xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 00:45:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[V1] [Hybrid] Remove code to override default CUDA graph configuration (#26226)

Browse Source 
Signed-off-by: Thomas Parnell <tpa@zurich.ibm.com>
This commit is contained in:
Thomas Parnell 2025-10-04 15:47:48 +02:00 committed by GitHub
parent 86ee949128
commit ed3aeb25a4
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 0 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
vllm/model_executor/models/config.py
View File

@ -4,7 +4,6 @@ from copy import deepcopy
from typing import TYPE_CHECKING from typing import TYPE_CHECKING
import vllm.envs as envs import vllm.envs as envs
from vllm.config.compilation import CUDAGraphMode
from vllm.logger import init_logger from vllm.logger import init_logger
from vllm.model_executor.models import ModelRegistry from vllm.model_executor.models import ModelRegistry
from vllm.utils import STR_DTYPE_TO_TORCH_DTYPE, cdiv from vllm.utils import STR_DTYPE_TO_TORCH_DTYPE, cdiv
@ -290,7 +289,6 @@ class MambaModelConfig(VerifyAndUpdateConfig):
model_config = vllm_config.model_config model_config = vllm_config.model_config
cache_config = vllm_config.cache_config cache_config = vllm_config.cache_config
compilation_config = vllm_config.compilation_config
# Set mamba block size to max_model_len (this may get # Set mamba block size to max_model_len (this may get
# override by prefix caching logic later) # override by prefix caching logic later)
@ -320,19 +318,6 @@ class MambaModelConfig(VerifyAndUpdateConfig):
"for hybrid models.") "for hybrid models.")
model_config.disable_cascade_attn = True model_config.disable_cascade_attn = True
# TODO(tdoublep): remove as full cuda graph support is added
FCG_NOT_SUPPORTED_MODELS = [
"Lfm2ForCausalLM",
"MiniMaxText01ForCausalLM",
]
if (model_config.architecture not in FCG_NOT_SUPPORTED_MODELS
and compilation_config.cudagraph_mode is None):
logger.info(
"Hybrid or mamba-based model detected: setting cudagraph mode "
"to FULL_AND_PIECEWISE in order to optimize performance.")
compilation_config.cudagraph_mode = CUDAGraphMode.FULL_AND_PIECEWISE
class HybridAttentionMambaModelConfig(VerifyAndUpdateConfig): class HybridAttentionMambaModelConfig(VerifyAndUpdateConfig):