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

Fix DeepSeek DP Attention + torch compile (#5367)

Browse Source 
Co-authored-by: ispobock <ispobaoke@163.com>
This commit is contained in:
fzyzcjy
2025-04-14 16:07:58 +08:00
committed by GitHub
parent fc72871975
commit defede5073
3 changed files with 8 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
python/sglang/srt/layers/dp_attention.py
View File

@@ -192,8 +192,7 @@ def _dp_gather(
if local_tokens.shape[0] > 0 and (is_partial or get_attention_tp_rank() == 0): if local_tokens.shape[0] > 0 and (is_partial or get_attention_tp_rank() == 0):
assert ( assert (
global_tokens.untyped_storage().data_ptr() local_tokens.untyped_storage() is not global_tokens.untyped_storage()
!= local_tokens.untyped_storage().data_ptr()
), "aliasing between global_tokens and local_tokens not allowed" ), "aliasing between global_tokens and local_tokens not allowed"
memcpy_triton( memcpy_triton(
global_tokens, local_tokens, 0, local_start_pos, local_num_tokens, False global_tokens, local_tokens, 0, local_start_pos, local_num_tokens, False
@@ -243,8 +242,7 @@ def dp_scatter(
assert global_tokens.is_contiguous() assert global_tokens.is_contiguous()
if local_tokens.shape[0] > 0: if local_tokens.shape[0] > 0:
assert ( assert (
local_tokens.untyped_storage().data_ptr() local_tokens.untyped_storage() is not global_tokens.untyped_storage()
!= global_tokens.untyped_storage().data_ptr()
), "aliasing between local_tokens and global_tokens not allowed" ), "aliasing between local_tokens and global_tokens not allowed"
memcpy_triton( memcpy_triton(
local_tokens, global_tokens, 0, local_start_pos, local_num_tokens, True local_tokens, global_tokens, 0, local_start_pos, local_num_tokens, True

5
test/srt/parse_results.py
View File

@@ -1,7 +1,8 @@
import json
import pandas as pd
import argparse import argparse
import json
import os import os
import pandas as pd
from tabulate import tabulate from tabulate import tabulate
# Parse command-line arguments # Parse command-line arguments

3
test/srt/test_dp_attention.py
View File

@@ -28,6 +28,9 @@ class TestDPAttentionDP2TP2(CustomTestCase):
"--enable-dp-attention", "--enable-dp-attention",
"--dp", "--dp",
"2", "2",
"--enable-torch-compile",
"--torch-compile-max-bs",
"2",
], ],
) )