mindspore 2.7.0__cp39-cp39-manylinux1_x86_64.whl → 2.7.1__cp39-cp39-manylinux1_x86_64.whl

mindspore/ops/auto_generate/gen_ops_prim.py

@@ -27,7 +27,6 @@ from mindspore.common.jit_context import jit_context
 from mindspore._checkparam import is_stub_tensor
 from mindspore._c_expression import pyboost_inplace_elu
 from mindspore._c_expression import pyboost_cummin_ext
-from mindspore._c_expression import pyboost_inner_comm_isend
 from mindspore._c_expression import pyboost_multi_scale_deformable_attn
 from mindspore._c_expression import pyboost_reduce_any
 from mindspore._c_expression import pyboost_inplace_floor_divide
@@ -41,11 +40,11 @@ from mindspore._c_expression import pyboost_elu_ext
 from mindspore._c_expression import pyboost_zeros
 from mindspore._c_expression import pyboost_expm1
 from mindspore._c_expression import pyboost_group_norm
-from mindspore._c_expression import pyboost_dist_comm_broadcast
 from mindspore._c_expression import pyboost_copy
 from mindspore._c_expression import pyboost_upsample_nearest3d
 from mindspore._c_expression import pyboost_inplace_adds_ext
 from mindspore._c_expression import pyboost_non_zero
+from mindspore._c_expression import pyboost_inplace_index_fill_tensor
 from mindspore._c_expression import pyboost_inplace_div
 from mindspore._c_expression import pyboost_expand_as
 from mindspore._c_expression import pyboost_batch_norm_reduce_grad
@@ -53,7 +52,6 @@ from mindspore._c_expression import pyboost_hardtanh
 from mindspore._c_expression import pyboost_prelu
 from mindspore._c_expression import pyboost_count_nonzero
 from mindspore._c_expression import pyboost_inplace_mul
-from mindspore._c_expression import pyboost_inner_comm_irecv
 from mindspore._c_expression import pyboost_mla
 from mindspore._c_expression import pyboost_remainder_scalar_tensor
 from mindspore._c_expression import pyboost_repeat
@@ -90,13 +88,13 @@ from mindspore._c_expression import pyboost_reshape_and_cache
 from mindspore._c_expression import pyboost_randn
 from mindspore._c_expression import pyboost_upsample_bilinear2d
 from mindspore._c_expression import pyboost_xlogy
-from mindspore._c_expression import pyboost_inner_comm_all_gather
 from mindspore._c_expression import pyboost_max
 from mindspore._c_expression import pyboost_matmul_reduce_scatter
 from mindspore._c_expression import pyboost_adaptive_avg_pool2d_ext
 from mindspore._c_expression import pyboost_log_softmax_ext
 from mindspore._c_expression import pyboost_inplace_grouped_matmul_add
 from mindspore._c_expression import pyboost_dot
+from mindspore._c_expression import pyboost_inplace_sign
 from mindspore._c_expression import pyboost_sort_ext
 from mindspore._c_expression import pyboost_randint
 from mindspore._c_expression import pyboost_reflection_pad_2d
@@ -125,13 +123,13 @@ from mindspore._c_expression import pyboost_atan2_ext
 from mindspore._c_expression import pyboost_inplace_fill_tensor
 from mindspore._c_expression import pyboost_scatter_value
 from mindspore._c_expression import pyboost_split_with_size_view
-from mindspore._c_expression import pyboost_dist_comm_barrier
 from mindspore._c_expression import pyboost_inplace_index_put
 from mindspore._c_expression import pyboost_view_as
+from mindspore._c_expression import pyboost_dequant_swiglu_quant
 from mindspore._c_expression import pyboost_hardtanh_grad
-from mindspore._c_expression import pyboost_dist_comm_gather
 from mindspore._c_expression import pyboost_lerp
 from mindspore._c_expression import pyboost_triu
+from mindspore._c_expression import pyboost_to_device
 from mindspore._c_expression import pyboost_logaddexp
 from mindspore._c_expression import pyboost_upsample_nearest3d_grad
 from mindspore._c_expression import pyboost_flash_attention_score
@@ -142,7 +140,6 @@ from mindspore._c_expression import pyboost_conv1d_ext
 from mindspore._c_expression import pyboost_tril_ext
 from mindspore._c_expression import pyboost_batch_norm_stats
 from mindspore._c_expression import pyboost_bitwise_and_tensor
-from mindspore._c_expression import pyboost_dist_comm_reduce_scatter_tensor_uneven
 from mindspore._c_expression import pyboost_replication_pad_2d
 from mindspore._c_expression import pyboost_max_pool_with_indices
 from mindspore._c_expression import pyboost_sub
@@ -151,7 +148,6 @@ from mindspore._c_expression import pyboost_fill_scalar
 from mindspore._c_expression import pyboost_addcmul_ext
 from mindspore._c_expression import pyboost_clone
 from mindspore._c_expression import pyboost_atan_ext
-from mindspore._c_expression import pyboost_inner_comm_all_to_all_v
 from mindspore._c_expression import pyboost_adaptive_max_pool2d
 from mindspore._c_expression import pyboost_addmm
 from mindspore._c_expression import pyboost_transpose_view
@@ -166,7 +162,6 @@ from mindspore._c_expression import pyboost_inplace_index_add
 from mindspore._c_expression import pyboost_split_tensor
 from mindspore._c_expression import pyboost_max_pool_grad_with_indices
 from mindspore._c_expression import pyboost_embedding
-from mindspore._c_expression import pyboost_dist_comm_reduce_scatter
 from mindspore._c_expression import pyboost_new_empty
 from mindspore._c_expression import pyboost_select_ext_view
 from mindspore._c_expression import pyboost_incre_flash_attention
@@ -195,6 +190,7 @@ from mindspore._c_expression import pyboost_randint_like
 from mindspore._c_expression import pyboost_gcd
 from mindspore._c_expression import pyboost_split
 from mindspore._c_expression import pyboost_mul
+from mindspore._c_expression import pyboost_to_other
 from mindspore._c_expression import pyboost_elu_grad_ext
 from mindspore._c_expression import pyboost_randperm_ext
 from mindspore._c_expression import pyboost_diagonal_view
@@ -220,7 +216,6 @@ from mindspore._c_expression import pyboost_empty_like
 from mindspore._c_expression import pyboost_silu
 from mindspore._c_expression import pyboost_nansum
 from mindspore._c_expression import pyboost_split_with_size
-from mindspore._c_expression import pyboost_dist_comm_all_gather_into_tensor
 from mindspore._c_expression import pyboost_l1_loss_backward_ext
 from mindspore._c_expression import pyboost_add_rms_norm
 from mindspore._c_expression import pyboost_inplace_silu
@@ -228,6 +223,7 @@ from mindspore._c_expression import pyboost_relu
 from mindspore._c_expression import pyboost_std_mean
 from mindspore._c_expression import pyboost_avg_pool2d_grad
 from mindspore._c_expression import pyboost_soft_margin_loss
+from mindspore._c_expression import pyboost_inplace_index_fill_scalar
 from mindspore._c_expression import pyboost_less_equal
 from mindspore._c_expression import pyboost_conv_transpose2d
 from mindspore._c_expression import pyboost_mean_ext
@@ -239,6 +235,7 @@ from mindspore._c_expression import pyboost_gather_d_grad_v2
 from mindspore._c_expression import pyboost_equal
 from mindspore._c_expression import pyboost_moe_distribute_dispatch
 from mindspore._c_expression import pyboost_inplace_floor_divides
+from mindspore._c_expression import pyboost_imag_view
 from mindspore._c_expression import pyboost_cumsum_ext
 from mindspore._c_expression import pyboost_layer_norm_grad_ext
 from mindspore._c_expression import pyboost_logsumexp
@@ -249,7 +246,6 @@ from mindspore._c_expression import pyboost_inplace_fill_scalar
 from mindspore._c_expression import pyboost_add_scalar
 from mindspore._c_expression import pyboost_softplus_ext
 from mindspore._c_expression import pyboost_hswish_grad
-from mindspore._c_expression import pyboost_dist_comm_gather_into_tensor
 from mindspore._c_expression import pyboost_diag_ext
 from mindspore._c_expression import pyboost_logaddexp2
 from mindspore._c_expression import pyboost_normal_float_float
@@ -280,9 +276,7 @@ from mindspore._c_expression import pyboost_max_pool_grad_with_mask
 from mindspore._c_expression import pyboost_l1_loss_ext
 from mindspore._c_expression import pyboost_gelu_ext
 from mindspore._c_expression import pyboost_squeeze
-from mindspore._c_expression import pyboost_dist_comm_irecv
 from mindspore._c_expression import pyboost_lerp_scalar
-from mindspore._c_expression import pyboost_inner_comm_all_reduce
 from mindspore._c_expression import pyboost_square
 from mindspore._c_expression import pyboost_conv2d_padding
 from mindspore._c_expression import pyboost_argmin_with_value
@@ -296,7 +290,6 @@ from mindspore._c_expression import pyboost_broadcast_to_view
 from mindspore._c_expression import pyboost_softshrink
 from mindspore._c_expression import pyboost_add_layer_norm_grad
 from mindspore._c_expression import pyboost_unstack_ext_view
-from mindspore._c_expression import pyboost_dist_comm_batch_isend_irecv
 from mindspore._c_expression import pyboost_grid_sampler_2d_grad
 from mindspore._c_expression import pyboost_rotary_position_embedding
 from mindspore._c_expression import pyboost_var_mean
@@ -307,6 +300,7 @@ from mindspore._c_expression import pyboost_inplace_divmods
 from mindspore._c_expression import pyboost_fill_tensor
 from mindspore._c_expression import pyboost_expand_dims
 from mindspore._c_expression import pyboost_inplace_threshold
+from mindspore._c_expression import pyboost_inplace_index_copy
 from mindspore._c_expression import pyboost_inplace_remainder_tensor_tensor
 from mindspore._c_expression import pyboost_erfinv
 from mindspore._c_expression import pyboost_triangular_solve
@@ -329,7 +323,6 @@ from mindspore._c_expression import pyboost_zeros_like_ext
 from mindspore._c_expression import pyboost_concat
 from mindspore._c_expression import pyboost_bitwise_xor_scalar
 from mindspore._c_expression import pyboost_fmod_tensor
-from mindspore._c_expression import pyboost_dist_comm_reduce_scatter_tensor
 from mindspore._c_expression import pyboost_cross_entropy_loss_grad
 from mindspore._c_expression import pyboost_upsample_nearest1d
 from mindspore._c_expression import pyboost_isfinite
@@ -339,19 +332,18 @@ from mindspore._c_expression import pyboost_tanh
 from mindspore._c_expression import pyboost_new_full
 from mindspore._c_expression import pyboost_muls
 from mindspore._c_expression import pyboost_dropout_ext
-from mindspore._c_expression import pyboost_dist_comm_scatter
 from mindspore._c_expression import pyboost_asin_ext
 from mindspore._c_expression import pyboost_remainder_tensor_tensor
 from mindspore._c_expression import pyboost_sinc
 from mindspore._c_expression import pyboost_inplace_divs
 from mindspore._c_expression import pyboost_conv2d_ext
-from mindspore._c_expression import pyboost_inner_comm_reduce_scatter
 from mindspore._c_expression import pyboost_mish_ext
 from mindspore._c_expression import pyboost_acos_ext
 from mindspore._c_expression import pyboost_trace_ext
 from mindspore._c_expression import pyboost_inplace_put
 from mindspore._c_expression import pyboost_glu
 from mindspore._c_expression import pyboost_add
+from mindspore._c_expression import pyboost_real_view
 from mindspore._c_expression import pyboost_nllloss_grad
 from mindspore._c_expression import pyboost_inplace_masked_fill_scalar
 from mindspore._c_expression import pyboost_rand_like_ext
@@ -369,6 +361,7 @@ from mindspore._c_expression import pyboost_reduce_max
 from mindspore._c_expression import pyboost_logical_or
 from mindspore._c_expression import pyboost_nllloss_2d_grad
 from mindspore._c_expression import pyboost_reflection_pad_3d_grad
+from mindspore._c_expression import pyboost_masked_fill_scalar
 from mindspore._c_expression import pyboost_inplace_clamp_tensor
 from mindspore._c_expression import pyboost_atanh
 from mindspore._c_expression import pyboost_replication_pad_1d
@@ -381,21 +374,19 @@ from mindspore._c_expression import pyboost_full_like
 from mindspore._c_expression import pyboost_softshrink_grad
 from mindspore._c_expression import pyboost_bmm_ext
 from mindspore._c_expression import pyboost_ring_attention_update
-from mindspore._c_expression import pyboost_dist_comm_all_to_all_v
 from mindspore._c_expression import pyboost_prelu_grad
 from mindspore._c_expression import pyboost_repeat_interleave_tensor
 from mindspore._c_expression import pyboost_add_layernorm_v2
+from mindspore._c_expression import pyboost_inplace_sigmoid
 from mindspore._c_expression import pyboost_bitwise_not
 from mindspore._c_expression import pyboost_matrix_inverse_ext
 from mindspore._c_expression import pyboost_adaptive_avg_pool2d_grad_ext
 from mindspore._c_expression import pyboost_transpose_ext_view
 from mindspore._c_expression import pyboost_logical_xor
-from mindspore._c_expression import pyboost_dist_comm_isend
 from mindspore._c_expression import pyboost_index_add_ext
 from mindspore._c_expression import pyboost_chunk
 from mindspore._c_expression import pyboost_inplace_remainder_tensor_scalar
 from mindspore._c_expression import pyboost_pow_scalar_tensor
-from mindspore._c_expression import pyboost_dist_comm_scatter_tensor
 from mindspore._c_expression import pyboost_kl_div_grad
 from mindspore._c_expression import pyboost_div
 from mindspore._c_expression import pyboost_median_ext
@@ -407,6 +398,7 @@ from mindspore._c_expression import pyboost_mse_loss_grad_ext
 from mindspore._c_expression import pyboost_sign
 from mindspore._c_expression import pyboost_transpose
 from mindspore._c_expression import pyboost_histc_ext
+from mindspore._c_expression import pyboost_to_dtype
 from mindspore._c_expression import pyboost_inplace_clamp_scalar
 from mindspore._c_expression import pyboost_grid_sampler_3d_grad
 from mindspore._c_expression import pyboost_inplace_hardtanh
@@ -428,7 +420,6 @@ from mindspore._c_expression import pyboost_clamp_scalar
 from mindspore._c_expression import pyboost_all_finite
 from mindspore._c_expression import pyboost_inner_index
 from mindspore._c_expression import pyboost_replication_pad_2d_grad
-from mindspore._c_expression import pyboost_dist_comm_reduce
 from mindspore._c_expression import pyboost_log10
 from mindspore._c_expression import pyboost_prod_ext
 from mindspore._c_expression import pyboost_argmin_ext
@@ -436,7 +427,6 @@ from mindspore._c_expression import pyboost_contiguous
 from mindspore._c_expression import pyboost_slice
 from mindspore._c_expression import pyboost_unique_consecutive
 from mindspore._c_expression import pyboost_convolution_grad
-from mindspore._c_expression import pyboost_dist_comm_all_gather_into_tensor_uneven
 from mindspore._c_expression import pyboost_one_hot_ext
 from mindspore._c_expression import pyboost_outer
 from mindspore._c_expression import pyboost_multi_scale_deformable_attn_grad
@@ -450,7 +440,6 @@ from mindspore._c_expression import pyboost_unique2
 from mindspore._c_expression import pyboost_upsample_bilinear2d_grad
 from mindspore._c_expression import pyboost_bincount_ext
 from mindspore._c_expression import pyboost_convolution_str
-from mindspore._c_expression import pyboost_dist_comm_all_gather
 from mindspore._c_expression import pyboost_isneginf
 from mindspore._c_expression import pyboost_gelu_grad_ext
 from mindspore._c_expression import pyboost_kl_div
@@ -471,13 +460,11 @@ from mindspore._c_expression import pyboost_layer_norm_ext
 from mindspore._c_expression import pyboost_uniform_ext
 from mindspore._c_expression import pyboost_erf
 from mindspore._c_expression import pyboost_argsort
-from mindspore._c_expression import pyboost_dist_comm_all_to_all_v_single
 from mindspore._c_expression import pyboost_inplace_scatter_value
 from mindspore._c_expression import pyboost_dropout_grad_ext
 from mindspore._c_expression import pyboost_gather_d
 from mindspore._c_expression import pyboost_binary_cross_entropy_grad
 from mindspore._c_expression import pyboost_ones_like_ext
-from mindspore._c_expression import pyboost_dist_comm_all_reduce
 from mindspore._c_expression import pyboost_moe_token_unpermute_grad
 from mindspore._c_expression import pyboost_new_ones
 from mindspore._c_expression import pyboost_ones
@@ -512,6 +499,7 @@ from mindspore._c_expression import pyboost_normal_tensor_float
 from mindspore._c_expression import pyboost_adamw
 from mindspore._c_expression import pyboost_masked_scatter
 from mindspore._c_expression import pyboost_multinomial_ext
+from mindspore._c_expression import pyboost_inplace_masked_scatter
 from mindspore._c_expression import pyboost_split_tensor_view
 from mindspore._c_expression import pyboost_mm_ext
 from mindspore._c_expression import pyboost_upsample_trilinear3d
@@ -519,6 +507,7 @@ from mindspore._c_expression import pyboost_bernoulli_ext
 from mindspore._c_expression import pyboost_hsigmoid
 from mindspore._c_expression import pyboost_type_as
 from mindspore._c_expression import pyboost_relu_grad
+from mindspore._c_expression import pyboost_inner_unique
 from mindspore._c_expression import pyboost_batch_norm_elemt_grad
 from mindspore._c_expression import pyboost_conv3d_padding
 from mindspore._c_expression import pyboost_group_norm_grad
@@ -552,6 +541,32 @@ from mindspore._c_expression import pyboost_moe_compute_expert_tokens
 from mindspore._c_expression import pyboost_fused_infer_attention_score
 from mindspore._c_expression import pyboost_moe_init_routing_quant_v2
 from mindspore._c_expression import pyboost_moe_init_routing
+from mindspore._c_expression import pyboost_inner_comm_isend
+from mindspore._c_expression import pyboost_dist_comm_broadcast
+from mindspore._c_expression import pyboost_inner_comm_irecv
+from mindspore._c_expression import pyboost_inner_comm_all_gather
+from mindspore._c_expression import pyboost_dist_comm_barrier
+from mindspore._c_expression import pyboost_dist_comm_gather
+from mindspore._c_expression import pyboost_dist_comm_reduce_scatter_tensor_uneven
+from mindspore._c_expression import pyboost_inner_comm_all_to_all_v
+from mindspore._c_expression import pyboost_dist_comm_reduce_scatter
+from mindspore._c_expression import pyboost_dist_comm_all_gather_into_tensor
+from mindspore._c_expression import pyboost_dist_comm_all_to_all_v_c
+from mindspore._c_expression import pyboost_dist_comm_gather_into_tensor
+from mindspore._c_expression import pyboost_dist_comm_irecv
+from mindspore._c_expression import pyboost_inner_comm_all_reduce
+from mindspore._c_expression import pyboost_dist_comm_batch_isend_irecv
+from mindspore._c_expression import pyboost_dist_comm_reduce_scatter_tensor
+from mindspore._c_expression import pyboost_dist_comm_scatter
+from mindspore._c_expression import pyboost_inner_comm_reduce_scatter
+from mindspore._c_expression import pyboost_dist_comm_all_to_all_v
+from mindspore._c_expression import pyboost_dist_comm_isend
+from mindspore._c_expression import pyboost_dist_comm_scatter_tensor
+from mindspore._c_expression import pyboost_dist_comm_reduce
+from mindspore._c_expression import pyboost_dist_comm_all_gather_into_tensor_uneven
+from mindspore._c_expression import pyboost_dist_comm_all_gather
+from mindspore._c_expression import pyboost_dist_comm_all_to_all_v_single
+from mindspore._c_expression import pyboost_dist_comm_all_reduce
 from mindspore._c_expression import pyboost_gmm_v2_backward_fusion
 from mindspore._c_expression import pyboost_any
 from mindspore._c_expression import pyboost_pixel_shuffle
@@ -560,6 +575,7 @@ from mindspore._c_expression import pyboost_gmm
 from mindspore._c_expression import pyboost_einsum_ext
 from mindspore._c_expression import pyboost_func_max_pool2d
 from mindspore._c_expression import pyboost_moe_token_unpermute
+from mindspore._c_expression import pyboost_dropout2d_ext
 from mindspore._c_expression import pyboost_gmm_v2
 from mindspore._c_expression import pyboost_func_dropout_ext
 from mindspore._c_expression import pyboost_inplace_exponential
@@ -639,28 +655,6 @@ class CumminExt(Primitive):
 cummin_ext_op=CumminExt()
 
 
-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 MultiScaleDeformableAttn(Primitive):
     r"""
     
@@ -1270,28 +1264,6 @@ class GroupNorm(Primitive):
 group_norm_op=GroupNorm()
 
 
-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 Copy(Primitive):
     r"""
     .. code-block::
@@ -1557,6 +1529,35 @@ class NonZero(Primitive):
 non_zero_op=NonZero()
 
 
+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 InplaceDiv(Primitive):
     r"""
     .. code-block::
@@ -1919,28 +1920,6 @@ class InplaceMul(Primitive):
 inplace_mul_op=InplaceMul()
 
 
-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 Acosh(Primitive):
     r"""
     .. code-block::
@@ -2462,32 +2441,6 @@ class Rsqrt(Primitive):
 rsqrt_op=Rsqrt()
 
 
-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 SequenceConcat(Primitive):
     r"""
     .. code-block::
@@ -3893,28 +3846,6 @@ class Xlogy(Primitive):
 xlogy_op=Xlogy()
 
 
-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 Max(Primitive):
     r"""
     .. code-block::
@@ -4043,42 +3974,6 @@ class LogSoftmaxExt(Primitive):
 log_softmax_ext_op=LogSoftmaxExt()
 
 
-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 InsertGemV2InBackward(Primitive):
     r"""
     
@@ -4209,6 +4104,43 @@ class Dot(Primitive):
 dot_op=Dot()
 
 
+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):
+        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_sign(self, [input])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input)
+        return res
+
+
+inplace_sign_op=InplaceSign()
+
+
 class IHFFT2(Primitive):
     r"""
     .. code-block::
@@ -5389,28 +5321,6 @@ class SplitWithSizeView(Primitive):
 split_with_size_view_op=SplitWithSizeView()
 
 
-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 InplaceIndexPut(Primitive):
     r"""
     .. code-block::
@@ -5510,30 +5420,49 @@ class ViewAs(Primitive):
 view_as_op=ViewAs()
 
 
-class EmbeddingFeatureMappingTableSize(Primitive):
+class DequantSwigluQuant(Primitive):
     r"""
     .. code-block::
 
-        prim = ops.EmbeddingFeatureMappingTableSize()
-        out = prim(table_name)
+        prim = ops.DequantSwigluQuant()
+        out = prim(x, weight_scale, activation_scale, bias, quant_scale, quant_offset, group_index, activate_left, quant_mode)
 
     is equivalent to
 
     .. code-block::
 
-        ops.embedding_feature_mapping_table_size(table_name)
+        ops.dequant_swiglu_quant(x, weight_scale, activation_scale, bias, quant_scale, quant_offset, group_index, activate_left, quant_mode)
 
-    Refer to :func:`mindspore.ops.embedding_feature_mapping_table_size` for more details.
+    Refer to :func:`mindspore.ops.dequant_swiglu_quant` for more details.
     """
+    __mindspore_signature__ =  (
+        sig.make_sig('x'),
+        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, table_name):
-        return super().__call__(table_name)
+    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
 
 
-embedding_feature_mapping_table_size_op=EmbeddingFeatureMappingTableSize()
+dequant_swiglu_quant_op=DequantSwigluQuant()
 
 
 class HardtanhGrad(Primitive):
@@ -5565,28 +5494,6 @@ class HardtanhGrad(Primitive):
 hardtanh_grad_op=HardtanhGrad()
 
 
-class DistCommGather(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 Lerp(Primitive):
     r"""
     .. code-block::
@@ -5650,6 +5557,36 @@ class Triu(Primitive):
         return res
 
 
+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 LogAddExp(Primitive):
     r"""
     .. code-block::
@@ -6238,60 +6175,6 @@ class FFTOrtho(Primitive):
 fft_ortho_op=FFTOrtho()
 
 
-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 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 ReplicationPad2D(Primitive):
     r"""
     
@@ -6870,28 +6753,6 @@ class AtanExt(Primitive):
 atan_ext_op=AtanExt()
 
 
-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 AdaptiveMaxPool2D(Primitive):
     r"""
     Performs 2D adaptive max pooling on a multi-plane input signal.
@@ -7063,53 +6924,6 @@ class Cummin(Primitive):
         return super().__call__(input, self.axis)
 
 
-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 KvScaleCache(Primitive):
     r"""
     .. code-block::
@@ -7590,28 +7404,6 @@ class Embedding(Primitive):
 embedding_op=Embedding()
 
 
-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 NewEmpty(Primitive):
     r"""
     
@@ -9260,6 +9052,35 @@ class Mul(Primitive):
 mul_op=Mul()
 
 
+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 ResizeNearestNeighborGrad(Primitive):
     r"""
     Compute gradient of `ResizeNearestNeighbor` operator.
@@ -9399,6 +9220,32 @@ class RandpermExt(Primitive):
 randperm_ext_op=RandpermExt()
 
 
+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 DiagonalView(Primitive):
     r"""
     
@@ -10209,52 +10056,27 @@ 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
 
 
 empty_like_op=EmptyLike()
 
 
-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 AssignSub(Primitive):
     r"""
     .. code-block::
@@ -10461,28 +10283,6 @@ class SplitWithSize(Primitive):
 split_with_size_op=SplitWithSize()
 
 
-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 L1LossBackwardExt(Primitive):
     r"""
     
@@ -10740,11 +10540,40 @@ class SoftMarginLoss(Primitive):
         return res
 
 
-class LessEqual(Primitive):
+class InplaceIndexFillScalar(Primitive):
     r"""
-    .. code-block::
-
-        prim = ops.LessEqual()
+    
+    """
+    __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_scalar(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_scalar_op=InplaceIndexFillScalar()
+
+
+class LessEqual(Primitive):
+    r"""
+    .. code-block::
+
+        prim = ops.LessEqual()
         out = prim(input, other)
 
     is equivalent to
@@ -11188,48 +11017,6 @@ class MoeDistributeDispatch(Primitive):
 moe_distribute_dispatch_op=MoeDistributeDispatch()
 
 
-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 InplaceFloorDivides(Primitive):
     r"""
     
@@ -11275,6 +11062,28 @@ class ACosGrad(Primitive):
 a_cos_grad_op=ACosGrad()
 
 
+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 CumsumExt(Primitive):
     r"""
     .. code-block::
@@ -11770,60 +11579,6 @@ class AsinGrad(Primitive):
 asin_grad_op=AsinGrad()
 
 
-class DistCommGatherIntoTensor(Primitive):
-    r"""
-    
-    """
-    @prim_arg_register
-    def __init__(self):
-        pass
-
-    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_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, other, input, rank_size, dst, rank_id, group)
-        return res
-
-
-dist_comm_gather_into_tensor_op=DistCommGatherIntoTensor()
-
-
-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 Svd(Primitive):
     r"""
     Computes the singular value decompositions of one or more matrices.
@@ -12039,32 +11794,6 @@ class ReflectionPad3D(Primitive):
 reflection_pad_3d_op=ReflectionPad3D()
 
 
-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 Sinh(Primitive):
     r"""
     .. code-block::
@@ -12950,20 +12679,21 @@ 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
 
 
@@ -13135,7 +12865,7 @@ 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: ``()`` .
 
@@ -13162,7 +12892,7 @@ 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.
@@ -13175,28 +12905,6 @@ class Squeeze(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 LerpScalar(Primitive):
     r"""
     .. code-block::
@@ -13230,28 +12938,6 @@ class LerpScalar(Primitive):
 lerp_scalar_op=LerpScalar()
 
 
-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 Square(Primitive):
     r"""
     .. code-block::
@@ -13994,28 +13680,6 @@ class UnstackExtView(Primitive):
 unstack_ext_view_op=UnstackExtView()
 
 
-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 GridSampler2DGrad(Primitive):
     r"""
     Computes gradients for GridSampler2D operation.
@@ -14486,49 +14150,33 @@ class Range(Primitive):
         return super().__call__(start, end, step, self.maxlen)
 
 
-class EmbeddingApplyRmsprop(Primitive):
+class InplaceIndexCopy(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),
+        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("_process_node_engine_id", 'PS')
+        self.add_prim_attr("side_effect_mem", True)
 
-    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)
+    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
 
 
-embedding_apply_rmsprop_op=EmbeddingApplyRmsprop()
+inplace_index_copy_op=InplaceIndexCopy()
 
 
 class MaximumGrad(Primitive):
@@ -15844,28 +15492,6 @@ class Atan(Primitive):
 atan_op=Atan()
 
 
-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 CrossEntropyLossGrad(Primitive):
     r"""
     .. code-block::
@@ -16279,31 +15905,9 @@ class LayerNormGradV3(Primitive):
         return super().__call__(x, dy, variance, mean, gamma, self.begin_norm_axis, self.begin_params_axis)
 
 
-class DistCommScatter(Primitive):
+class AsinExt(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 AsinExt(Primitive):
-    r"""
-    .. code-block::
+    .. code-block::
 
         prim = ops.AsinExt()
         out = prim(input)
@@ -16539,56 +16143,6 @@ class Sinc(Primitive):
 sinc_op=Sinc()
 
 
-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 InplaceDivs(Primitive):
     r"""
     .. code-block::
@@ -16692,28 +16246,6 @@ class Conv2DExt(Primitive):
 conv2d_ext_op=Conv2DExt()
 
 
-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 MishExt(Primitive):
     r"""
     .. code-block::
@@ -17028,6 +16560,28 @@ class Gather(Primitive):
         return super().__call__(input_params, input_indices, axis, self.batch_dims)
 
 
+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 NLLLossGrad(Primitive):
     r"""
     Computes the gradients of `NLLLoss`.
@@ -17947,6 +17501,28 @@ class ReflectionPad3DGrad(Primitive):
 reflection_pad_3d_grad_op=ReflectionPad3DGrad()
 
 
+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 InplaceClampTensor(Primitive):
     r"""
     .. code-block::
@@ -18393,28 +17969,6 @@ class RingAttentionUpdate(Primitive):
 ring_attention_update_op=RingAttentionUpdate()
 
 
-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 PReLUGrad(Primitive):
     r"""
     Gradients of PReLU operation.
@@ -18572,6 +18126,43 @@ class AddLayerNormV2(Primitive):
 add_layernorm_v2_op=AddLayerNormV2()
 
 
+class InplaceSigmoid(Primitive):
+    r"""
+    .. code-block::
+
+        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 BitwiseNot(Primitive):
     r"""
     
@@ -18793,28 +18384,6 @@ class LogicalXor(Primitive):
 logical_xor_op=LogicalXor()
 
 
-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 ExtractImagePatches(Primitive):
     r"""
     .. code-block::
@@ -18980,28 +18549,6 @@ class PowScalarTensor(Primitive):
 pow_scalar_tensor_op=PowScalarTensor()
 
 
-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 KLDivGrad(Primitive):
     r"""
     .. code-block::
@@ -19526,13 +19073,13 @@ 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.
     """
@@ -19540,14 +19087,14 @@ 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
 
 
@@ -19647,6 +19194,35 @@ class HistcExt(Primitive):
 histc_ext_op=HistcExt()
 
 
+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 InplaceClampScalar(Primitive):
     r"""
     .. code-block::
@@ -20536,40 +20112,6 @@ class CholeskyGrad(Primitive):
 cholesky_grad_op=CholeskyGrad()
 
 
-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 ReplicationPad2DGrad(Primitive):
     r"""
     
@@ -20592,38 +20134,9 @@ class ReplicationPad2DGrad(Primitive):
 replication_pad_2d_grad_op=ReplicationPad2DGrad()
 
 
-class DistCommReduce(Primitive):
+class Log10(Primitive):
     r"""
-    
-    """
-    __mindspore_signature__ =  (
-        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):
-        self.add_prim_attr("side_effect_mem", True)
-
-    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
-
-
-dist_comm_reduce_op=DistCommReduce()
-
-
-class Log10(Primitive):
-    r"""
-    .. code-block::
+    .. code-block::
 
         prim = ops.Log10()
         out = prim(input)
@@ -20999,28 +20512,6 @@ class ConvolutionGrad(Primitive):
 convolution_grad_op=ConvolutionGrad()
 
 
-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 OneHotExt(Primitive):
     r"""
     Computes a one-hot tensor.
@@ -22038,7 +21529,7 @@ 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.
@@ -22319,28 +21810,6 @@ class ConvolutionStr(Primitive):
 convolution_str_op=ConvolutionStr()
 
 
-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 IsNegInf(Primitive):
     r"""
     .. code-block::
@@ -23396,28 +22865,6 @@ class ArgSort(Primitive):
 argsort_op=ArgSort()
 
 
-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 BiasAddGrad(Primitive):
     r"""
     Computes gradients of BiasAdd.
@@ -23623,34 +23070,6 @@ class OnesLikeExt(Primitive):
 ones_like_ext_op=OnesLikeExt()
 
 
-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 MoeTokenUnpermuteGrad(Primitive):
     r"""
     .. code-block::
@@ -25453,6 +24872,60 @@ class OnesLike(Primitive):
 ones_like_op=OnesLike()
 
 
+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):
+        self.add_prim_attr("side_effect_mem", True)
+
+    def __call__(self, input, mask, source):
+        # Add for jit context.
+        if jit_context() and jit_context().compiled:
+            return jit_context().default_output()
+        res = pyboost_inplace_masked_scatter(self, [input, mask, source])
+        # Add for jit context.
+        if jit_context():
+            return jit_context().run_op(self, res, input, mask, source)
+        return res
+
+
+inplace_masked_scatter_op=InplaceMaskedScatter()
+
+
 class SplitTensorView(Primitive):
     r"""
     
@@ -25860,49 +25333,32 @@ class ReluGrad(Primitive):
 relu_grad_op=ReluGrad()
 
 
-class EmbeddingApplyFtrl(Primitive):
+class InnerUnique(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),
+        sig.make_sig('input'),
+        sig.make_sig('sorted', default=True),
+        sig.make_sig('return_inverse', default=False),
     )
 
     @prim_arg_register
     def __init__(self):
-        self.add_prim_attr("_process_node_engine_id", 'PS')
+        pass
 
-    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)
+    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
 
 
-embedding_apply_ftrl_op=EmbeddingApplyFtrl()
+inner_unique_op=InnerUnique()
 
 
 class ResizeNearestNeighbor(Primitive):
@@ -26672,48 +26128,6 @@ class AsinhExt(Primitive):
 asinh_ext_op=AsinhExt()
 
 
-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 InplaceBernoulliScalar(Primitive):
     r"""
     .. code-block::
@@ -26945,13 +26359,13 @@ 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.
     """
@@ -26972,20 +26386,21 @@ 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
 
 
@@ -28106,36 +27521,29 @@ class MoeInitRouting(Primitive):
 moe_init_routing_op=MoeInitRouting()
 
 
-class GmmV2BackwardFusion(Primitive):
+class InnerCommIsend(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, 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_fusion(self, [grad, weight, group_list, group_list_type])
+        res = pyboost_inner_comm_isend(self, [input, dst, group, tag])
         # 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, input, dst, group, tag)
         return res
 
 
-gmm_v2_backward_fusion_op=GmmV2BackwardFusion()
+inner_comm_isend_op=InnerCommIsend()
 
 
-class Any(Primitive):
+class DistCommBroadcast(Primitive):
     r"""
     
     """
@@ -28143,21 +27551,21 @@ class Any(Primitive):
     def __init__(self):
         pass
 
-    def __call__(self, input):
+    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_any(self, [input])
+        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)
+            return jit_context().run_op(self, res, input, src, rank_id, group)
         return res
 
 
-any_op=Any()
+dist_comm_broadcast_op=DistCommBroadcast()
 
 
-class PixelShuffle(Primitive):
+class InnerCommIrecv(Primitive):
     r"""
     
     """
@@ -28165,56 +27573,648 @@ class PixelShuffle(Primitive):
     def __init__(self):
         pass
 
-    def __call__(self, input, upscale_factor):
+    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_pixel_shuffle(self, [input, upscale_factor])
+        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, input, upscale_factor)
+            return jit_context().run_op(self, res, tag, src, shape, group, dtype_to_type_id('InnerCommIrecv', 'dtype', dtype))
         return res
 
 
-pixel_shuffle_op=PixelShuffle()
+inner_comm_irecv_op=InnerCommIrecv()
 
 
-class GmmV2Backward(Primitive):
+class InnerCommAllGather(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, rank_size, group):
         # 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_inner_comm_all_gather(self, [input, rank_size, group])
         # 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, rank_size, group)
         return res
 
 
-gmm_v2_backward_op=GmmV2Backward()
+inner_comm_all_gather_op=InnerCommAllGather()
 
 
-class Gmm(Primitive):
+class DistCommBarrier(Primitive):
     r"""
     
     """
-    __mindspore_signature__ =  (
-        sig.make_sig('x'),
+    @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 DistCommGather(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 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 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 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 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 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 DistCommGatherIntoTensor(Primitive):
+    r"""
+    
+    """
+    @prim_arg_register
+    def __init__(self):
+        pass
+
+    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_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, other, input, rank_size, dst, rank_id, group)
+        return res
+
+
+dist_comm_gather_into_tensor_op=DistCommGatherIntoTensor()
+
+
+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 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 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 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 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 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 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 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 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 DistCommReduce(Primitive):
+    r"""
+    
+    """
+    __mindspore_signature__ =  (
+        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):
+        self.add_prim_attr("side_effect_mem", True)
+
+    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
+
+
+dist_comm_reduce_op=DistCommReduce()
+
+
+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 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 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 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 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 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 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 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 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),
@@ -28335,6 +28335,28 @@ class MoeTokenUnpermute(Primitive):
 moe_token_unpermute_op=MoeTokenUnpermute()
 
 
+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 GmmV2(Primitive):
     r"""