chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[DP-Attn] Clarify MLP sync / idle batch preparation logic (#12843)

Browse Source
This commit is contained in:
Liangsheng Yin
2025-11-08 23:23:14 +08:00
committed by GitHub
parent 6fee2c535c
commit 243ea585fc
7 changed files with 229 additions and 177 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
python/sglang/bench_one_batch.py
View File

@@ -68,7 +68,7 @@ from sglang.srt.distributed.parallel_state import destroy_distributed_environmen
from sglang.srt.entrypoints.engine import _set_envs_and_config
from sglang.srt.layers.moe import initialize_moe_config
from sglang.srt.managers.schedule_batch import Req, ScheduleBatch
from sglang.srt.managers.scheduler import Scheduler
from sglang.srt.managers.scheduler_dp_attn_mixin import prepare_mlp_sync_batch_raw
from sglang.srt.model_executor.forward_batch_info import ForwardBatch
from sglang.srt.model_executor.model_runner import ModelRunner
from sglang.srt.sampling.sampling_params import SamplingParams
@@ -391,15 +391,13 @@ def decode(input_token_ids, batch, model_runner):
def _maybe_prepare_mlp_sync_batch(batch: ScheduleBatch, model_runner):
if require_mlp_sync(model_runner.server_args):
Scheduler.prepare_mlp_sync_batch_raw(
prepare_mlp_sync_batch_raw(
batch,
dp_size=model_runner.server_args.dp_size,
attn_tp_size=1,
tp_group=model_runner.tp_group,
get_idle_batch=None,
disable_cuda_graph=model_runner.server_args.disable_cuda_graph,
spec_algorithm=SpeculativeAlgorithm.NONE,
speculative_num_draft_tokens=None,
require_mlp_tp_gather=require_mlp_tp_gather(model_runner.server_args),
disable_overlap_schedule=model_runner.server_args.disable_overlap_schedule,
offload_tags=set(),