[Minor]Raise Error when deepep num dispatch token per rank is smaller than cuda graph bs (#14065)
This commit is contained in:
@@ -23,6 +23,7 @@ from sglang.srt.layers.moe.utils import (
|
|||||||
get_moe_runner_backend,
|
get_moe_runner_backend,
|
||||||
is_tbo_enabled,
|
is_tbo_enabled,
|
||||||
)
|
)
|
||||||
|
from sglang.srt.server_args import get_global_server_args
|
||||||
from sglang.srt.utils import (
|
from sglang.srt.utils import (
|
||||||
get_bool_env_var,
|
get_bool_env_var,
|
||||||
get_int_env_var,
|
get_int_env_var,
|
||||||
@@ -311,6 +312,15 @@ class _DeepEPDispatcherImplBase:
|
|||||||
self.num_max_dispatch_tokens_per_rank = get_int_env_var(
|
self.num_max_dispatch_tokens_per_rank = get_int_env_var(
|
||||||
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK", 128
|
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK", 128
|
||||||
)
|
)
|
||||||
|
if (
|
||||||
|
not get_global_server_args().disable_cuda_graph
|
||||||
|
and get_global_server_args().cuda_graph_max_bs
|
||||||
|
> self.num_max_dispatch_tokens_per_rank
|
||||||
|
):
|
||||||
|
raise ValueError(
|
||||||
|
f"Maximum captured batch size {get_global_server_args().cuda_graph_max_bs} is larger than num_max_dispatch_tokens_per_rank for deepep {self.num_max_dispatch_tokens_per_rank}. Please set a smaller value for --cuda-graph-max-bs server argument, or increase SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK."
|
||||||
|
)
|
||||||
|
|
||||||
# DeepEP internode_ll dispatch uses FINISHED_SUM_TAG=1024
|
# DeepEP internode_ll dispatch uses FINISHED_SUM_TAG=1024
|
||||||
# and the logic requires num-tokens-sent-from-one-rank-to-another-rank less than it
|
# and the logic requires num-tokens-sent-from-one-rank-to-another-rank less than it
|
||||||
assert self.num_max_dispatch_tokens_per_rank <= 1024
|
assert self.num_max_dispatch_tokens_per_rank <= 1024
|
||||||
|
|||||||
Reference in New Issue
Block a user