[FIX]remove ServerArgs duplicate code (#6485)
This commit is contained in:
@@ -324,12 +324,6 @@ class ServerArgs:
|
|||||||
if self.grammar_backend is None:
|
if self.grammar_backend is None:
|
||||||
self.grammar_backend = "xgrammar"
|
self.grammar_backend = "xgrammar"
|
||||||
|
|
||||||
if self.pp_size > 1:
|
|
||||||
self.disable_overlap_schedule = True
|
|
||||||
logger.warning(
|
|
||||||
"Overlap scheduler is disabled because of using pipeline parallelism."
|
|
||||||
)
|
|
||||||
|
|
||||||
# Data parallelism attention
|
# Data parallelism attention
|
||||||
if self.enable_dp_attention:
|
if self.enable_dp_attention:
|
||||||
self.schedule_conservativeness = self.schedule_conservativeness * 0.3
|
self.schedule_conservativeness = self.schedule_conservativeness * 0.3
|
||||||
@@ -1444,8 +1438,6 @@ class ServerArgs:
|
|||||||
|
|
||||||
# FIXME pp constraints
|
# FIXME pp constraints
|
||||||
if self.pp_size > 1:
|
if self.pp_size > 1:
|
||||||
logger.warning(f"Turn off overlap scheule for pipeline parallelism.")
|
|
||||||
self.disable_overlap_schedule = True
|
|
||||||
assert (
|
assert (
|
||||||
self.disable_overlap_schedule
|
self.disable_overlap_schedule
|
||||||
and self.speculative_algorithm is None
|
and self.speculative_algorithm is None
|
||||||
|
|||||||
Reference in New Issue
Block a user