minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

Fix DSA partial DP-TP mode log to use derived attn_tp_size (#39871)

Browse Source 
Co-authored-by: Hun-ho Kim <hunho.kim@samsung.com>
This commit is contained in:
hunhokim
2026-09-18 15:54:45 -07:00
committed by GitHub
co-authored by Hun-ho Kim
parent f3851486cb
commit 803f0c93d2
1 changed files with 8 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/arg_groups/model_hook.py
+8 -2
View File
@@ -37,7 +37,7 @@ from sglang.srt.connector import ConnectorType
from sglang.srt.environ import envs from sglang.srt.environ import envs
from sglang.srt.hardware_backend.mlx.runtime import use_mlx from sglang.srt.hardware_backend.mlx.runtime import use_mlx
from sglang.srt.model_executor.cuda_graph_config import Backend, Phase, with_phase from sglang.srt.model_executor.cuda_graph_config import Backend, Phase, with_phase
from sglang.srt.runtime_context import get_platform from sglang.srt.runtime_context import derive_attention_widths, get_platform
from sglang.srt.utils.common import ( from sglang.srt.utils.common import (
get_quantization_config, get_quantization_config,
is_mps, is_mps,
@@ -285,9 +285,15 @@ def handle_model_specific_adjustments(server_args: Any):
else: else:
# Pure TP and partial DP Attention mode is active for DSA, logging a warning # Pure TP and partial DP Attention mode is active for DSA, logging a warning
if cfg.dp_size < cfg.tp_size: if cfg.dp_size < cfg.tp_size:
_, attn_tp_size = derive_attention_widths(
tp_size=cfg.tp_size,
attn_cp_size=cfg.attn_cp_size,
dp_size=cfg.dp_size,
enable_dp_attention=cfg.enable_dp_attention,
)
logger.warning( logger.warning(
f"DSA with TP mode is active, dp_size={cfg.dp_size}, tp_size={cfg.tp_size}, " f"DSA with TP mode is active, dp_size={cfg.dp_size}, tp_size={cfg.tp_size}, "
f"attn_tp_size={cfg.tp_size}, attention weights will be sharded across {cfg.tp_size} ranks." f"attn_tp_size={attn_tp_size}, attention weights will be sharded across {attn_tp_size} ranks."
) )
# The DSA page-size selection moved to the override registry # The DSA page-size selection moved to the override registry