Skip to content

GitLab

"src/vscode:/vscode.git/clone" did not exist on "d63e6fccb11f1ec6e47940efb565bd8aeeda18b8"
Unverified Commit 61ce91ed authored by fzyzcjy's avatar fzyzcjy Committed by GitHub
Browse files

Tiny support customize DeepEP max dispatch tokens per rank (#6934)

parent e6b7053b
Hide whitespace changes
Inline Side-by-side
Showing with 4 additions and 2 deletions
python/sglang/srt/layers/moe/ep_moe/token_dispatcher.py
View file @ 61ce91ed
...@@ -6,7 +6,7 @@ from sglang.srt.managers.expert_distribution import ( ...@@ -6,7 +6,7 @@ from sglang.srt.managers.expert_distribution import (
get_global_expert_distribution_recorder, get_global_expert_distribution_recorder,
) )
from sglang.srt.managers.schedule_batch import global_server_args_dict from sglang.srt.managers.schedule_batch import global_server_args_dict
from sglang.srt.utils import DeepEPMode, load_json_config from sglang.srt.utils import DeepEPMode, get_int_env_var, load_json_config
try: try:
from deep_ep import Buffer, Config from deep_ep import Buffer, Config
...@@ -188,7 +188,9 @@ class _DeepEPDispatcherImplBase: ...@@ -188,7 +188,9 @@ class _DeepEPDispatcherImplBase:
self.deepep_mode = deepep_mode self.deepep_mode = deepep_mode
self.params_bytes = 2 self.params_bytes = 2
self.num_max_dispatch_tokens_per_rank = 128 self.num_max_dispatch_tokens_per_rank = get_int_env_var(
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK", 128
)
self.handle = None self.handle = None
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment