biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Deprecation] Remove deprecated plugin and compilation fields for v0.13 release (#30396)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-12-11 11:59:35 +08:00
committed by GitHub
parent d1e1fb4363
commit 5a87d8b9b1
10 changed files with 22 additions and 238 deletions
Download Patch File Download Diff File Expand all files Collapse all files

81
vllm/config/compilation.py
View File

@@ -17,7 +17,6 @@ from vllm.config.utils import (
Range,
config,
get_hash_factors,
handle_deprecated,
hash_factors,
)
from vllm.logger import init_logger
@@ -127,27 +126,6 @@ class PassConfig:
fuse_allreduce_rms: bool = Field(default=None)
"""Enable flashinfer allreduce fusion."""
# Deprecated flags
enable_fusion: bool = Field(default=None)
"""Deprecated in: v0.12.0. Use fuse_norm_quant and fuse_act_quant
instead. Will be removed in v0.13.0 or v1.0.0, whichever is sooner.
"""
enable_attn_fusion: bool = Field(default=None)
"""Deprecated in: v0.12.0. Use fuse_attn_quant instead.
Will be removed in v0.13.0 or v1.0.0, whichever is sooner."""
enable_noop: bool = Field(default=None)
"""Deprecated in: v0.12.0. Use eliminate_noops instead.
Will be removed in v0.13.0 or v1.0.0, whichever is sooner."""
enable_sequence_parallelism: bool = Field(default=None)
"""Deprecated in: v0.12.0. Use enable_sp instead.
Will be removed in v0.13.0 or v1.0.0, whichever is sooner."""
enable_async_tp: bool = Field(default=None)
"""Deprecated in: v0.12.0. Use fuse_gemm_comms instead.
Will be removed in v0.13.0 or v1.0.0, whichever is sooner."""
enable_fi_allreduce_fusion: bool = Field(default=None)
"""Deprecated in: v0.12.0. Use fuse_allreduce_rms instead.
Will be removed in v0.13.0 or v1.0.0, whichever is sooner."""
fi_allreduce_fusion_max_size_mb: float | None = None
"""The threshold of the communicated tensor sizes under which
vllm should use flashinfer fused allreduce. Specified as a
@@ -206,15 +184,7 @@ class PassConfig:
Any future fields that don't affect compilation should be excluded.
"""
ignored_fields = [
"enable_fusion",
"enable_attn_fusion",
"enable_noop",
"enable_sequence_parallelism",
"enable_async_tp",
"enable_fi_allreduce_fusion",
]
return hash_factors(get_hash_factors(self, ignored_factors=ignored_fields))
return hash_factors(get_hash_factors(self, set()))
@field_validator(
"fuse_norm_quant",
@@ -224,12 +194,6 @@ class PassConfig:
"enable_sp",
"fuse_gemm_comms",
"fuse_allreduce_rms",
"enable_fusion",
"enable_attn_fusion",
"enable_noop",
"enable_sequence_parallelism",
"enable_async_tp",
"enable_fi_allreduce_fusion",
mode="wrap",
)
@classmethod
@@ -242,49 +206,6 @@ class PassConfig:
def __post_init__(self) -> None:
# Handle deprecation and defaults
# Map old flags to new flags and issue warnings
handle_deprecated(
self,
"enable_fusion",
["fuse_norm_quant", "fuse_act_quant"],
"v0.13.0 or v1.0.0, whichever is sooner",
)
handle_deprecated(
self,
"enable_attn_fusion",
"fuse_attn_quant",
"v0.13.0 or v1.0.0, whichever is sooner",
)
handle_deprecated(
self,
"enable_sequence_parallelism",
"enable_sp",
"v0.13.0 or v1.0.0, whichever is sooner",
)
handle_deprecated(
self,
"enable_async_tp",
"fuse_gemm_comms",
"v0.13.0 or v1.0.0, whichever is sooner",
)
handle_deprecated(
self,
"enable_fi_allreduce_fusion",
"fuse_allreduce_rms",
"v0.13.0 or v1.0.0, whichever is sooner",
)
handle_deprecated(
self,
"enable_noop",
"eliminate_noops",
"v0.13.0 or v1.0.0, whichever is sooner",
)
if not self.eliminate_noops:
if self.fuse_norm_quant or self.fuse_act_quant:
logger.warning_once(