elastic_ep: Fix issues with repeated scale up/down cycles (#37131)
Signed-off-by: Itay Alroy <ialroy@nvidia.com> Co-authored-by: Ron Tourgeman <rtourgeman@nvidia.com>
This commit is contained in:
co-authored by
Ron Tourgeman
parent
e5ed6c6c13
commit
c57d38d603
@@ -382,9 +382,10 @@ class RayDistributedExecutor(Executor):
|
||||
all_kwargs.append(kwargs)
|
||||
self.collective_rpc("init_worker", args=(all_kwargs,))
|
||||
|
||||
is_eep_new_worker = envs.VLLM_ELASTIC_EP_SCALE_UP_LAUNCH
|
||||
if not is_eep_new_worker:
|
||||
self.collective_rpc("init_device")
|
||||
self.collective_rpc("init_device")
|
||||
if envs.VLLM_ELASTIC_EP_SCALE_UP_LAUNCH:
|
||||
self.collective_rpc("elastic_ep_execute", args=("load_model",))
|
||||
else:
|
||||
self.collective_rpc("load_model")
|
||||
|
||||
def _update_block_size(worker):
|
||||
|
||||
Reference in New Issue
Block a user