EngineX/xc-llm-kunlun
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

提交vllm0.11.0开发分支

Browse Source
This commit is contained in:
chenyili
2025-12-10 17:51:24 +08:00
parent deab7dd0b6
commit 7c22d621fb
175 changed files with 31856 additions and 8683 deletions
Download Patch File Download Diff File Expand all files Collapse all files

112
vllm_kunlun/platforms/envs.py
View File

@@ -4,28 +4,26 @@ import os
from typing import TYPE_CHECKING, Any, Callable, Optional
if TYPE_CHECKING:
VLLM_MULTI_LOGPATH: str = ("./log",)
ENABLE_VLLM_MULTI_LOG: bool = (False,)
ENABLE_VLLM_INFER_HOOK: bool = (False,)
ENABLE_VLLM_OPS_HOOK: bool = (False,)
ENABLE_VLLM_MODULE_HOOK: bool = False
VLLM_MULTI_LOGPATH : str = "./log",
ENABLE_VLLM_MULTI_LOG : bool = False,
ENABLE_VLLM_INFER_HOOK : bool = False,
ENABLE_VLLM_OPS_HOOK : bool = False,
ENABLE_VLLM_MODULE_HOOK : bool = False
def maybe_convert_int(value: Optional[str]) -> Optional[int]:
"""
If the value is None, return None; otherwise, convert the string to an integer and return it.
如果值是None，则返回None；否则将字符串转换为整数并返回。
Args:
value (Optional[str], optional): The optional string to convert. Defaults to None.
value (Optional[str], optional): 要转换的可选字符串. Defaults to None.
Returns:
Optional[int]: If the value is None, return None; otherwise, convert the string to an integer and return it.
Optional[int]: 如果值是None，则返回None；否则将字符串转换为整数并返回.
"""
if value is None:
return None
return int(value)
# The begin-* and end* here are used by the documentation generator
# to extract the used env vars.
@@ -33,56 +31,59 @@ def maybe_convert_int(value: Optional[str]) -> Optional[int]:
xvllm_environment_variables: dict[str, Callable[[], Any]] = {
# path to the logs of redirect-output, abstrac of related are ok
"VLLM_MULTI_LOGPATH": lambda: os.environ.get("VLLM_MULTI_LOGPATH", "./logs"),
# turn on / off multi-log of multi nodes & multi cards
"ENABLE_VLLM_MULTI_LOG": lambda: (
os.environ.get("ENABLE_VLLM_MULTI_LOG", "False").lower() in ("true", "1")
),
# turn on / off XVLLM infer stage log ability
"ENABLE_VLLM_INFER_HOOK": lambda: (
os.environ.get("ENABLE_VLLM_INFER_HOOK", "False").lower() in ("true", "1")
),
# turn on / off XVLLM infer_ops log ability
"ENABLE_VLLM_OPS_HOOK": lambda: (
os.environ.get("ENABLE_VLLM_OPS_HOOK", "False").lower() in ("true", "1")
),
"ENABLE_VLLM_MODULE_HOOK": lambda: (
os.environ.get("ENABLE_VLLM_MODULE_HOOK", "False").lower() in ("true", "1")
),
"VLLM_MULTI_LOGPATH":
lambda: os.environ.get("VLLM_MULTI_LOGPATH", "./logs"),
# turn on / off multi-log of multi nodes & multi cards
"ENABLE_VLLM_MULTI_LOG":
lambda: (os.environ.get("ENABLE_VLLM_MULTI_LOG", "False").lower() in
("true", "1")),
# turn on / off XVLLM infer stage log ability
"ENABLE_VLLM_INFER_HOOK":
lambda: (os.environ.get("ENABLE_VLLM_INFER_HOOK", "False").lower() in
("true", "1")),
# turn on / off XVLLM infer_ops log ability
"ENABLE_VLLM_OPS_HOOK":
lambda: (os.environ.get("ENABLE_VLLM_OPS_HOOK", "False").lower() in
("true", "1")),
"ENABLE_VLLM_MODULE_HOOK":
lambda: (os.environ.get("ENABLE_VLLM_MODULE_HOOK", "False").lower() in
("true", "1")),
# fuse sorted op with fused_moe kernel
"ENABLE_VLLM_MOE_FC_SORTED": lambda: (
os.environ.get("ENABLE_VLLM_MOE_FC_SORTED", "False").lower() in ("true", "1")
),
"ENABLE_VLLM_MOE_FC_SORTED":
lambda: (os.environ.get("ENABLE_VLLM_MOE_FC_SORTED", "False").lower() in
("true", "1")),
# enable custom dpsk scaling rope
"ENABLE_CUSTOM_DPSK_SCALING_ROPE": lambda: (
os.environ.get("ENABLE_CUSTOM_DPSK_SCALING_ROPE", "False").lower()
in ("true", "1")
),
"ENABLE_CUSTOM_DPSK_SCALING_ROPE":
lambda: (os.environ.get("ENABLE_CUSTOM_DPSK_SCALING_ROPE", "False").lower() in
("true", "1")),
# fuse qkv split & qk norm & qk rope
# only works for qwen3 dense and qwen3 moe models
"ENABLE_VLLM_FUSED_QKV_SPLIT_NORM_ROPE": lambda: (
os.environ.get("ENABLE_VLLM_FUSED_QKV_SPLIT_NORM_ROPE", "False").lower()
in ("true", "1")
),
"ENABLE_VLLM_FUSED_QKV_SPLIT_NORM_ROPE":
lambda: (os.environ.get("ENABLE_VLLM_FUSED_QKV_SPLIT_NORM_ROPE", "False").lower() in
("true", "1")),
}
# end-env-vars-definition
def __getattr__(name: str):
"""
This function is called when an attribute that doesn't exist is accessed.
If the attribute is one of the xvllm_environment_variables, return the corresponding value.
Otherwise, raise an AttributeError.
当调用不存在的属性时，该函数被调用。如果属性是xvllm_environment_variables中的一个，则返回相应的值。否则引发AttributeError异常。
Args:
name (str): The name of the attribute to retrieve.
name (str): 要获取的属性名称。
Raises:
AttributeError (Exception): If the attribute is not one of xvllm_environment_variables, this exception is raised.
AttributeError (Exception): 如果属性不是xvllm_environment_variables中的一个，则会引发此异常。
Returns:
Any, optional: If the attribute is one of xvllm_environment_variables, the corresponding value is returned; otherwise, None is returned.
Any, optional: 如果属性是xvllm_environment_variables中的一个，则返回相应的值；否则返回None。
"""
# lazy evaluation of environment variables
if name in xvllm_environment_variables:
@@ -92,14 +93,13 @@ def __getattr__(name: str):
def __dir__():
"""
Returns a list of all visible variable names.
返回一个包含所有可见的变量名称的列表。
返回值（list）：一个包含所有可见的变量名称的列表，这些变量是通过`xvllm_environment_variables`字典定义的。
Returns:
list: A list of all visible variable names, which are defined through the `xvllm_environment_variables` dictionary.
Returns:
List[str]: A list of all visible variable names.
These variables are defined through the `xvllm_environment_variables` dictionary.
List[str]: 一个包含所有可见的变量名称的列表。
这些变量是通过`xvllm_environment_variables`字典定义的。
"""
return list(xvllm_environment_variables.keys())