinSpaceExt(Primitive):
 lin_space_ext_op=LinSpaceExt()
 
 
-class InnerCommIsend(Primitive):
-    r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, input, dst, group, tag):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_inner_comm_isend(self, [input, dst, group, tag])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, input, dst, group, tag)
-        return res
-
-
-inner_comm_isend_op=InnerCommIsend()
-
-
 class Argmax(Primitive):
     r"""
     Returns the indices of the maximum value along a specified `axis` of a Tensor.
@@ -7760,40 +7551,6 @@ class SearchSorted(Primitive):
         return res
 
 
-class EmbeddingFeatureMappingFileSize(Primitive):
-    r"""
-    .. code-block::
-
-        prim = ops.EmbeddingFeatureMappingFileSize()
-        out = prim(file_path, table_name, global_step, embedding_dim, only_offset_flag)
-
-    is equivalent to
-
-    .. code-block::
-
-        ops.embedding_feature_mapping_file_size(file_path, table_name, global_step, embedding_dim, only_offset_flag)
-
-    Refer to :func:`mindspore.ops.embedding_feature_mapping_file_size` for more details.
-    """
-    __mindspore_signature__ =  (
-        sig.make_sig('file_path'),
-        sig.make_sig('table_name'),
-        sig.make_sig('global_step'),
-        sig.make_sig('embedding_dim'),
-        sig.make_sig('only_offset_flag', default=True),
-    )
-
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, file_path, table_name, global_step, embedding_dim, only_offset_flag=True):
-        return super().__call__(file_path, table_name, global_step, embedding_dim, only_offset_flag)
-
-
-embedding_feature_mapping_file_size_op=EmbeddingFeatureMappingFileSize()
-
-
 class InplaceScatterSrc(Primitive):
     r"""
     InplaceScatterSrc is for scatter_ when using a Tensor as the source element without reduce.
@@ -8770,32 +8527,6 @@ class ArgMaxExt(Primitive):
 argmax_ext_op=ArgMaxExt()
 
 
-class EmbeddingFeatureMappingTableSize(Primitive):
-    r"""
-    .. code-block::
-
-        prim = ops.EmbeddingFeatureMappingTableSize()
-        out = prim(table_name)
-
-    is equivalent to
-
-    .. code-block::
-
-        ops.embedding_feature_mapping_table_size(table_name)
-
-    Refer to :func:`mindspore.ops.embedding_feature_mapping_table_size` for more details.
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, table_name):
-        return super().__call__(table_name)
-
-
-embedding_feature_mapping_table_size_op=EmbeddingFeatureMappingTableSize()
-
-
 class Atan(Primitive):
     r"""
     .. code-block::
@@ -9049,32 +8780,6 @@ class ReduceAny(Primitive):
         return res
 
 
-class EmbeddingFeatureMappingV2(Primitive):
-    r"""
-    .. code-block::
-
-        prim = ops.EmbeddingFeatureMappingV2()
-        out = prim(table_name, feature_id, table_total_size, table_actual_size)
-
-    is equivalent to
-
-    .. code-block::
-
-        ops.embedding_feature_mapping_v2(table_name, feature_id, table_total_size, table_actual_size)
-
-    Refer to :func:`mindspore.ops.embedding_feature_mapping_v2` for more details.
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, table_name, feature_id, table_total_size, table_actual_size):
-        return super().__call__(table_name, feature_id, table_total_size, table_actual_size)
-
-
-embedding_feature_mapping_v2_op=EmbeddingFeatureMappingV2()
-
-
 class Eye(Primitive):
     r"""
     Returns a tensor with ones on the diagonal and zeros in the rest.
@@ -9418,91 +9123,40 @@ class Angle(Primitive):
 angle_op=Angle()
 
 
-class DistCommReduce(Primitive):
+class FFTShift(Primitive):
     r"""
-    
+    .. code-block::
+
+        prim = ops.FFTShift()
+        out = prim(input, dim)
+
+    is equivalent to
+
+    .. code-block::
+
+        ops.fftshift(input, dim)
+
+    Refer to :func:`mindspore.ops.fftshift` for more details.
     """
     __mindspore_signature__ =  (
-        sig.make_sig('input', sig.sig_rw.RW_WRITE),
-        sig.make_sig('op_type'),
-        sig.make_sig('dst'),
-        sig.make_sig('group'),
+        sig.make_sig('input'),
+        sig.make_sig('dim', default=None),
     )
 
     @prim_arg_register
     def __init__(self):
-        self.add_prim_attr("side_effect_mem", True)
+        pass
 
-    def __call__(self, input, op_type, dst, group):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_dist_comm_reduce(self, [input, op_type, dst, group])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, input, op_type, dst, group)
-        return res
+    def __call__(self, input, dim=None):
+        return super().__call__(input, dim)
 
 
-dist_comm_reduce_op=DistCommReduce()
+fftshift_op=FFTShift()
 
 
-class DistCommGather(Primitive):
+class LogicalOr(Primitive):
     r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, input, gather_list, rank_size, dst, rank_id, group):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_dist_comm_gather(self, [input, gather_list, rank_size, dst, rank_id, group])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, input, gather_list, rank_size, dst, rank_id, group)
-        return res
-
-
-dist_comm_gather_op=DistCommGather()
-
-
-class FFTShift(Primitive):
-    r"""
-    .. code-block::
-
-        prim = ops.FFTShift()
-        out = prim(input, dim)
-
-    is equivalent to
-
-    .. code-block::
-
-        ops.fftshift(input, dim)
-
-    Refer to :func:`mindspore.ops.fftshift` for more details.
-    """
-    __mindspore_signature__ =  (
-        sig.make_sig('input'),
-        sig.make_sig('dim', default=None),
-    )
-
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, input, dim=None):
-        return super().__call__(input, dim)
-
-
-fftshift_op=FFTShift()
-
-
-class LogicalOr(Primitive):
-    r"""
-    Computes the "logical OR" of two tensors element-wise.
+    Computes the "logical OR" of two tensors element-wise.
 
     Refer to :func:`mindspore.ops.logical_or` for more details.
 
@@ -11051,28 +10705,6 @@ class PowScalarTensor(Primitive):
 pow_scalar_tensor_op=PowScalarTensor()
 
 
-class DistCommAllToAllV(Primitive):
-    r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, other, input, group, send_numel_list, recv_numel_list, rank_size):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_dist_comm_all_to_all_v(self, [other, input, group, send_numel_list, recv_numel_list, rank_size])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, other, input, group, send_numel_list, recv_numel_list, rank_size)
-        return res
-
-
-dist_comm_all_to_all_v_op=DistCommAllToAllV()
-
-
 class SoftShrink(Primitive):
     r"""
     .. code-block::
@@ -11595,28 +11227,6 @@ class InplaceBernoulliScalar(Primitive):
 inplace_bernoulli_scalar_op=InplaceBernoulliScalar()
 
 
-class InnerCommAllToAllV(Primitive):
-    r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, input, group, send_numel_list, recv_numel_list, rank_size, split_sizes_empty):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_inner_comm_all_to_all_v(self, [input, group, send_numel_list, recv_numel_list, rank_size, split_sizes_empty])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, input, group, send_numel_list, recv_numel_list, rank_size, split_sizes_empty)
-        return res
-
-
-inner_comm_all_to_all_v_op=InnerCommAllToAllV()
-
-
 class SilentCheckV2(Primitive):
     r"""
     
@@ -11951,7 +11561,7 @@ class ReduceSum(Primitive):
           [-rank(`x`), rank(`x`)).
 
     Outputs:
-        Tensor, has the same dtype as the `x`.
+        Tensor.
 
         - If `axis` is ``()`` , `keep_dims` is ``False`` , and `skip_mode` is ``False`` ,
           the output is a 0-D tensor representing the sum of all elements in the input tensor.
@@ -12124,6 +11734,36 @@ class ACos(Primitive):
 acos_op=ACos()
 
 
+class ToDevice(Primitive):
+    r"""
+    
+    """
+    __mindspore_signature__ =  (
+        sig.make_sig('input'),
+        sig.make_sig('device', default=None),
+        sig.make_sig('dtype', default=None),
+        sig.make_sig('non_blocking', default=False),
+        sig.make_sig('copy', default=False),
+    )
+
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, input, device=None, dtype=None, non_blocking=False, copy=False):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_to_device(self, [input, device if device is None else str_to_enum('ToDevice', 'device', device), dtype if dtype is None else dtype_to_type_id('ToDevice', 'dtype', dtype), non_blocking, copy])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input, device if device is None else str_to_enum('ToDevice', 'device', device), dtype if dtype is None else dtype_to_type_id('ToDevice', 'dtype', dtype), non_blocking, copy)
+        return res
+
+
+to_device_op=ToDevice()
+
+
 class ReluGrad(Primitive):
     r"""
     Computes gradient for the ReLU activation.
@@ -13121,38 +12761,6 @@ class Trace(Primitive):
 trace_op=Trace()
 
 
-class EmbeddingTableEvict(Primitive):
-    r"""
-    .. code-block::
-
-        prim = ops.EmbeddingTableEvict()
-        out = prim(var_handle, global_step, steps_to_live)
-
-    is equivalent to
-
-    .. code-block::
-
-        ops.embedding_table_evict(var_handle, global_step, steps_to_live)
-
-    Refer to :func:`mindspore.ops.embedding_table_evict` for more details.
-    """
-    __mindspore_signature__ =  (
-        sig.make_sig('var_handle'),
-        sig.make_sig('global_step'),
-        sig.make_sig('steps_to_live', default=0),
-    )
-
-    @prim_arg_register
-    def __init__(self):
-        self.add_prim_attr("_process_node_engine_id", 'PS')
-
-    def __call__(self, var_handle, global_step, steps_to_live=0):
-        return super().__call__(var_handle, global_step, steps_to_live)
-
-
-embedding_table_evict_op=EmbeddingTableEvict()
-
-
 class Qr(Primitive):
     r"""
     Returns the QR decomposition of one or more matrices.
@@ -13279,28 +12887,6 @@ class DiagExt(Primitive):
 diag_ext_op=DiagExt()
 
 
-class InnerCommIrecv(Primitive):
-    r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, tag, src, shape, group, dtype):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_inner_comm_irecv(self, [tag, src, shape, group, dtype_to_type_id('InnerCommIrecv', 'dtype', dtype)])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, tag, src, shape, group, dtype_to_type_id('InnerCommIrecv', 'dtype', dtype))
-        return res
-
-
-inner_comm_irecv_op=InnerCommIrecv()
-
-
 class GreaterEqual(Primitive):
     r"""
     .. code-block::
@@ -13336,48 +12922,6 @@ class GreaterEqual(Primitive):
 greater_equal_op=GreaterEqual()
 
 
-class EmbeddingApplySgd(Primitive):
-    r"""
-    .. code-block::
-
-        prim = ops.EmbeddingApplySgd()
-        out = prim(var_handle, lr, grad, keys, global_step, embedding_dim, mask_zero, padding_key, padding_key_mask, completion_key, completion_key_mask, _embedding_dim, _max_key_num)
-
-    is equivalent to
-
-    .. code-block::
-
-        ops.embedding_apply_sgd(var_handle, lr, grad, keys, global_step, embedding_dim, mask_zero, padding_key, padding_key_mask, completion_key, completion_key_mask, _embedding_dim, _max_key_num)
-
-    Refer to :func:`mindspore.ops.embedding_apply_sgd` for more details.
-    """
-    __mindspore_signature__ =  (
-        sig.make_sig('var_handle'),
-        sig.make_sig('lr'),
-        sig.make_sig('grad'),
-        sig.make_sig('keys'),
-        sig.make_sig('global_step'),
-        sig.make_sig('embedding_dim'),
-        sig.make_sig('mask_zero', default=(0,)),
-        sig.make_sig('padding_key', default=(0,)),
-        sig.make_sig('padding_key_mask', default=(1,)),
-        sig.make_sig('completion_key', default=(0,)),
-        sig.make_sig('completion_key_mask', default=(1,)),
-        sig.make_sig('_embedding_dim', default=1),
-        sig.make_sig('_max_key_num', default=1),
-    )
-
-    @prim_arg_register
-    def __init__(self):
-        self.add_prim_attr("_process_node_engine_id", 'PS')
-
-    def __call__(self, var_handle, lr, grad, keys, global_step, embedding_dim, mask_zero=(0,), padding_key=(0,), padding_key_mask=(1,), completion_key=(0,), completion_key_mask=(1,), _embedding_dim=1, _max_key_num=1):
-        return super().__call__(var_handle, lr, grad, keys, global_step, embedding_dim, mask_zero, padding_key, padding_key_mask, completion_key, completion_key_mask, _embedding_dim, _max_key_num)
-
-
-embedding_apply_sgd_op=EmbeddingApplySgd()
-
-
 class IDCTN(Primitive):
     r"""
     
@@ -13772,6 +13316,28 @@ class ArgMaxWithValue(Primitive):
         return res
 
 
+class RealView(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, input):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_real_view(self, [input])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input)
+        return res
+
+
+real_view_op=RealView()
+
+
 class EmptyLike(Primitive):
     r"""
     
@@ -13780,20 +13346,21 @@ class EmptyLike(Primitive):
         sig.make_sig('input'),
         sig.make_sig('dtype', default=None),
         sig.make_sig('device', default=None),
+        sig.make_sig('pin_memory', default=False),
     )
 
     @prim_arg_register
     def __init__(self):
         pass
 
-    def __call__(self, input, dtype=None, device=None):
+    def __call__(self, input, dtype=None, device=None, pin_memory=False):
         # Add for jit context.
         if jit_context() and jit_context().compiled:
             return jit_context().default_output()
-        res = pyboost_empty_like(self, [input, dtype if dtype is None else dtype_to_type_id('EmptyLike', 'dtype', dtype), device if device is None else str_to_enum('EmptyLike', 'device', device)])
+        res = pyboost_empty_like(self, [input, dtype if dtype is None else dtype_to_type_id('EmptyLike', 'dtype', dtype), device if device is None else str_to_enum('EmptyLike', 'device', device), pin_memory])
         # Add for jit context.
         if jit_context():
-            return jit_context().run_op(self, res, input, dtype if dtype is None else dtype_to_type_id('EmptyLike', 'dtype', dtype), device if device is None else str_to_enum('EmptyLike', 'device', device))
+            return jit_context().run_op(self, res, input, dtype if dtype is None else dtype_to_type_id('EmptyLike', 'dtype', dtype), device if device is None else str_to_enum('EmptyLike', 'device', device), pin_memory)
         return res
 
 
@@ -14439,6 +14006,32 @@ class InplaceScatterValueReduce(Primitive):
 inplace_scatter_value_reduce_op=InplaceScatterValueReduce()
 
 
+class FormatCast(Primitive):
+    r"""
+    .. code-block::
+
+        prim = ops.FormatCast()
+        out = prim(input, acl_format)
+
+    is equivalent to
+
+    .. code-block::
+
+        ops.format_cast(input, acl_format)
+
+    Refer to :func:`mindspore.ops.format_cast` for more details.
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, input, acl_format):
+        return super().__call__(input, acl_format)
+
+
+format_cast_op=FormatCast()
+
+
 class Rsqrt(Primitive):
     r"""
     .. code-block::
@@ -14506,13 +14099,13 @@ class Transpose(Primitive):
     .. code-block::
 
         prim = ops.Transpose()
-        out = prim(input, input_perm)
+        out = prim(input, dims)
 
     is equivalent to
 
     .. code-block::
 
-        ops.transpose(input, input_perm)
+        ops.transpose(input, dims)
 
     Refer to :func:`mindspore.ops.transpose` for more details.
     """
@@ -14520,14 +14113,14 @@ class Transpose(Primitive):
     def __init__(self):
         pass
 
-    def __call__(self, input, input_perm):
+    def __call__(self, input, dims):
         # Add for jit context.
         if jit_context() and jit_context().compiled:
             return jit_context().default_output()
-        res = pyboost_transpose(self, [input, input_perm])
+        res = pyboost_transpose(self, [input, dims])
         # Add for jit context.
         if jit_context():
-            return jit_context().run_op(self, res, input, input_perm)
+            return jit_context().run_op(self, res, input, dims)
         return res
 
 
@@ -15798,26 +15391,33 @@ class ReduceMin(Primitive):
         return res
 
 
-class DistCommAllToAllVSingle(Primitive):
+class InplaceIndexFillScalar(Primitive):
     r"""
     
     """
+    __mindspore_signature__ =  (
+        sig.make_sig('input', sig.sig_rw.RW_WRITE),
+        sig.make_sig('dim'),
+        sig.make_sig('index'),
+        sig.make_sig('value'),
+    )
+
     @prim_arg_register
     def __init__(self):
         pass
 
-    def __call__(self, other, input, group, send_numel_list, recv_numel_list, rank_size, split_sizes_empty):
+    def __call__(self, input, dim, index, value):
         # Add for jit context.
         if jit_context() and jit_context().compiled:
             return jit_context().default_output()
-        res = pyboost_dist_comm_all_to_all_v_single(self, [other, input, group, send_numel_list, recv_numel_list, rank_size, split_sizes_empty])
+        res = pyboost_inplace_index_fill_scalar(self, [input, dim, index, value])
         # Add for jit context.
         if jit_context():
-            return jit_context().run_op(self, res, other, input, group, send_numel_list, recv_numel_list, rank_size, split_sizes_empty)
+            return jit_context().run_op(self, res, input, dim, index, value)
         return res
 
 
-dist_comm_all_to_all_v_single_op=DistCommAllToAllVSingle()
+inplace_index_fill_scalar_op=InplaceIndexFillScalar()
 
 
 class Conv3DExt(Primitive):
@@ -16772,7 +16372,7 @@ class Squeeze(Primitive):
         This is an experimental API that is subject to change or deletion.
 
     Args:
-        axis (Union[int, tuple(int)]): Specifies the dimension indexes of shape to be removed, which will remove
+        axis (Union[int, tuple(int), list(int)]): Specifies the dimension indexes of shape to be removed, which will remove
             all the dimensions of size 1 in the given axis parameter. If specified, it must be int32 or int64.
             Default: ``()`` .
 
@@ -16799,7 +16399,7 @@ class Squeeze(Primitive):
     """
     @prim_arg_register
     def __init__(self, axis=()):
-        self._set_prim_arg("axis", type_it('Squeeze', 'axis', axis, OpDtype.DT_INT, OpDtype.DT_TUPLE_INT))
+        self._set_prim_arg("axis", type_it('Squeeze', 'axis', axis, (OpDtype.DT_INT, OpDtype.DT_LIST_INT), OpDtype.DT_TUPLE_INT))
 
     def __call__(self, input):
         # Add for jit context.
@@ -17187,37 +16787,82 @@ class UpsampleTrilinear3D(Primitive):
         return res
 
 
-class ResizeNearestNeighborV2Grad(Primitive):
+class DequantSwigluQuant(Primitive):
     r"""
-    Compute gradient of `ResizeNearestNeighborV2` operator.
+    .. code-block::
 
-    Args:
-        grads (Tensor): A 4-D Tensor with shape [batch, channel, height, width].
-        size (Union[tuple[int], Tensor]): The size for the input image. 2 elements: [`height, width`].
-        align_corners (bool): Whether the centers of the 4 corner pixels of the input
-            and output tensors are aligned. Default: ``False``.
-        half_pixel_centers (bool): Default: ``False``.
+        prim = ops.DequantSwigluQuant()
+        out = prim(x, weight_scale, activation_scale, bias, quant_scale, quant_offset, group_index, activate_left, quant_mode)
 
-    Outputs:
-        A 4-D Tensor , with the same shape and data type as `image`.
-    """
-    @prim_arg_register
-    def __init__(self, align_corners=False, half_pixel_centers=False):
-        self._set_prim_arg("align_corners", align_corners)
-        self._set_prim_arg("half_pixel_centers", half_pixel_centers)
+    is equivalent to
 
-    def __call__(self, grads, size):
-        return super().__call__(grads, size, self.align_corners, self.half_pixel_centers)
+    .. code-block::
 
+        ops.dequant_swiglu_quant(x, weight_scale, activation_scale, bias, quant_scale, quant_offset, group_index, activate_left, quant_mode)
 
-class UpsampleLinear1D(Primitive):
-    r"""
-    
+    Refer to :func:`mindspore.ops.dequant_swiglu_quant` for more details.
     """
     __mindspore_signature__ =  (
         sig.make_sig('x'),
-        sig.make_sig('output_size', default=None),
-        sig.make_sig('scales', default=None),
+        sig.make_sig('weight_scale'),
+        sig.make_sig('activation_scale'),
+        sig.make_sig('bias', default=None),
+        sig.make_sig('quant_scale', default=None),
+        sig.make_sig('quant_offset', default=None),
+        sig.make_sig('group_index', default=None),
+        sig.make_sig('activate_left', default=False),
+        sig.make_sig('quant_mode', default='static'),
+    )
+
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, x, weight_scale, activation_scale, bias=None, quant_scale=None, quant_offset=None, group_index=None, activate_left=False, quant_mode='static'):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_dequant_swiglu_quant(self, [x, weight_scale, activation_scale, bias, quant_scale, quant_offset, group_index, activate_left, str_to_enum('DequantSwigluQuant', 'quant_mode', quant_mode)])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, x, weight_scale, activation_scale, bias, quant_scale, quant_offset, group_index, activate_left, str_to_enum('DequantSwigluQuant', 'quant_mode', quant_mode))
+        return res
+
+
+dequant_swiglu_quant_op=DequantSwigluQuant()
+
+
+class ResizeNearestNeighborV2Grad(Primitive):
+    r"""
+    Compute gradient of `ResizeNearestNeighborV2` operator.
+
+    Args:
+        grads (Tensor): A 4-D Tensor with shape [batch, channel, height, width].
+        size (Union[tuple[int], Tensor]): The size for the input image. 2 elements: [`height, width`].
+        align_corners (bool): Whether the centers of the 4 corner pixels of the input
+            and output tensors are aligned. Default: ``False``.
+        half_pixel_centers (bool): Default: ``False``.
+
+    Outputs:
+        A 4-D Tensor , with the same shape and data type as `image`.
+    """
+    @prim_arg_register
+    def __init__(self, align_corners=False, half_pixel_centers=False):
+        self._set_prim_arg("align_corners", align_corners)
+        self._set_prim_arg("half_pixel_centers", half_pixel_centers)
+
+    def __call__(self, grads, size):
+        return super().__call__(grads, size, self.align_corners, self.half_pixel_centers)
+
+
+class UpsampleLinear1D(Primitive):
+    r"""
+    
+    """
+    __mindspore_signature__ =  (
+        sig.make_sig('x'),
+        sig.make_sig('output_size', default=None),
+        sig.make_sig('scales', default=None),
         sig.make_sig('align_corners', default=False),
     )
 
@@ -17239,26 +16884,41 @@ class UpsampleLinear1D(Primitive):
 upsample_linear1d_op=UpsampleLinear1D()
 
 
-class DistCommAllGatherIntoTensor(Primitive):
+class InplaceSign(Primitive):
     r"""
-    
+    .. code-block::
+
+        prim = ops.InplaceSign()
+        out = prim(input)
+
+    is equivalent to
+
+    .. code-block::
+
+        ops.inplace_sign(input)
+
+    Refer to :func:`mindspore.ops.inplace_sign` for more details.
     """
+    __mindspore_signature__ =  (
+        sig.make_sig('input', sig.sig_rw.RW_WRITE),
+    )
+
     @prim_arg_register
     def __init__(self):
-        pass
+        self.add_prim_attr("side_effect_mem", True)
 
-    def __call__(self, other, input, rank_size, group):
+    def __call__(self, input):
         # Add for jit context.
         if jit_context() and jit_context().compiled:
             return jit_context().default_output()
-        res = pyboost_dist_comm_all_gather_into_tensor(self, [other, input, rank_size, group])
+        res = pyboost_inplace_sign(self, [input])
         # Add for jit context.
         if jit_context():
-            return jit_context().run_op(self, res, other, input, rank_size, group)
+            return jit_context().run_op(self, res, input)
         return res
 
 
-dist_comm_all_gather_into_tensor_op=DistCommAllGatherIntoTensor()
+inplace_sign_op=InplaceSign()
 
 
 class GroupNorm(Primitive):
@@ -17625,56 +17285,6 @@ class AvgPool2DGrad(Primitive):
 avg_pool2d_grad_op=AvgPool2DGrad()
 
 
-class EmbeddingApplyAdamW(Primitive):
-    r"""
-    .. code-block::
-
-        prim = ops.EmbeddingApplyAdamW()
-        out = prim(var_handle, beta1_power, beta2_power, lr, weight_decay, beta1, beta2, epsilon, grad, keys, max_grad_norm, global_step, embedding_dim, ams_grad, mask_zero, padding_key, padding_key_mask, completion_key, completion_key_mask, _embedding_dim, _max_key_num)
-
-    is equivalent to
-
-    .. code-block::
-
-        ops.embedding_apply_adam_w(var_handle, beta1_power, beta2_power, lr, weight_decay, beta1, beta2, epsilon, grad, keys, max_grad_norm, global_step, embedding_dim, ams_grad, mask_zero, padding_key, padding_key_mask, completion_key, completion_key_mask, _embedding_dim, _max_key_num)
-
-    Refer to :func:`mindspore.ops.embedding_apply_adam_w` for more details.
-    """
-    __mindspore_signature__ =  (
-        sig.make_sig('var_handle'),
-        sig.make_sig('beta1_power'),
-        sig.make_sig('beta2_power'),
-        sig.make_sig('lr'),
-        sig.make_sig('weight_decay'),
-        sig.make_sig('beta1'),
-        sig.make_sig('beta2'),
-        sig.make_sig('epsilon'),
-        sig.make_sig('grad'),
-        sig.make_sig('keys'),
-        sig.make_sig('max_grad_norm'),
-        sig.make_sig('global_step'),
-        sig.make_sig('embedding_dim'),
-        sig.make_sig('ams_grad', default=(0,)),
-        sig.make_sig('mask_zero', default=(0,)),
-        sig.make_sig('padding_key', default=(0,)),
-        sig.make_sig('padding_key_mask', default=(1,)),
-        sig.make_sig('completion_key', default=(0,)),
-        sig.make_sig('completion_key_mask', default=(1,)),
-        sig.make_sig('_embedding_dim', default=1),
-        sig.make_sig('_max_key_num', default=1),
-    )
-
-    @prim_arg_register
-    def __init__(self):
-        self.add_prim_attr("_process_node_engine_id", 'PS')
-
-    def __call__(self, var_handle, beta1_power, beta2_power, lr, weight_decay, beta1, beta2, epsilon, grad, keys, max_grad_norm, global_step, embedding_dim, ams_grad=(0,), mask_zero=(0,), padding_key=(0,), padding_key_mask=(1,), completion_key=(0,), completion_key_mask=(1,), _embedding_dim=1, _max_key_num=1):
-        return super().__call__(var_handle, beta1_power, beta2_power, lr, weight_decay, beta1, beta2, epsilon, grad, keys, max_grad_norm, global_step, embedding_dim, ams_grad, mask_zero, padding_key, padding_key_mask, completion_key, completion_key_mask, _embedding_dim, _max_key_num)
-
-
-embedding_apply_adam_w_op=EmbeddingApplyAdamW()
-
-
 class SilentCheckV3(Primitive):
     r"""
     
@@ -17739,53 +17349,6 @@ class Triu(Primitive):
         return res
 
 
-class EmbeddingApplyAdam(Primitive):
-    r"""
-    .. code-block::
-
-        prim = ops.EmbeddingApplyAdam()
-        out = prim(var_handle, beta1_power, beta2_power, lr, beta1, beta2, epsilon, grad, keys, global_step, embedding_dim, mask_zero, padding_key, padding_key_mask, completion_key, completion_key_mask, _embedding_dim, _max_key_num)
-
-    is equivalent to
-
-    .. code-block::
-
-        ops.embedding_apply_adam(var_handle, beta1_power, beta2_power, lr, beta1, beta2, epsilon, grad, keys, global_step, embedding_dim, mask_zero, padding_key, padding_key_mask, completion_key, completion_key_mask, _embedding_dim, _max_key_num)
-
-    Refer to :func:`mindspore.ops.embedding_apply_adam` for more details.
-    """
-    __mindspore_signature__ =  (
-        sig.make_sig('var_handle'),
-        sig.make_sig('beta1_power'),
-        sig.make_sig('beta2_power'),
-        sig.make_sig('lr'),
-        sig.make_sig('beta1'),
-        sig.make_sig('beta2'),
-        sig.make_sig('epsilon'),
-        sig.make_sig('grad'),
-        sig.make_sig('keys'),
-        sig.make_sig('global_step'),
-        sig.make_sig('embedding_dim'),
-        sig.make_sig('mask_zero', default=(0,)),
-        sig.make_sig('padding_key', default=(0,)),
-        sig.make_sig('padding_key_mask', default=(1,)),
-        sig.make_sig('completion_key', default=(0,)),
-        sig.make_sig('completion_key_mask', default=(1,)),
-        sig.make_sig('_embedding_dim', default=1),
-        sig.make_sig('_max_key_num', default=1),
-    )
-
-    @prim_arg_register
-    def __init__(self):
-        self.add_prim_attr("_process_node_engine_id", 'PS')
-
-    def __call__(self, var_handle, beta1_power, beta2_power, lr, beta1, beta2, epsilon, grad, keys, global_step, embedding_dim, mask_zero=(0,), padding_key=(0,), padding_key_mask=(1,), completion_key=(0,), completion_key_mask=(1,), _embedding_dim=1, _max_key_num=1):
-        return super().__call__(var_handle, beta1_power, beta2_power, lr, beta1, beta2, epsilon, grad, keys, global_step, embedding_dim, mask_zero, padding_key, padding_key_mask, completion_key, completion_key_mask, _embedding_dim, _max_key_num)
-
-
-embedding_apply_adam_op=EmbeddingApplyAdam()
-
-
 class UpsampleNearest3DGrad(Primitive):
     r"""
     Upsample the 3-D gradient data  with the nearest neighbor interpolation algorithm.
@@ -18601,28 +18164,6 @@ class ReduceMax(Primitive):
         return res
 
 
-class DistCommBarrier(Primitive):
-    r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, group):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_dist_comm_barrier(self, [group])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, group)
-        return res
-
-
-dist_comm_barrier_op=DistCommBarrier()
-
-
 class InnerInplaceIndexPut(Primitive):
     r"""
     Returns a Tensor. According to the index number of `indices` ,
@@ -18700,6 +18241,28 @@ class InnerInplaceIndexPut(Primitive):
 inner_inplace_index_put_op=InnerInplaceIndexPut()
 
 
+class ImagView(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, input):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_imag_view(self, [input])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input)
+        return res
+
+
+imag_view_op=ImagView()
+
+
 class FFTOrtho(Primitive):
     r"""
     
@@ -19078,60 +18641,6 @@ class FFNExt(Primitive):
         return res
 
 
-class DistCommReduceScatter(Primitive):
-    r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, other, input_list, rank_size, op_type, group):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_dist_comm_reduce_scatter(self, [other, input_list, rank_size, op_type, group])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, other, input_list, rank_size, op_type, group)
-        return res
-
-
-dist_comm_reduce_scatter_op=DistCommReduceScatter()
-
-
-class EmbeddingFeatureMappingFind(Primitive):
-    r"""
-    .. code-block::
-
-        prim = ops.EmbeddingFeatureMappingFind()
-        out = prim(table_name, feature_size, num)
-
-    is equivalent to
-
-    .. code-block::
-
-        ops.embedding_feature_mapping_find(table_name, feature_size, num)
-
-    Refer to :func:`mindspore.ops.embedding_feature_mapping_find` for more details.
-    """
-    __mindspore_signature__ =  (
-        sig.make_sig('table_name'),
-        sig.make_sig('feature_size'),
-        sig.make_sig('num', default=1),
-    )
-
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, table_name, feature_size, num=1):
-        return super().__call__(table_name, feature_size, num)
-
-
-embedding_feature_mapping_find_op=EmbeddingFeatureMappingFind()
-
-
 class BitwiseAndScalar(Primitive):
     r"""
     Returns bitwise `and` of tensor and scalar element-wise.
@@ -19527,6 +19036,35 @@ class LstsqV2(Primitive):
 lstsq_v2_op=LstsqV2()
 
 
+class InplaceIndexCopy(Primitive):
+    r"""
+    
+    """
+    __mindspore_signature__ =  (
+        sig.make_sig('input', sig.sig_rw.RW_WRITE),
+        sig.make_sig('dim'),
+        sig.make_sig('index'),
+        sig.make_sig('tensor'),
+    )
+
+    @prim_arg_register
+    def __init__(self):
+        self.add_prim_attr("side_effect_mem", True)
+
+    def __call__(self, input, dim, index, tensor):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_inplace_index_copy(self, [input, dim, index, tensor])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input, dim, index, tensor)
+        return res
+
+
+inplace_index_copy_op=InplaceIndexCopy()
+
+
 class HardtanhGrad(Primitive):
     r"""
     
@@ -20179,28 +19717,6 @@ class TransposeView(Primitive):
 transpose_view_op=TransposeView()
 
 
-class DistCommAllGatherIntoTensorUneven(Primitive):
-    r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, other, input, output_split_sizes, rank_size, group):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_dist_comm_all_gather_into_tensor_uneven(self, [other, input, output_split_sizes, rank_size, group])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, other, input, output_split_sizes, rank_size, group)
-        return res
-
-
-dist_comm_all_gather_into_tensor_uneven_op=DistCommAllGatherIntoTensorUneven()
-
-
 class Erfinv(Primitive):
     r"""
     .. code-block::
@@ -20673,29 +20189,7 @@ class AcosExt(Primitive):
 acos_ext_op=AcosExt()
 
 
-class DistCommScatter(Primitive):
-    r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, other, scatter_list, rank_size, src, rank_id, group):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_dist_comm_scatter(self, [other, scatter_list, rank_size, src, rank_id, group])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, other, scatter_list, rank_size, src, rank_id, group)
-        return res
-
-
-dist_comm_scatter_op=DistCommScatter()
-
-
-class IndexSelect(Primitive):
+class IndexSelect(Primitive):
     r"""
     .. code-block::
 
@@ -20761,6 +20255,34 @@ class SeLUExt(Primitive):
 selu_ext_op=SeLUExt()
 
 
+class InnerUnique(Primitive):
+    r"""
+    
+    """
+    __mindspore_signature__ =  (
+        sig.make_sig('input'),
+        sig.make_sig('sorted', default=True),
+        sig.make_sig('return_inverse', default=False),
+    )
+
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, input, sorted=True, return_inverse=False):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_inner_unique(self, [input, sorted, return_inverse])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input, sorted, return_inverse)
+        return res
+
+
+inner_unique_op=InnerUnique()
+
+
 class Conv3DPadding(Primitive):
     r"""
     
@@ -21498,6 +21020,35 @@ class Acosh(Primitive):
 acosh_op=Acosh()
 
 
+class InplaceIndexFillTensor(Primitive):
+    r"""
+    
+    """
+    __mindspore_signature__ =  (
+        sig.make_sig('input', sig.sig_rw.RW_WRITE),
+        sig.make_sig('dim'),
+        sig.make_sig('index'),
+        sig.make_sig('value'),
+    )
+
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, input, dim, index, value):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_inplace_index_fill_tensor(self, [input, dim, index, value])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input, dim, index, value)
+        return res
+
+
+inplace_index_fill_tensor_op=InplaceIndexFillTensor()
+
+
 class SoftplusExt(Primitive):
     r"""
     .. code-block::
@@ -21784,54 +21335,6 @@ class LpNormV2(Primitive):
 lp_norm_v2_op=LpNormV2()
 
 
-class DistCommBatchIsendIrecv(Primitive):
-    r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, input, group, op_types, remotes_ranks):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_dist_comm_batch_isend_irecv(self, [input, group, op_types, remotes_ranks])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, input, group, op_types, remotes_ranks)
-        return res
-
-
-dist_comm_batch_isend_irecv_op=DistCommBatchIsendIrecv()
-
-
-class EmbeddingFeatureMappingExport(Primitive):
-    r"""
-    .. code-block::
-
-        prim = ops.EmbeddingFeatureMappingExport()
-        out = prim(file_path, table_name, global_step, values, embedding_dim, feature_id, offset_id)
-
-    is equivalent to
-
-    .. code-block::
-
-        ops.embedding_feature_mapping_export(file_path, table_name, global_step, values, embedding_dim, feature_id, offset_id)
-
-    Refer to :func:`mindspore.ops.embedding_feature_mapping_export` for more details.
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, file_path, table_name, global_step, values, embedding_dim, feature_id, offset_id):
-        return super().__call__(file_path, table_name, global_step, values, embedding_dim, feature_id, offset_id)
-
-
-embedding_feature_mapping_export_op=EmbeddingFeatureMappingExport()
-
-
 class Mm(Primitive):
     r"""
     .. code-block::
@@ -21898,51 +21401,6 @@ class Frac(Primitive):
 frac_op=Frac()
 
 
-class EmbeddingApplyFtrl(Primitive):
-    r"""
-    .. code-block::
-
-        prim = ops.EmbeddingApplyFtrl()
-        out = prim(var_handle, lr, lr_power, lambda1, lambda2, grad, keys, global_step, embedding_dim, mask_zero, padding_key, padding_key_mask, completion_key, completion_key_mask, _embedding_dim, _max_key_num)
-
-    is equivalent to
-
-    .. code-block::
-
-        ops.embedding_apply_ftrl(var_handle, lr, lr_power, lambda1, lambda2, grad, keys, global_step, embedding_dim, mask_zero, padding_key, padding_key_mask, completion_key, completion_key_mask, _embedding_dim, _max_key_num)
-
-    Refer to :func:`mindspore.ops.embedding_apply_ftrl` for more details.
-    """
-    __mindspore_signature__ =  (
-        sig.make_sig('var_handle'),
-        sig.make_sig('lr'),
-        sig.make_sig('lr_power'),
-        sig.make_sig('lambda1'),
-        sig.make_sig('lambda2'),
-        sig.make_sig('grad'),
-        sig.make_sig('keys'),
-        sig.make_sig('global_step'),
-        sig.make_sig('embedding_dim'),
-        sig.make_sig('mask_zero', default=(0,)),
-        sig.make_sig('padding_key', default=(0,)),
-        sig.make_sig('padding_key_mask', default=(1,)),
-        sig.make_sig('completion_key', default=(0,)),
-        sig.make_sig('completion_key_mask', default=(1,)),
-        sig.make_sig('_embedding_dim', default=1),
-        sig.make_sig('_max_key_num', default=1),
-    )
-
-    @prim_arg_register
-    def __init__(self):
-        self.add_prim_attr("_process_node_engine_id", 'PS')
-
-    def __call__(self, var_handle, lr, lr_power, lambda1, lambda2, grad, keys, global_step, embedding_dim, mask_zero=(0,), padding_key=(0,), padding_key_mask=(1,), completion_key=(0,), completion_key_mask=(1,), _embedding_dim=1, _max_key_num=1):
-        return super().__call__(var_handle, lr, lr_power, lambda1, lambda2, grad, keys, global_step, embedding_dim, mask_zero, padding_key, padding_key_mask, completion_key, completion_key_mask, _embedding_dim, _max_key_num)
-
-
-embedding_apply_ftrl_op=EmbeddingApplyFtrl()
-
-
 class Cosh(Primitive):
     r"""
     .. code-block::
@@ -22466,42 +21924,6 @@ class InplaceClampScalar(Primitive):
 inplace_clamp_scalar_op=InplaceClampScalar()
 
 
-class EmbeddingFeatureMappingImport(Primitive):
-    r"""
-    .. code-block::
-
-        prim = ops.EmbeddingFeatureMappingImport()
-        out = prim(file_path, teble_name, feature_size, global_step, embedding_dim, only_offset_flag, num)
-
-    is equivalent to
-
-    .. code-block::
-
-        ops.embedding_feature_mapping_import(file_path, teble_name, feature_size, global_step, embedding_dim, only_offset_flag, num)
-
-    Refer to :func:`mindspore.ops.embedding_feature_mapping_import` for more details.
-    """
-    __mindspore_signature__ =  (
-        sig.make_sig('file_path'),
-        sig.make_sig('teble_name'),
-        sig.make_sig('feature_size'),
-        sig.make_sig('global_step'),
-        sig.make_sig('embedding_dim'),
-        sig.make_sig('only_offset_flag', default=True),
-        sig.make_sig('num', default=1),
-    )
-
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, file_path, teble_name, feature_size, global_step, embedding_dim, only_offset_flag=True, num=1):
-        return super().__call__(file_path, teble_name, feature_size, global_step, embedding_dim, only_offset_flag, num)
-
-
-embedding_feature_mapping_import_op=EmbeddingFeatureMappingImport()
-
-
 class ReplicationPad2DGrad(Primitive):
     r"""
     
@@ -23410,28 +22832,6 @@ class FFTWithSize(Primitive):
         return super().__call__(x, self.signal_ndim, self.inverse, self.real, self.norm, self.onesided, self.signal_sizes)
 
 
-class DistCommBroadcast(Primitive):
-    r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, input, src, rank_id, group):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_dist_comm_broadcast(self, [input, src, rank_id, group])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, input, src, rank_id, group)
-        return res
-
-
-dist_comm_broadcast_op=DistCommBroadcast()
-
-
 class InplaceLog(Primitive):
     r"""
     
@@ -23514,28 +22914,6 @@ class Real(Primitive):
 real_op=Real()
 
 
-class InnerCommReduceScatter(Primitive):
-    r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, input, rank_size, op_type, group):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_inner_comm_reduce_scatter(self, [input, rank_size, op_type, group])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, input, rank_size, op_type, group)
-        return res
-
-
-inner_comm_reduce_scatter_op=InnerCommReduceScatter()
-
-
 class Divs(Primitive):
     r"""
     .. code-block::
@@ -24956,6 +24334,35 @@ class DropoutGradExt(Primitive):
 dropout_grad_ext_op=DropoutGradExt()
 
 
+class ToDtype(Primitive):
+    r"""
+    
+    """
+    __mindspore_signature__ =  (
+        sig.make_sig('input'),
+        sig.make_sig('dtype', default=None),
+        sig.make_sig('non_blocking', default=False),
+        sig.make_sig('copy', default=False),
+    )
+
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, input, dtype=None, non_blocking=False, copy=False):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_to_dtype(self, [input, dtype if dtype is None else dtype_to_type_id('ToDtype', 'dtype', dtype), non_blocking, copy])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input, dtype if dtype is None else dtype_to_type_id('ToDtype', 'dtype', dtype), non_blocking, copy)
+        return res
+
+
+to_dtype_op=ToDtype()
+
+
 class TensorCopySlices(Primitive):
     r"""
     Copy continues memory.
@@ -26070,6 +25477,35 @@ class SelectExtView(Primitive):
 select_ext_view_op=SelectExtView()
 
 
+class ToOther(Primitive):
+    r"""
+    
+    """
+    __mindspore_signature__ =  (
+        sig.make_sig('input'),
+        sig.make_sig('other'),
+        sig.make_sig('non_blocking', default=False),
+        sig.make_sig('copy', default=False),
+    )
+
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, input, other, non_blocking=False, copy=False):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_to_other(self, [input, other, non_blocking, copy])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input, other, non_blocking, copy)
+        return res
+
+
+to_other_op=ToOther()
+
+
 class DCTN(Primitive):
     r"""
     
@@ -26271,31 +25707,9 @@ class HFFT2(Primitive):
 hfft2_op=HFFT2()
 
 
-class DistCommAllGather(Primitive):
+class Meshgrid(Primitive):
     r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    def __call__(self, tensor_list, input, rank_size, group):
-        # Add for jit context.
-        if jit_context() and jit_context().compiled:
-            return jit_context().default_output()
-        res = pyboost_dist_comm_all_gather(self, [tensor_list, input, rank_size, group])
-        # Add for jit context.
-        if jit_context():
-            return jit_context().run_op(self, res, tensor_list, input, rank_size, group)
-        return res
-
-
-dist_comm_all_gather_op=DistCommAllGather()
-
-
-class Meshgrid(Primitive):
-    r"""
-    Generates coordinate matrices from given coordinate tensors.
+    Generates coordinate matrices from given coordinate tensors.
 
     Refer to :func:`mindspore.ops.meshgrid` for more details.
 
@@ -27540,13 +26954,13 @@ class GroupedMatmulV4(Primitive):
     .. code-block::
 
         prim = ops.GroupedMatmulV4()
-        out = prim(x, weight, bias, scale, offset, antiquant_scale, antiquant_offset, pre_token_scale, group_list, activation_input, activation_quant_scale, activation_quant_offset, split_item, group_type, group_list_type, act_type)
+        out = prim(x, weight, bias, scale, offset, antiquant_scale, antiquant_offset, pre_token_scale, group_list, activation_input, activation_quant_scale, activation_quant_offset, split_item, group_type, group_list_type, act_type, output_dtype)
 
     is equivalent to
 
     .. code-block::
 
-        ops.grouped_matmul_v4(x, weight, bias, scale, offset, antiquant_scale, antiquant_offset, pre_token_scale, group_list, activation_input, activation_quant_scale, activation_quant_offset, split_item, group_type, group_list_type, act_type)
+        ops.grouped_matmul_v4(x, weight, bias, scale, offset, antiquant_scale, antiquant_offset, pre_token_scale, group_list, activation_input, activation_quant_scale, activation_quant_offset, split_item, group_type, group_list_type, act_type, output_dtype)
 
     Refer to :func:`mindspore.ops.grouped_matmul_v4` for more details.
     """
@@ -27567,20 +26981,21 @@ class GroupedMatmulV4(Primitive):
         sig.make_sig('group_type', default=-1),
         sig.make_sig('group_list_type', default=0),
         sig.make_sig('act_type', default=0),
+        sig.make_sig('output_dtype', default=None),
     )
 
     @prim_arg_register
     def __init__(self):
         pass
 
-    def __call__(self, x, weight, bias=None, scale=None, offset=None, antiquant_scale=None, antiquant_offset=None, pre_token_scale=None, group_list=None, activation_input=None, activation_quant_scale=None, activation_quant_offset=None, split_item=0, group_type=-1, group_list_type=0, act_type=0):
+    def __call__(self, x, weight, bias=None, scale=None, offset=None, antiquant_scale=None, antiquant_offset=None, pre_token_scale=None, group_list=None, activation_input=None, activation_quant_scale=None, activation_quant_offset=None, split_item=0, group_type=-1, group_list_type=0, act_type=0, output_dtype=None):
         # Add for jit context.
         if jit_context() and jit_context().compiled:
             return jit_context().default_output()
-        res = pyboost_grouped_matmul_v4(self, [x, weight, bias, scale, offset, antiquant_scale, antiquant_offset, pre_token_scale, group_list, activation_input, activation_quant_scale, activation_quant_offset, split_item, group_type, group_list_type, act_type])
+        res = pyboost_grouped_matmul_v4(self, [x, weight, bias, scale, offset, antiquant_scale, antiquant_offset, pre_token_scale, group_list, activation_input, activation_quant_scale, activation_quant_offset, split_item, group_type, group_list_type, act_type, output_dtype if output_dtype is None else dtype_to_type_id('GroupedMatmulV4', 'output_dtype', output_dtype)])
         # Add for jit context.
         if jit_context():
-            return jit_context().run_op(self, res, x, weight, bias, scale, offset, antiquant_scale, antiquant_offset, pre_token_scale, group_list, activation_input, activation_quant_scale, activation_quant_offset, split_item, group_type, group_list_type, act_type)
+            return jit_context().run_op(self, res, x, weight, bias, scale, offset, antiquant_scale, antiquant_offset, pre_token_scale, group_list, activation_input, activation_quant_scale, activation_quant_offset, split_item, group_type, group_list_type, act_type, output_dtype if output_dtype is None else dtype_to_type_id('GroupedMatmulV4', 'output_dtype', output_dtype))
         return res
 
 
@@ -28106,97 +27521,73 @@ class TransposeBatchMatmulTranspose(Primitive):
 transpose_batch_matmul_transpose_op=TransposeBatchMatmulTranspose()
 
 
-class GmmV2Backward(Primitive):
+class DistCommIsend(Primitive):
     r"""
     
     """
-    __mindspore_signature__ =  (
-        sig.make_sig('grad'),
-        sig.make_sig('x'),
-        sig.make_sig('weight'),
-        sig.make_sig('group_list', default=None),
-        sig.make_sig('group_list_type', default=0),
-    )
-
     @prim_arg_register
     def __init__(self):
         pass
 
-    def __call__(self, grad, x, weight, group_list=None, group_list_type=0):
+    def __call__(self, input, dst, group, tag):
         # Add for jit context.
         if jit_context() and jit_context().compiled:
             return jit_context().default_output()
-        res = pyboost_gmm_v2_backward(self, [grad, x, weight, group_list, group_list_type])
+        res = pyboost_dist_comm_isend(self, [input, dst, group, tag])
         # Add for jit context.
         if jit_context():
-            return jit_context().run_op(self, res, grad, x, weight, group_list, group_list_type)
+            return jit_context().run_op(self, res, input, dst, group, tag)
         return res
 
 
-gmm_v2_backward_op=GmmV2Backward()
+dist_comm_isend_op=DistCommIsend()
 
 
-class GmmV2(Primitive):
+class DistCommIrecv(Primitive):
     r"""
     
     """
-    __mindspore_signature__ =  (
-        sig.make_sig('x'),
-        sig.make_sig('weight'),
-        sig.make_sig('bias', default=None),
-        sig.make_sig('group_list', default=None),
-        sig.make_sig('group_type', default=0),
-        sig.make_sig('group_list_type', default=0),
-    )
-
     @prim_arg_register
     def __init__(self):
         pass
 
-    def __call__(self, x, weight, bias=None, group_list=None, group_type=0, group_list_type=0):
+    def __call__(self, input, tag, src, group):
         # Add for jit context.
         if jit_context() and jit_context().compiled:
             return jit_context().default_output()
-        res = pyboost_gmm_v2(self, [x, weight, bias, group_list, group_type, group_list_type])
+        res = pyboost_dist_comm_irecv(self, [input, tag, src, group])
         # Add for jit context.
         if jit_context():
-            return jit_context().run_op(self, res, x, weight, bias, group_list, group_type, group_list_type)
+            return jit_context().run_op(self, res, input, tag, src, group)
         return res
 
 
-gmm_v2_op=GmmV2()
+dist_comm_irecv_op=DistCommIrecv()
 
 
-class GmmV2BackwardFusion(Primitive):
+class DistCommReduceScatterTensor(Primitive):
     r"""
     
     """
-    __mindspore_signature__ =  (
-        sig.make_sig('grad'),
-        sig.make_sig('weight'),
-        sig.make_sig('group_list', default=None),
-        sig.make_sig('group_list_type', default=0),
-    )
-
     @prim_arg_register
     def __init__(self):
         pass
 
-    def __call__(self, grad, weight, group_list=None, group_list_type=0):
+    def __call__(self, other, input, rank_size, op_type, group):
         # Add for jit context.
         if jit_context() and jit_context().compiled:
             return jit_context().default_output()
-        res = pyboost_gmm_v2_backward_fusion(self, [grad, weight, group_list, group_list_type])
+        res = pyboost_dist_comm_reduce_scatter_tensor(self, [other, input, rank_size, op_type, group])
         # Add for jit context.
         if jit_context():
-            return jit_context().run_op(self, res, grad, weight, group_list, group_list_type)
+            return jit_context().run_op(self, res, other, input, rank_size, op_type, group)
         return res
 
 
-gmm_v2_backward_fusion_op=GmmV2BackwardFusion()
+dist_comm_reduce_scatter_tensor_op=DistCommReduceScatterTensor()
 
 
-class PixelShuffle(Primitive):
+class DistCommGatherIntoTensor(Primitive):
     r"""
     
     """
@@ -28204,50 +27595,43 @@ class PixelShuffle(Primitive):
     def __init__(self):
         pass
 
-    def __call__(self, input, upscale_factor):
+    def __call__(self, other, input, rank_size, dst, rank_id, group):
         # Add for jit context.
         if jit_context() and jit_context().compiled:
             return jit_context().default_output()
-        res = pyboost_pixel_shuffle(self, [input, upscale_factor])
+        res = pyboost_dist_comm_gather_into_tensor(self, [other, input, rank_size, dst, rank_id, group])
         # Add for jit context.
         if jit_context():
-            return jit_context().run_op(self, res, input, upscale_factor)
+            return jit_context().run_op(self, res, other, input, rank_size, dst, rank_id, group)
         return res
 
 
-pixel_shuffle_op=PixelShuffle()
+dist_comm_gather_into_tensor_op=DistCommGatherIntoTensor()
 
 
-class GmmBackwardFusion(Primitive):
+class DistCommReduceScatterTensorUneven(Primitive):
     r"""
     
     """
-    __mindspore_signature__ =  (
-        sig.make_sig('grad'),
-        sig.make_sig('weight'),
-        sig.make_sig('group_list', default=None),
-        sig.make_sig('group_list_type', default=0),
-    )
-
     @prim_arg_register
     def __init__(self):
         pass
 
-    def __call__(self, grad, weight, group_list=None, group_list_type=0):
+    def __call__(self, other, input, input_split_size, rank_size, op_type, group):
         # Add for jit context.
         if jit_context() and jit_context().compiled:
             return jit_context().default_output()
-        res = pyboost_gmm_backward_fusion(self, [grad, weight, group_list, group_list_type])
+        res = pyboost_dist_comm_reduce_scatter_tensor_uneven(self, [other, input, input_split_size, rank_size, op_type, group])
         # Add for jit context.
         if jit_context():
-            return jit_context().run_op(self, res, grad, weight, group_list, group_list_type)
+            return jit_context().run_op(self, res, other, input, input_split_size, rank_size, op_type, group)
         return res
 
 
-gmm_backward_fusion_op=GmmBackwardFusion()
+dist_comm_reduce_scatter_tensor_uneven_op=DistCommReduceScatterTensorUneven()
 
 
-class EinsumExt(Primitive):
+class InnerCommAllReduce(Primitive):
     r"""
     
     """
@@ -28255,123 +27639,144 @@ class EinsumExt(Primitive):
     def __init__(self):
         pass
 
-    def __call__(self, equation, operands):
+    def __call__(self, input, op_type, group):
         # Add for jit context.
         if jit_context() and jit_context().compiled:
             return jit_context().default_output()
-        res = pyboost_einsum_ext(self, [equation, operands])
+        res = pyboost_inner_comm_all_reduce(self, [input, op_type, group])
         # Add for jit context.
         if jit_context():
-            return jit_context().run_op(self, res, equation, operands)
+            return jit_context().run_op(self, res, input, op_type, group)
         return res
 
 
-einsum_ext_op=EinsumExt()
+inner_comm_all_reduce_op=InnerCommAllReduce()
 
 
-class InplaceExponential(Primitive):
+class DistCommScatterTensor(Primitive):
     r"""
     
     """
-    __mindspore_signature__ =  (
-        sig.make_sig('input', sig.sig_rw.RW_WRITE),
-        sig.make_sig('lambd'),
-        sig.make_sig('seed'),
-        sig.make_sig('offset'),
-    )
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, other, input, rank_size, src, rank_id, group):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_dist_comm_scatter_tensor(self, [other, input, rank_size, src, rank_id, group])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, other, input, rank_size, src, rank_id, group)
+        return res
+
 
+dist_comm_scatter_tensor_op=DistCommScatterTensor()
+
+
+class InnerCommAllGather(Primitive):
+    r"""
+    
+    """
     @prim_arg_register
     def __init__(self):
-        self.add_prim_attr("side_effect_mem", True)
+        pass
 
-    def __call__(self, input, lambd, seed, offset):
+    def __call__(self, input, rank_size, group):
         # Add for jit context.
         if jit_context() and jit_context().compiled:
             return jit_context().default_output()
-        res = pyboost_inplace_exponential(self, [input, lambd, seed, offset])
+        res = pyboost_inner_comm_all_gather(self, [input, rank_size, group])
         # Add for jit context.
         if jit_context():
-            return jit_context().run_op(self, res, input, lambd, seed, offset)
+            return jit_context().run_op(self, res, input, rank_size, group)
         return res
 
 
-inplace_exponential_op=InplaceExponential()
+inner_comm_all_gather_op=InnerCommAllGather()
 
 
-class FuncMaxPool2D(Primitive):
+class DistCommAllReduce(Primitive):
     r"""
     
     """
     __mindspore_signature__ =  (
-        sig.make_sig('input'),
-        sig.make_sig('kernel_size'),
-        sig.make_sig('stride', default=None),
-        sig.make_sig('padding', default=0),
-        sig.make_sig('dilation', default=(1, 1)),
-        sig.make_sig('ceil_mode', default=False),
-        sig.make_sig('return_indices', default=False),
+        sig.make_sig('input', sig.sig_rw.RW_WRITE),
+        sig.make_sig('op_type'),
+        sig.make_sig('group'),
     )
 
     @prim_arg_register
     def __init__(self):
-        pass
+        self.add_prim_attr("side_effect_mem", True)
 
-    def __call__(self, input, kernel_size, stride=None, padding=0, dilation=(1, 1), ceil_mode=False, return_indices=False):
+    def __call__(self, input, op_type, group):
         # Add for jit context.
         if jit_context() and jit_context().compiled:
             return jit_context().default_output()
-        res = pyboost_func_max_pool2d(self, [input, to_kernel_size('FuncMaxPool2D', 'kernel_size', kernel_size), stride if stride is None else to_strides('FuncMaxPool2D', 'stride', stride), to_output_padding('FuncMaxPool2D', 'padding', padding), to_dilations('FuncMaxPool2D', 'dilation', dilation), ceil_mode, return_indices])
+        res = pyboost_dist_comm_all_reduce(self, [input, op_type, group])
         # Add for jit context.
         if jit_context():
-            return jit_context().run_op(self, res, input, to_kernel_size('FuncMaxPool2D', 'kernel_size', kernel_size), stride if stride is None else to_strides('FuncMaxPool2D', 'stride', stride), to_output_padding('FuncMaxPool2D', 'padding', padding), to_dilations('FuncMaxPool2D', 'dilation', dilation), ceil_mode, return_indices)
+            return jit_context().run_op(self, res, input, op_type, group)
         return res
 
 
-func_max_pool2d_op=FuncMaxPool2D()
+dist_comm_all_reduce_op=DistCommAllReduce()
 
 
-class MoeTokenUnpermute(Primitive):
+class InnerCommIsend(Primitive):
     r"""
-    .. code-block::
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
 
-        prim = ops.MoeTokenUnpermute()
-        out = prim(permuted_tokens, sorted_indices, probs, padded_mode, restore_shape)
+    def __call__(self, input, dst, group, tag):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_inner_comm_isend(self, [input, dst, group, tag])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input, dst, group, tag)
+        return res
 
-    is equivalent to
 
-    .. code-block::
+inner_comm_isend_op=InnerCommIsend()
 
-        ops.moe_token_unpermute(permuted_tokens, sorted_indices, probs, padded_mode, restore_shape)
 
-    Refer to :func:`mindspore.ops.moe_token_unpermute` for more details.
+class DistCommReduce(Primitive):
+    r"""
+    
     """
     __mindspore_signature__ =  (
-        sig.make_sig('permuted_tokens'),
-        sig.make_sig('sorted_indices'),
-        sig.make_sig('probs', default=None),
-        sig.make_sig('padded_mode', default=False),
-        sig.make_sig('restore_shape', default=None),
+        sig.make_sig('input', sig.sig_rw.RW_WRITE),
+        sig.make_sig('op_type'),
+        sig.make_sig('dst'),
+        sig.make_sig('group'),
     )
 
     @prim_arg_register
     def __init__(self):
-        pass
+        self.add_prim_attr("side_effect_mem", True)
 
-    def __call__(self, permuted_tokens, sorted_indices, probs=None, padded_mode=False, restore_shape=None):
+    def __call__(self, input, op_type, dst, group):
         # Add for jit context.
         if jit_context() and jit_context().compiled:
             return jit_context().default_output()
-        res = pyboost_moe_token_unpermute(self, [permuted_tokens, sorted_indices, probs, padded_mode, restore_shape])
+        res = pyboost_dist_comm_reduce(self, [input, op_type, dst, group])
         # Add for jit context.
         if jit_context():
-            return jit_context().run_op(self, res, permuted_tokens, sorted_indices, probs, padded_mode, restore_shape)
+            return jit_context().run_op(self, res, input, op_type, dst, group)
         return res
 
 
-moe_token_unpermute_op=MoeTokenUnpermute()
+dist_comm_reduce_op=DistCommReduce()
 
 
-class Any(Primitive):
+class DistCommGather(Primitive):
     r"""
     
     """
@@ -28379,49 +27784,666 @@ class Any(Primitive):
     def __init__(self):
         pass
 
-    def __call__(self, input):
+    def __call__(self, input, gather_list, rank_size, dst, rank_id, group):
         # Add for jit context.
         if jit_context() and jit_context().compiled:
             return jit_context().default_output()
-        res = pyboost_any(self, [input])
+        res = pyboost_dist_comm_gather(self, [input, gather_list, rank_size, dst, rank_id, group])
         # Add for jit context.
         if jit_context():
-            return jit_context().run_op(self, res, input)
+            return jit_context().run_op(self, res, input, gather_list, rank_size, dst, rank_id, group)
         return res
 
 
-any_op=Any()
+dist_comm_gather_op=DistCommGather()
 
 
-class Gmm(Primitive):
+class DistCommAllToAllV(Primitive):
     r"""
     
     """
-    __mindspore_signature__ =  (
-        sig.make_sig('x'),
-        sig.make_sig('weight'),
-        sig.make_sig('bias', default=None),
-        sig.make_sig('group_list', default=None),
-        sig.make_sig('group_type', default=0),
-        sig.make_sig('group_list_type', default=0),
-    )
-
     @prim_arg_register
     def __init__(self):
         pass
 
-    def __call__(self, x, weight, bias=None, group_list=None, group_type=0, group_list_type=0):
+    def __call__(self, other, input, group, send_numel_list, recv_numel_list, rank_size):
         # Add for jit context.
         if jit_context() and jit_context().compiled:
             return jit_context().default_output()
-        res = pyboost_gmm(self, [x, weight, bias, group_list, group_type, group_list_type])
+        res = pyboost_dist_comm_all_to_all_v(self, [other, input, group, send_numel_list, recv_numel_list, rank_size])
         # Add for jit context.
         if jit_context():
-            return jit_context().run_op(self, res, x, weight, bias, group_list, group_type, group_list_type)
+            return jit_context().run_op(self, res, other, input, group, send_numel_list, recv_numel_list, rank_size)
         return res
 
 
-gmm_op=Gmm()
+dist_comm_all_to_all_v_op=DistCommAllToAllV()
+
+
+class InnerCommAllToAllV(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, input, group, send_numel_list, recv_numel_list, rank_size, split_sizes_empty):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_inner_comm_all_to_all_v(self, [input, group, send_numel_list, recv_numel_list, rank_size, split_sizes_empty])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input, group, send_numel_list, recv_numel_list, rank_size, split_sizes_empty)
+        return res
+
+
+inner_comm_all_to_all_v_op=InnerCommAllToAllV()
+
+
+class InnerCommIrecv(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, tag, src, shape, group, dtype):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_inner_comm_irecv(self, [tag, src, shape, group, dtype_to_type_id('InnerCommIrecv', 'dtype', dtype)])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, tag, src, shape, group, dtype_to_type_id('InnerCommIrecv', 'dtype', dtype))
+        return res
+
+
+inner_comm_irecv_op=InnerCommIrecv()
+
+
+class DistCommAllToAllVSingle(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, other, input, group, send_numel_list, recv_numel_list, rank_size, split_sizes_empty):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_dist_comm_all_to_all_v_single(self, [other, input, group, send_numel_list, recv_numel_list, rank_size, split_sizes_empty])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, other, input, group, send_numel_list, recv_numel_list, rank_size, split_sizes_empty)
+        return res
+
+
+dist_comm_all_to_all_v_single_op=DistCommAllToAllVSingle()
+
+
+class DistCommAllGatherIntoTensor(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, other, input, rank_size, group):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_dist_comm_all_gather_into_tensor(self, [other, input, rank_size, group])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, other, input, rank_size, group)
+        return res
+
+
+dist_comm_all_gather_into_tensor_op=DistCommAllGatherIntoTensor()
+
+
+class DistCommBarrier(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, group):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_dist_comm_barrier(self, [group])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, group)
+        return res
+
+
+dist_comm_barrier_op=DistCommBarrier()
+
+
+class DistCommReduceScatter(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, other, input_list, rank_size, op_type, group):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_dist_comm_reduce_scatter(self, [other, input_list, rank_size, op_type, group])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, other, input_list, rank_size, op_type, group)
+        return res
+
+
+dist_comm_reduce_scatter_op=DistCommReduceScatter()
+
+
+class DistCommAllGatherIntoTensorUneven(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, other, input, output_split_sizes, rank_size, group):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_dist_comm_all_gather_into_tensor_uneven(self, [other, input, output_split_sizes, rank_size, group])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, other, input, output_split_sizes, rank_size, group)
+        return res
+
+
+dist_comm_all_gather_into_tensor_uneven_op=DistCommAllGatherIntoTensorUneven()
+
+
+class DistCommScatter(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, other, scatter_list, rank_size, src, rank_id, group):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_dist_comm_scatter(self, [other, scatter_list, rank_size, src, rank_id, group])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, other, scatter_list, rank_size, src, rank_id, group)
+        return res
+
+
+dist_comm_scatter_op=DistCommScatter()
+
+
+class DistCommBatchIsendIrecv(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, input, group, op_types, remotes_ranks):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_dist_comm_batch_isend_irecv(self, [input, group, op_types, remotes_ranks])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input, group, op_types, remotes_ranks)
+        return res
+
+
+dist_comm_batch_isend_irecv_op=DistCommBatchIsendIrecv()
+
+
+class DistCommAllToAllVC(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, other, input, group, send_count_matrix, rank_size, rank_id):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_dist_comm_all_to_all_v_c(self, [other, input, group, send_count_matrix, rank_size, rank_id])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, other, input, group, send_count_matrix, rank_size, rank_id)
+        return res
+
+
+dist_comm_all_to_all_v_c_op=DistCommAllToAllVC()
+
+
+class DistCommBroadcast(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, input, src, rank_id, group):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_dist_comm_broadcast(self, [input, src, rank_id, group])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input, src, rank_id, group)
+        return res
+
+
+dist_comm_broadcast_op=DistCommBroadcast()
+
+
+class InnerCommReduceScatter(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, input, rank_size, op_type, group):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_inner_comm_reduce_scatter(self, [input, rank_size, op_type, group])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input, rank_size, op_type, group)
+        return res
+
+
+inner_comm_reduce_scatter_op=InnerCommReduceScatter()
+
+
+class DistCommAllGather(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, tensor_list, input, rank_size, group):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_dist_comm_all_gather(self, [tensor_list, input, rank_size, group])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, tensor_list, input, rank_size, group)
+        return res
+
+
+dist_comm_all_gather_op=DistCommAllGather()
+
+
+class GmmV2Backward(Primitive):
+    r"""
+    
+    """
+    __mindspore_signature__ =  (
+        sig.make_sig('grad'),
+        sig.make_sig('x'),
+        sig.make_sig('weight'),
+        sig.make_sig('group_list', default=None),
+        sig.make_sig('group_list_type', default=0),
+    )
+
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, grad, x, weight, group_list=None, group_list_type=0):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_gmm_v2_backward(self, [grad, x, weight, group_list, group_list_type])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, grad, x, weight, group_list, group_list_type)
+        return res
+
+
+gmm_v2_backward_op=GmmV2Backward()
+
+
+class GmmV2(Primitive):
+    r"""
+    
+    """
+    __mindspore_signature__ =  (
+        sig.make_sig('x'),
+        sig.make_sig('weight'),
+        sig.make_sig('bias', default=None),
+        sig.make_sig('group_list', default=None),
+        sig.make_sig('group_type', default=0),
+        sig.make_sig('group_list_type', default=0),
+    )
+
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, x, weight, bias=None, group_list=None, group_type=0, group_list_type=0):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_gmm_v2(self, [x, weight, bias, group_list, group_type, group_list_type])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, x, weight, bias, group_list, group_type, group_list_type)
+        return res
+
+
+gmm_v2_op=GmmV2()
+
+
+class GmmV2BackwardFusion(Primitive):
+    r"""
+    
+    """
+    __mindspore_signature__ =  (
+        sig.make_sig('grad'),
+        sig.make_sig('weight'),
+        sig.make_sig('group_list', default=None),
+        sig.make_sig('group_list_type', default=0),
+    )
+
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, grad, weight, group_list=None, group_list_type=0):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_gmm_v2_backward_fusion(self, [grad, weight, group_list, group_list_type])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, grad, weight, group_list, group_list_type)
+        return res
+
+
+gmm_v2_backward_fusion_op=GmmV2BackwardFusion()
+
+
+class PixelShuffle(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, input, upscale_factor):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_pixel_shuffle(self, [input, upscale_factor])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input, upscale_factor)
+        return res
+
+
+pixel_shuffle_op=PixelShuffle()
+
+
+class GmmBackwardFusion(Primitive):
+    r"""
+    
+    """
+    __mindspore_signature__ =  (
+        sig.make_sig('grad'),
+        sig.make_sig('weight'),
+        sig.make_sig('group_list', default=None),
+        sig.make_sig('group_list_type', default=0),
+    )
+
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, grad, weight, group_list=None, group_list_type=0):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_gmm_backward_fusion(self, [grad, weight, group_list, group_list_type])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, grad, weight, group_list, group_list_type)
+        return res
+
+
+gmm_backward_fusion_op=GmmBackwardFusion()
+
+
+class EinsumExt(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, equation, operands):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_einsum_ext(self, [equation, operands])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, equation, operands)
+        return res
+
+
+einsum_ext_op=EinsumExt()
+
+
+class InplaceExponential(Primitive):
+    r"""
+    
+    """
+    __mindspore_signature__ =  (
+        sig.make_sig('input', sig.sig_rw.RW_WRITE),
+        sig.make_sig('lambd'),
+        sig.make_sig('seed'),
+        sig.make_sig('offset'),
+    )
+
+    @prim_arg_register
+    def __init__(self):
+        self.add_prim_attr("side_effect_mem", True)
+
+    def __call__(self, input, lambd, seed, offset):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_inplace_exponential(self, [input, lambd, seed, offset])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input, lambd, seed, offset)
+        return res
+
+
+inplace_exponential_op=InplaceExponential()
+
+
+class FuncMaxPool2D(Primitive):
+    r"""
+    
+    """
+    __mindspore_signature__ =  (
+        sig.make_sig('input'),
+        sig.make_sig('kernel_size'),
+        sig.make_sig('stride', default=None),
+        sig.make_sig('padding', default=0),
+        sig.make_sig('dilation', default=(1, 1)),
+        sig.make_sig('ceil_mode', default=False),
+        sig.make_sig('return_indices', default=False),
+    )
+
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, input, kernel_size, stride=None, padding=0, dilation=(1, 1), ceil_mode=False, return_indices=False):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_func_max_pool2d(self, [input, to_kernel_size('FuncMaxPool2D', 'kernel_size', kernel_size), stride if stride is None else to_strides('FuncMaxPool2D', 'stride', stride), to_output_padding('FuncMaxPool2D', 'padding', padding), to_dilations('FuncMaxPool2D', 'dilation', dilation), ceil_mode, return_indices])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input, to_kernel_size('FuncMaxPool2D', 'kernel_size', kernel_size), stride if stride is None else to_strides('FuncMaxPool2D', 'stride', stride), to_output_padding('FuncMaxPool2D', 'padding', padding), to_dilations('FuncMaxPool2D', 'dilation', dilation), ceil_mode, return_indices)
+        return res
+
+
+func_max_pool2d_op=FuncMaxPool2D()
+
+
+class MoeTokenUnpermute(Primitive):
+    r"""
+    .. code-block::
+
+        prim = ops.MoeTokenUnpermute()
+        out = prim(permuted_tokens, sorted_indices, probs, padded_mode, restore_shape)
+
+    is equivalent to
+
+    .. code-block::
+
+        ops.moe_token_unpermute(permuted_tokens, sorted_indices, probs, padded_mode, restore_shape)
+
+    Refer to :func:`mindspore.ops.moe_token_unpermute` for more details.
+    """
+    __mindspore_signature__ =  (
+        sig.make_sig('permuted_tokens'),
+        sig.make_sig('sorted_indices'),
+        sig.make_sig('probs', default=None),
+        sig.make_sig('padded_mode', default=False),
+        sig.make_sig('restore_shape', default=None),
+    )
+
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, permuted_tokens, sorted_indices, probs=None, padded_mode=False, restore_shape=None):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_moe_token_unpermute(self, [permuted_tokens, sorted_indices, probs, padded_mode, restore_shape])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, permuted_tokens, sorted_indices, probs, padded_mode, restore_shape)
+        return res
+
+
+moe_token_unpermute_op=MoeTokenUnpermute()
+
+
+class Any(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, input):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_any(self, [input])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input)
+        return res
+
+
+any_op=Any()
+
+
+class Gmm(Primitive):
+    r"""
+    
+    """
+    __mindspore_signature__ =  (
+        sig.make_sig('x'),
+        sig.make_sig('weight'),
+        sig.make_sig('bias', default=None),
+        sig.make_sig('group_list', default=None),
+        sig.make_sig('group_type', default=0),
+        sig.make_sig('group_list_type', default=0),
+    )
+
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, x, weight, bias=None, group_list=None, group_type=0, group_list_type=0):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_gmm(self, [x, weight, bias, group_list, group_type, group_list_type])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, x, weight, bias, group_list, group_type, group_list_type)
+        return res
+
+
+gmm_op=Gmm()
+
+
+class Dropout2dExt(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    def __call__(self, input, p, training, inplace, seed, offset):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_dropout2d_ext(self, [input, p, training, inplace, seed, offset])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input, p, training, inplace, seed, offset)
+        return res
+
+
+dropout2d_ext_op=Dropout2dExt()
 
 
 class FuncDropoutExt(Primitive):