[CodeStyle] black -> ruff format migration - part 2

python/paddle/apy/sys/__builtin_registry_item__.py

@@ -16,7 +16,6 @@
 
 
 class RegistryEntry:
-
     def __init__(self):
         self.__tag_name__ = None
         self.__nice__ = None
@@ -48,15 +47,13 @@ def __call__(self, tag_name, nice):
 
 
 class RegistryObject:
-
     def __init__(self, tag_name, nice):
         self.tag_name = tag_name
         self.nice = nice
         self.value = None
 
 
 class RegisterItemDecorator:
-
     def __init__(self, register_obj):
         self.register_obj = register_obj
 

python/paddle/base/framework.py

@@ -8546,7 +8546,6 @@ def set_op_roles(block, op_role, always_forward_ops):
 # there would be always_forward_ops in your region, you should use "auto_complete_op_role"
 @signature_safe_contextmanager
 def pir_op_role_guard(op_role: int - 1) -> Generator[None, None, None]:
-
     if paddle.framework.in_pir_mode():
         original_op_rope = pir.get_op_role()
         pir.set_op_role(op_role)
@@ -8559,7 +8558,6 @@ def pir_op_role_guard(op_role: int - 1) -> Generator[None, None, None]:
 
 @signature_safe_contextmanager
 def pir_chunk_id_guard(chunk_id: int - 1) -> Generator[None, None, None]:
-
     if paddle.framework.in_pir_mode():
         original_chunk_id = pir.get_chunk_id()
         pir.set_chunk_id(chunk_id)
@@ -8572,7 +8570,6 @@ def pir_chunk_id_guard(chunk_id: int - 1) -> Generator[None, None, None]:
 
 @signature_safe_contextmanager
 def pir_op_name_guard(op_name: str) -> Generator[None, None, None]:
-
     if paddle.framework.in_pir_mode() and core._is_bwd_prim_enabled():
         original_comp_op_name = pir.get_comp_op_name()
         pir.set_comp_op_name(op_name)

python/paddle/base/variable_index.py

@@ -805,7 +805,6 @@ def get_tensor_with_basic_indexing(
                     attrs['decrease_axis'],
                 )
         else:
-
             target_block = paddle.static.default_main_program().current_block()
 
             slice_out_var = target_block.create_var(

python/paddle/decomposition/recompute.py

@@ -752,7 +752,6 @@ def partition_joint_graph(
 def replace_mid_values_with_forward_subgraph(
     program, saved_values, mid_values, fwd_op_end_idx, backward_op_start_idx
 ):
-
     def _extract_forward_recompute_subgraph_for_backward(
         saved_values, mid_values
     ):

python/paddle/distributed/auto_parallel/pipelining/stage.py

@@ -621,7 +621,6 @@ def forward_maybe_with_nosync(self, *args, **kwargs):
     def backward_maybe_with_nosync(
         self, backward_type, bwd_kwargs: dict, last_backward=False
     ) -> tuple[tuple[paddle.Tensor | None, ...], list[dict[str, Any] | None]]:
-
         def perform_backward(
             backward_type,
         ) -> Callable[
@@ -1245,7 +1244,6 @@ def _prepare_forward_infra(
         args: tuple[Any, ...],
         kwargs: dict[str, Any] | None = None,
     ) -> tuple[Any, ...]:
-
         assert num_microbatches is not None, "num_microbatches must be provided"
 
         outputs: tuple[Any, ...] = ()

python/paddle/distributed/auto_parallel/pipelining/utils.py

@@ -152,7 +152,6 @@ def _get_stage_mesh(stage_index, pp_group_size, style=None):
     if style is not None:
         raise ValueError(f"Unknown style: {style}, style can be None, v.")
     else:
-
         pp_idx = stage_index % pp_group_size
         return _get_pp_mesh(pp_idx)
 

python/paddle/distributed/auto_parallel/process_mesh.py

@@ -447,7 +447,6 @@ def get_group(
                 if hasattr(fleet.fleet, "_hcg"):
                     hcg = fleet.get_hybrid_communicate_group()
                     if hcg is not None:
-
                         parallel_group_map = {
                             "pp": hcg.get_pipe_parallel_group,
                             "dp": hcg.get_data_parallel_group,

python/paddle/distributed/auto_parallel/static/pir_pass.py

@@ -232,7 +232,6 @@ def apply_partition_pass(program, block=None):
 
 
 class ReshardPasses:
-
     @staticmethod
     def decompose_reshard_pass(dist_program):
         # split composed reshard op into atomic reshard ops, which would increase the opportunity of reshard Re-Use in following fold_reshard_pass.
@@ -445,7 +444,6 @@ def remove_sub_block_unused_inputs(op):
 
 
 class RemovePasses:
-
     @staticmethod
     def remove_other_rank_op_pass(dist_program):
         # pruning op and value not belong to cur rank
@@ -1855,7 +1853,6 @@ def fuse_attention_ffn_qkv_pass(
 
                 # Fuse params and init pir program fusion params.
                 with paddle.base.dygraph.guard():
-
                     dyparam_dtype = concated_dy_param_list[0].dtype
                     for param in concated_dy_param_list:
                         assert (

python/paddle/distributed/auto_parallel/static/reshard_funcs/global_to_sub_mesh_func.py

@@ -23,7 +23,6 @@
 
 class GlobalToSubMeshFunction(ReshardFunction):
     def is_suitable(self, src_dist_attr, dst_dist_attr):
-
         # NOTE we could allow the src_dist_attr is not replicated and reshard it as replicated before go through the global_to_sub logic
         # but the dst_dist_attr should be replicated otherwise there will be un-defined result when change the mesh.
         if not is_replicated(dst_dist_attr):
@@ -39,7 +38,6 @@ def is_suitable(self, src_dist_attr, dst_dist_attr):
             return out_mesh in sub_meshes
 
     def reshard(self, src_dist_attr, dst_dist_attr, src_value, dst_type):
-
         # reshard operand as replicated before change the mesh.
         if not is_replicated(src_dist_attr):
             tmp_dist_attr = (

python/paddle/distributed/auto_parallel/static/tuner/to_distributed_api_patterns.py

@@ -553,7 +553,6 @@ def apply(
         value_states,
         attention_mask,
     ):
-
         bsz, q_len, num_heads, head_dim = query_states.shape
         _, kv_seq_len, _, _ = value_states.shape
 
@@ -1263,7 +1262,6 @@ def apply(x, w1, b1, w2, b2):
 
 
 def match_pattern(pattern, program):
-
     def _compare_op_node(src, tgt):
         """Compare whether two op nodes are equivalent."""
         if src.name() != tgt.name():

python/paddle/distributed/communicator.py

@@ -30,6 +30,7 @@
 Communicator is used for async distribute training in distribute_transpiler mode.
 It's a wrapper of a cpp class Communicator and should be used inside fleet API.
 """
+
 import paddle
 from paddle.distributed.ps.utils.public import DistributedMode
 from paddle.framework import core

python/paddle/distributed/fleet/base/role_maker.py

@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """Definition of Role Makers."""
+
 from __future__ import annotations
 
 import os

python/paddle/distributed/fleet/fleet.py

@@ -39,7 +39,6 @@
 from .utils.log_util import logger, set_log_level
 
 if TYPE_CHECKING:
-
     from collections.abc import (
         Callable,
         Iterable,

python/paddle/distributed/fleet/meta_parallel/dualpipev.py

@@ -759,7 +759,6 @@ def forward_backward_pipeline(
                         main_stage=True,
                     )
             else:
-
                 self._forward_backward_pass(
                     0,
                     1,

python/paddle/distributed/fleet/meta_parallel/zero_bubble_utils.py

@@ -27,7 +27,6 @@
 
 
 class WeightGradStore:
-
     enabled = False
     cache = []
     funcs_queue = queue.Queue()
@@ -55,7 +54,6 @@ def clear(cls) -> None:
 
 
 class EventStore:
-
     event = None
 
     @classmethod

python/paddle/distributed/launch/job/container.py

@@ -168,7 +168,6 @@ def status(self):
             return Status.FAILED
 
     def __str__(self):
-
         need_print = os.environ.get('FLAGS_print_launcher_env', 'false').lower()
         if need_print == 'true' or need_print == '1':
             return f'Container rank {self._rank} status {self.status} cmd {self._entrypoint} code {self.exit_code} log {self.errfile} \nenv {self._env}'

python/paddle/distributed/passes/pipeline_scheduler_pass/pipeline_1f1b.py

@@ -34,7 +34,6 @@
 
 @register_pass("pipeline_scheduler_1F1B")
 class Pipeline1F1BPass(PipelinePassBase):
-
     def __init__(self):
         super().__init__()
         self.jobs_in_stable_phase = [self.BACKWARD, self.FORWARD]

python/paddle/distributed/passes/pipeline_scheduler_pass/pipeline_pass_base.py

@@ -27,7 +27,6 @@
 
 
 class PipelinePassBase(PassBase):
-
     # Pipeline stages
     RECV_FORWARD = "recv_forward"
     SEND_BACKWARD = "send_backward"

python/paddle/distributed/transpiler/geo_sgd_transpiler.py

@@ -24,6 +24,7 @@
 4. append sum ops that should run on current server instance.
 5. add listen_and_serv op
 """
+
 import collections
 
 from paddle import framework

python/paddle/incubate/cc/ap/apy_to_axpr_json.py

@@ -31,7 +31,6 @@ def convert_python_stmts_to_axpr_json(python_code_stmts_str):
 
 @dataclass
 class AnfExpr:
-
     def DumpToFileAsJson(self, file_name):
         with open(file_name, "w") as f:
             json.dump(self.value, f, indent=2)