apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 08:37:06 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[LLM] Update Multinode Deployment (#2830)
Some checks failed
Deploy GitHub Pages / deploy (push) Has been cancelled
Details

Browse Source 
* [LLM] fix multinode bugs

* [LLM] update multinode deployment

* [LLM] update multinode deployment

* [LLM] update multinode deployment

* [LLM] update multinode deployment

* [LLM] update multinode deployment

* [LLM] fix ci bugs

* Update fastdeploy/engine/args_utils.py

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

* [LLM] update random port

* [LLM] update random port

* [LLM] fix ci bugs

* fix ci bugs

---------

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
This commit is contained in:
ltd0924
2025-07-16 23:42:54 +08:00
committed by GitHub
parent d245d1ca6c
commit 9c25dcca0b
11 changed files with 108 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
fastdeploy/worker/worker_process.py
View File

@@ -136,9 +136,9 @@ class PaddleDisWorkerProc():
model_weights_status:
"""
# init worker_ready_signal
max_chips_per_node = 16 if current_platform.is_iluvatar() else 8
self.max_chips_per_node = 16 if current_platform.is_iluvatar() else 8
array_size = min(
max_chips_per_node, self.parallel_config.tensor_parallel_size *
self.max_chips_per_node, self.parallel_config.tensor_parallel_size *
self.parallel_config.expert_parallel_size)
workers_ready = np.zeros(shape=[array_size], dtype=np.int32)
self.worker_ready_signal = IPCSignal(
@@ -148,10 +148,10 @@ class PaddleDisWorkerProc():
suffix=self.parallel_config.engine_pid,
create=False)
self.worker_ready_signal.value[self.local_rank %
max_chips_per_node] = 1
self.max_chips_per_node] = 1
# init worker_healthy_live_signal
workers_alive = np.zeros(shape=[self.ranks], dtype=np.int32)
workers_alive = np.zeros(shape=[array_size], dtype=np.int32)
self.worker_healthy_live_signal = IPCSignal(
name="worker_healthy_live_signal",
array=workers_alive,
@@ -205,7 +205,7 @@ class PaddleDisWorkerProc():
Tmp loop function for ep utill DP is supported
"""
while True:
self.worker_healthy_live_signal.value[self.local_rank] = int(
self.worker_healthy_live_signal.value[self.local_rank % self.max_chips_per_node] = int(
time.time())
if self.fd_config.parallel_config.tensor_parallel_rank == 0 and self.task_queue.num_tasks(