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

Support sequence parallelism combined with pipeline parallelism (#18243)

Browse Source 
Signed-off-by: cascade812 <cascade812@outlook.com>
This commit is contained in:
cascade
2025-05-17 15:47:25 -07:00
committed by GitHub
parent 66e63e86ec
commit 9ab2c02ff8
3 changed files with 74 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
vllm/config.py
View File

@@ -4287,18 +4287,6 @@ class VllmConfig:
self.compilation_config.level = CompilationLevel.PIECEWISE
self.compilation_config.set_splitting_ops_for_v1()
if self.parallel_config is not None and \
self.parallel_config.tensor_parallel_size > 1 and \
self.parallel_config.pipeline_parallel_size > 1 and \
self.compilation_config is not None and \
self.compilation_config.pass_config is not None and \
self.compilation_config.pass_config.enable_sequence_parallelism:
logger.warning_once(
"Sequence parallelism is not supported with pipeline "
"parallelism. Disabling sequence parallelism.")
self.compilation_config.pass_config.\
enable_sequence_parallelism = False
self._set_cudagraph_sizes()
if self.cache_config is not None and \