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

Support data parallelism (static) (#480)

Browse Source 
Co-authored-by: Ying Sheng <ying.sheng@databricks.com>
Co-authored-by: Lianmin Zheng <lianminzheng@gmail.com>
Co-authored-by: Liangsheng Yin <hnyls2002@gmail.com>
Co-authored-by: Zhiqiang Xie <xiezhq@stanford.edu>
This commit is contained in:
Ying Sheng
2024-05-27 21:24:10 -07:00
committed by GitHub
parent 565d727409
commit 0463f7fb52
32 changed files with 580 additions and 181 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
python/sglang/srt/layers/radix_attention.py
View File

@@ -5,7 +5,7 @@ from torch import nn
from sglang.srt.layers.context_flashattention_nopad import context_attention_fwd
from sglang.srt.layers.extend_attention import extend_attention_fwd
from sglang.srt.layers.token_attention import token_attention_fwd
from sglang.srt.managers.router.model_runner import ForwardMode, InputMetadata
from sglang.srt.managers.controller.model_runner import ForwardMode, InputMetadata
class RadixAttention(nn.Module):
@@ -20,7 +20,7 @@ class RadixAttention(nn.Module):
assert np.allclose(scaling, 1.0 / (head_dim**0.5))
from sglang.srt.managers.router.model_runner import global_server_args_dict
from sglang.srt.managers.controller.model_runner import global_server_args_dict
if global_server_args_dict.get("enable_flashinfer", False):
self.prefill_forward = self.prefill_forward_flashinfer