Skip to content

Commit 53e8949

Browse files
author
lt
committed
fix some bug
1 parent d4d0716 commit 53e8949

File tree

2 files changed

+1
-9
lines changed

2 files changed

+1
-9
lines changed

vllm_ascend/eplb/eplb_updator.py

Lines changed: 1 addition & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,3 @@
1-
#
21
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
32
#
43
# Licensed under the Apache License, Version 2.0 (the "License");
@@ -76,7 +75,7 @@ def init_eplb(self, expert_map_path):
7675
shared_dict = self.shared_dict,
7776
planner_q = self.planner_block_queue,
7877
block_update_q = self.block_update_queue,
79-
redundant_enable = self.redundant_enable,
78+
redundant_enable = self.redundant_enable,
8079
policy_type = 6,
8180
enable_d2d = True
8281
)
@@ -104,7 +103,6 @@ def wakeup_eplb_worker(self):
104103
self.planner_block_queue.put(1)
105104

106105
def forward_before(self):
107-
108106
# Batch after eplb process being triggered, get update info provided by eplb process
109107
if self.update_in_flight and self.weight_update_counter == 0 and self.wait_worker_iterations == self.num_wait_worker_iterations:
110108
self.wait_worker_iterations = 0

vllm_ascend/worker/worker_v1.py

Lines changed: 0 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -178,11 +178,6 @@ def execute_model(
178178
output = self.model_runner.execute_model(scheduler_output)
179179
return output if self.is_driver_worker else None
180180

181-
def get_expert_load(self) -> str:
182-
""" todo 一共几个worker"""
183-
moe_load = self.model_runner.do_get_expert_load()
184-
return moe_load
185-
186181
def load_model(self) -> None:
187182
if self.vllm_config.model_config.enable_sleep_mode:
188183
allocator = CaMemAllocator.get_instance()
@@ -215,7 +210,6 @@ def compile_or_warm_up_model(self) -> None:
215210
set_random_seed(self.model_config.seed)
216211

217212
def get_expert_load(self) -> str:
218-
""" todo 一共几个worker"""
219213
moe_load = self.model_runner.do_get_expert_load()
220214
return moe_load
221215

0 commit comments

Comments
 (0)