apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-06 17:17:14 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[LLM] fix multinode bugs (#2945)

Browse Source 
* [LLM] fix multinode bugs

* [LLM] fix multinode bugs

* [LLM] fix multinode bugs

* [LLM] fix ci bugs

* fix ci bugs

* fix ci bugs
This commit is contained in:
ltd0924
2025-07-22 20:23:37 +08:00
committed by GitHub
parent 69be77c8c0
commit b0f1e0eef4
9 changed files with 68 additions and 87 deletions
Download Patch File Download Diff File Expand all files Collapse all files

40
fastdeploy/engine/args_utils.py
View File

@@ -124,19 +124,9 @@ class EngineArgs:
Ratio of tokens to process in a block.
"""
dist_init_ip: Optional[str] = None
ips: Optional[List[str]] = None
"""
The master node ip of multinode deployment
"""
nnodes: int = 1
"""
The number of nodes in multinode deployment
"""
node_rank: int = 0
"""
The rank of the current node in multinode deployment
The ips of multinode deployment
"""
swap_space: float = None
@@ -495,25 +485,11 @@ class EngineArgs:
# Cluster system parameters group
system_group = parser.add_argument_group("System Configuration")
system_group.add_argument(
"--dist-init-ip",
default=EngineArgs.dist_init_ip,
"--ips",
type=lambda s: s.split(",") if s else None,
default=EngineArgs.ips,
help=
"IP addresses of master node.")
system_group.add_argument(
"--nnodes",
type=int,
default=EngineArgs.nnodes,
help=
"The number of all nodes.")
system_group.add_argument(
"--node-rank",
type=int,
default=EngineArgs.node_rank,
help=
"node rank id (range [0, nnodes)).")
"IP addresses of all nodes participating in distributed inference.")
# Performance tuning parameters group
@@ -813,9 +789,7 @@ class EngineArgs:
max_num_seqs=self.max_num_seqs,
speculative_config=speculative_cfg,
max_num_batched_tokens=self.max_num_batched_tokens,
dist_init_ip=self.dist_init_ip,
nnodes=self.nnodes,
node_rank=self.node_rank,
ips=self.ips,
use_warmup=self.use_warmup,
engine_worker_queue_port=self.engine_worker_queue_port,
limit_mm_per_prompt=self.limit_mm_per_prompt,