EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

[Feature] Integrate DeepEP into SGLang (#4232)

Browse Source 
Co-authored-by: Cheng Wan <cwan39@gatech.edu>
Co-authored-by: Xuting Zhou <xutingz@nvidia.com>
This commit is contained in:
Jinyan Chen
2025-03-19 23:16:31 +08:00
committed by GitHub
parent f9c53cbb42
commit f44db16c8e
12 changed files with 1228 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
python/sglang/srt/model_executor/model_runner.py
View File

@@ -145,6 +145,7 @@ class ModelRunner:
"enable_nan_detection": server_args.enable_nan_detection,
"enable_dp_attention": server_args.enable_dp_attention,
"enable_ep_moe": server_args.enable_ep_moe,
"enable_deepep_moe": server_args.enable_deepep_moe,
"device": server_args.device,
"speculative_accept_threshold_single": server_args.speculative_accept_threshold_single,
"speculative_accept_threshold_acc": server_args.speculative_accept_threshold_acc,
@@ -277,6 +278,12 @@ class ModelRunner:
server_args.chunked_prefill_size = -1
server_args.disable_radix_cache = True
if server_args.enable_deepep_moe:
logger.info("DeepEP is turned on.")
assert (
server_args.enable_dp_attention == True
), "Currently DeepEP is bind to Attention DP. Set '--enable-dp-attention --enable-deepep-moe'"
def init_torch_distributed(self):
logger.info("Init torch distributed begin.")