Upgrade to vllm 0.17.0 corex v4.1 overlay

This commit is contained in:
2026-04-29 19:38:22 +08:00
parent 8fac6062e4
commit 938d0854a5
430 changed files with 35969 additions and 14511 deletions

View File

@@ -87,8 +87,12 @@ class WorkerBase:
"""Get specifications for KV cache implementation."""
raise NotImplementedError
def compile_or_warm_up_model(self) -> None:
"""Prepare model for execution through compilation/warmup."""
def compile_or_warm_up_model(self) -> float:
"""Prepare model for execution through compilation/warmup.
Returns:
The accumulated compilation time in seconds.
"""
raise NotImplementedError
def check_health(self) -> None:
@@ -213,13 +217,8 @@ class WorkerWrapperBase:
It is only used during the initialization of the executor,
to adjust the rpc_rank of workers after we create all workers.
"""
# if self.rpc_rank in rank_mapping:
# self.rpc_rank = rank_mapping[self.rpc_rank]
old_rank = self.rpc_rank
if old_rank in rank_mapping:
self.rpc_rank = rank_mapping[old_rank]
if self.global_rank == old_rank:
self.global_rank = rank_mapping[old_rank]
if self.rpc_rank in rank_mapping:
self.rpc_rank = rank_mapping[self.rpc_rank]
def update_environment_variables(
self,