fix lint.

Author: Weichao Luo

lightllm/server/router/model_infer/mode_backend/base_backend.py

@@ -524,7 +524,7 @@ def _post_handle(
                 eos_ids=self.eos_id,
                 extra_post_req_handle_func=extra_post_req_handle_func,
                 is_master_in_dp=self.is_master_in_dp,
-                call_post_handle_for_chunk=call_post_handle_for_chunk
+                call_post_handle_for_chunk=call_post_handle_for_chunk,
             )
 
         g_infer_context.req_manager.req_sampling_params_manager.update_reqs_token_counter(

lightllm/server/router/model_infer/mode_backend/chunked_prefill/impl.py

@@ -138,7 +138,7 @@ def prefill_normal(
             next_token_logprobs=next_token_logprobs_cpu,
             run_reqs_update_packs=update_packs,
             extra_post_req_handle_func=self.extra_post_req_handle_func,
-            call_post_handle_for_chunk=self.call_post_handle_for_chunk
+            call_post_handle_for_chunk=self.call_post_handle_for_chunk,
         )
         # 第四阶段
         event_pack.notify_pre_post_handle()
@@ -185,7 +185,7 @@ def decode_normal(
             next_token_ids=next_token_ids_cpu,
             next_token_logprobs=next_token_logprobs_cpu,
             run_reqs_update_packs=update_packs,
-            extra_post_req_handle_func=self.extra_post_req_handle_func
+            extra_post_req_handle_func=self.extra_post_req_handle_func,
         )
 
         # 第四阶段
@@ -351,7 +351,7 @@ def decode_mtp(
             next_token_ids=next_token_ids_cpu[select_mask],
             next_token_logprobs=next_token_logprobs_cpu[select_mask],
             run_reqs_update_packs=update_packs,
-            extra_post_req_handle_func=self.extra_post_req_handle_func
+            extra_post_req_handle_func=self.extra_post_req_handle_func,
         )
         if len(need_free_mem_indexes) > 0:
             g_infer_state_lock.acquire()

lightllm/server/router/model_infer/mode_backend/dp_backend/impl.py

@@ -160,7 +160,7 @@ def prefill_normal(
                 next_token_logprobs=next_token_logprobs_cpu,
                 run_reqs_update_packs=update_packs,
                 extra_post_req_handle_func=self.extra_post_req_handle_func,
-                call_post_handle_for_chunk=self.call_post_handle_for_chunk
+                call_post_handle_for_chunk=self.call_post_handle_for_chunk,
             )
             # 第四阶段
             event_pack.notify_pre_post_handle()
@@ -291,7 +291,7 @@ def prefill_overlap(self, event_pack: OverlapEventPack, prefill_reqs: List[Infer
                 next_token_logprobs=next_token_logprobs_cpu,
                 run_reqs_update_packs=update_packs,
                 extra_post_req_handle_func=self.extra_post_req_handle_func,
-                call_post_handle_for_chunk=self.call_post_handle_for_chunk
+                call_post_handle_for_chunk=self.call_post_handle_for_chunk,
             )
             # 第四阶段
             event_pack.notify_pre_post_handle()
@@ -664,7 +664,7 @@ def prefill_overlap_mtp(self, event_pack: OverlapEventPack, prefill_reqs: List[I
                 next_token_logprobs=next_token_logprobs_cpu,
                 run_reqs_update_packs=update_packs,
                 extra_post_req_handle_func=self.extra_post_req_handle_func,
-                call_post_handle_for_chunk=self.call_post_handle_for_chunk
+                call_post_handle_for_chunk=self.call_post_handle_for_chunk,
             )
             event_pack.notify_pre_post_handle()
         else:

lightllm/server/router/model_infer/mode_backend/pd_nixl/impl_for_pd_base.py

@@ -190,7 +190,6 @@ def _handle_chunked_transfer(self, req: InferReq, next_token_id: int = None, nex
                     transfer_state.next_token_logprob = next_token_logprob
 
     async def _transfer_kv_to_remote_paged_batch(self, transfer_reqs: List[KVMoveRequest]):
-        start = time.time()
         requests_by_agents = dict()
         transfer_pages = self.page_scheduer.borrow(len(transfer_reqs))
         # first copy the kv to transfer pages & build notification
@@ -216,7 +215,9 @@ async def _transfer_kv_to_remote_paged_batch(self, transfer_reqs: List[KVMoveReq
                     receive_page = transfer_state.free_page_ids.pop(0)
                     requests_by_agents[decode_id][0].append(page_index)
                     requests_by_agents[decode_id][1].append(receive_page)
-                    is_last = transfer_state.is_finished and start_kv_len + trans_kv_len == transfer_state.current_kv_len
+                    is_last = (
+                        transfer_state.is_finished and start_kv_len + trans_kv_len == transfer_state.current_kv_len
+                    )
 
                     requests_by_agents[decode_id][2].append(
                         RemotePrefillStatus(
@@ -241,8 +242,6 @@ async def _transfer_kv_to_remote_paged_batch(self, transfer_reqs: List[KVMoveReq
             # transfer
             self.nixl_agent.write_blocks_paged(decode_id, transfer_pages, receive_pages, notifications)
 
-        # logger.info(f"transfer kv to remote paged batch: {len(transfer_reqs)} " f"took: {time.time() - start} seconds")
-
     async def _handle_transfer_loop(self):
         while True:
             free_transfer_pages = self.page_scheduer.current_size()
@@ -424,8 +423,7 @@ def _post_remote_prefill(self, req: InferReq, success: bool = True):
         assert group_req_id in self.request_to_first_token, f"{group_req_id} not in request_to_first_token dict"
         token_id, token_logprob = self.request_to_first_token.pop(group_req_id)
 
-
-        #(TODO) figure out how to update req_to_next_token_ids
+        # (TODO) figure out how to update req_to_next_token_ids
         # req.cur_output_len += 1
 
         # pack = InferReqUpdatePack(req, req.cur_output_len)
@@ -439,9 +437,7 @@ def _post_remote_prefill(self, req: InferReq, success: bool = True):
         # )
         return token_id
 
-    def _decode_filter_reqs(
-        self, prefill_reqs: List[InferReq], decode_reqs: List[InferReq]
-    ):
+    def _decode_filter_reqs(self, prefill_reqs: List[InferReq], decode_reqs: List[InferReq]):
         new_prefill_reqs: List[InferReq] = []
         remote_prefill_reqs: List[InferReq] = []
         failed_prefill_reqs: List[InferReq] = []
@@ -534,5 +530,6 @@ def _prefill_abort_remote(self, req_objs: List[InferReq]):
 class PDNIXLBackendBaseChunked(PDNIXLBackendBase, ChunkedPrefillBackend):
     pass
 
+
 class PDNIXLBackendBaseDPChunked(PDNIXLBackendBase, DPChunkedPrefillBackend):
-    pass
+    pass

lightllm/server/router/model_infer/mode_backend/pd_nixl/impl_for_pd_decode.py

@@ -73,15 +73,16 @@ def _pre_handle_finished_reqs(self, finished_reqs: List[InferReq]):
 
         finished_reqs = new_finished_reqs
 
-
     def _get_classed_reqs(
         self,
         req_ids: List[int] = None,
         no_decode: bool = False,
         strict_prefill: bool = False,
         recover_paused: bool = False,
     ):
-        prefill_reqs, decode_reqs = super(type(self), self)._get_classed_reqs(req_ids, no_decode, strict_prefill, recover_paused)
+        prefill_reqs, decode_reqs = super(type(self), self)._get_classed_reqs(
+            req_ids, no_decode, strict_prefill, recover_paused
+        )
         prefill_reqs, decode_reqs, failed_reqs, _ = self._decode_filter_reqs(prefill_reqs, decode_reqs)
 
         if failed_reqs:

lightllm/server/router/model_infer/mode_backend/pd_nixl/impl_for_pd_decode_dp.py

@@ -1,7 +1,6 @@
 from types import MethodType
 import torch.multiprocessing as mp
 from lightllm.utils.log_utils import init_logger
-from lightllm.utils.envs_utils import get_env_start_args
 
 from .impl_for_pd_base import PDNIXLBackendBaseDPChunked
 from .impl_for_pd_decode import PDNIXLBackendForDecodeNode
@@ -20,4 +19,3 @@ def __init__(self, prefill_task_queue: mp.Queue, prefill_done_queue: mp.Queue, n
         self._trigger_remote_prefill = MethodType(PDNIXLBackendForDecodeNode._trigger_remote_prefill, self)
         self._pre_handle_finished_reqs = MethodType(PDNIXLBackendForDecodeNode._pre_handle_finished_reqs, self)
         self._get_classed_reqs = MethodType(PDNIXLBackendForDecodeNode._get_classed_reqs, self)
-

lightllm/server/router/model_infer/mode_backend/pd_nixl/impl_for_pd_prefill_dp.py

@@ -2,8 +2,6 @@
 import torch.multiprocessing as mp
 from lightllm.server.router.model_infer.infer_batch import g_infer_context
 from lightllm.utils.log_utils import init_logger
-from lightllm.utils.envs_utils import get_env_start_args
-# from lightllm.server.router.model_infer.mode_backend.dp_backend.impl import DPChunkedPrefillBackend
 
 from .impl_for_pd_base import PDNIXLBackendBaseDPChunked
 from .impl_for_pd_prefill import PDNIXLBackendForPrefillNode