A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[Misc] update dict parse to EPLBConfig from json dumps to dict unpacking (#23305)
Signed-off-by: rongfu.leng <rongfu.leng@daocloud.io>
rongfu.leng committed
Aug 24, 2025 at 16:06 UTC
1b9b16649c10453fe25ff28313dffa175194a84b
2 files changed
+2
-10
vllm/config/parallel.py
+1
-8
@@ -6,7 +6,7 @@ from dataclasses import field
6
from typing import TYPE_CHECKING, Any, Literal, Optional, Union
7
8
import torch
9
-from pydantic import TypeAdapter, model_validator
9
+from pydantic import model_validator
10
from pydantic.dataclasses import dataclass
11
from torch.distributed import ProcessGroup, ReduceOp
12
from typing_extensions import Self
@@ -56,13 +56,6 @@ class EPLBConfig:
56
This is turned off by default since it will cause communication overhead.
57
"""
58
59
- @classmethod
60
- def from_cli(cls, cli_value: str) -> "EPLBConfig":
61
- """Parse the CLI value for the compilation config.
62
- -O1, -O2, -O3, etc. is handled in FlexibleArgumentParser.
63
- """
64
- return TypeAdapter(EPLBConfig).validate_json(cli_value)
65
-
59
60
@config
61
@dataclass
vllm/engine/arg_utils.py
+1
-2
@@ -455,8 +455,7 @@ class EngineArgs:
455
self.compilation_config = CompilationConfig(
456
**self.compilation_config)
457
if isinstance(self.eplb_config, dict):
458
- self.eplb_config = EPLBConfig.from_cli(json.dumps(
459
- self.eplb_config))
458
+ self.eplb_config = EPLBConfig(**self.eplb_config)
459
# Setup plugins
460
from vllm.plugins import load_general_plugins
461
load_general_plugins()