Skip to content

[GDN] Remove FlashInfer GDN decode + no_buffer guard and default to FlashInfer on SM100+ #21861

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
YAMY1234 wants to merge 3 commits into
base: main
Choose a base branch
from
+19 −14
Open

[GDN] Remove FlashInfer GDN decode + no_buffer guard and default to FlashInfer on SM100+ #21861

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
f9a4e2c
Remove FlashInfer GDN decode + no_buffer incompatibility guard
YAMY1234 Apr 1, 2026
c92b80d
Default to FlashInfer GDN decode on SM100+ with bf16 mamba state
YAMY1234 Apr 1, 2026
8a868f6
Merge branch 'main' into remove-gdn-no-buffer-guard
YAMY1234 Apr 2, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
33 changes: 19 additions & 14 deletions python/sglang/srt/server_args.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2145,19 +2145,6 @@ def _handle_mamba_radix_cache(
not self.enable_mamba_extra_buffer()
), f"mamba extra_buffer is not supported for {model_arch} model"

# FlashInfer GDN decode is incompatible with no_buffer scheduling.
# See https://github.com/sgl-project/sglang/issues/20791
if (
self.linear_attn_decode_backend == "flashinfer"
and self.mamba_scheduler_strategy == "no_buffer"
):
raise ValueError(
"FlashInfer GDN decode (--linear-attn-decode-backend flashinfer) is not "
"compatible with --mamba-scheduler-strategy no_buffer. "
"Please use --mamba-scheduler-strategy extra_buffer instead. "
"See https://github.com/sgl-project/sglang/issues/20791"
)

if self.enable_mamba_extra_buffer(): # extra_buffer
if self.disable_radix_cache:
raise ValueError(
Expand Down Expand Up @@ -2577,9 +2564,27 @@ def _handle_mamba_backend(self):
)

def _handle_linear_attn_backend(self):
# SM100+ FlashInfer GDN decode requires bf16 state; SM90 uses float32.
import torch

# SM100+: default to FlashInfer GDN decode when the user hasn't
# explicitly chosen a decode backend and mamba-ssm-dtype is bf16
# (required by FlashInfer GDN on SM100+).
# Fixed in FlashInfer v0.6.7: flashinfer-ai/flashinfer#2810
# Excluded when MTP speculative decoding is enabled because
# FlashInfer GDN MTP verify is not yet supported on SM100+.
if (
self.linear_attn_decode_backend is None
and is_sm100_supported()
and self.mamba_ssm_dtype == "bfloat16"
and self.speculative_algorithm is None
):
self.linear_attn_decode_backend = "flashinfer"
logger.info(
"SM100+ detected with mamba-ssm-dtype=bfloat16, "
"defaulting --linear-attn-decode-backend to flashinfer."
)

# SM100+ FlashInfer GDN decode requires bf16 state; SM90 uses float32.
decode = self.linear_attn_decode_backend or self.linear_attn_backend
if (
decode == "flashinfer"
Expand Down
Loading