Skip to content

[NVIDIA] Relax previous restraints for using flashinfer gdn decode #22818

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

Closed
kaixih wants to merge 1 commit into from
+24 −25
Closed

[NVIDIA] Relax previous restraints for using flashinfer gdn decode #22818

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
13 changes: 0 additions & 13 deletions python/sglang/srt/server_args.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2188,19 +2188,6 @@ 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
36 changes: 24 additions & 12 deletions test/registered/4-gpu-models/test_qwen35_models.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,15 +29,11 @@

class TestQwen35FP4(CustomTestCase):
def test_gsm8k(self):
base_args = [
common_args = [
"--tp-size",
"4",
"--chunked-prefill-size",
"2048",
"--mamba-scheduler-strategy",
"extra_buffer",
"--mamba-track-interval",
"128",
"--mamba-ssm-dtype",
"bfloat16",
"--max-running-requests",
Expand All @@ -51,19 +47,35 @@ def test_gsm8k(self):
"--model-loader-extra-config",
'{"enable_multithread_load": true,"num_threads": 64}',
]
extra_buffer_args = common_args + [
"--mamba-scheduler-strategy",
"extra_buffer",
"--mamba-track-interval",
"128",
]
no_buffer_args = common_args + [
"--mamba-scheduler-strategy",
"no_buffer",
]

variants = [
ModelLaunchSettings(
QWEN35_FP4_MODEL,
extra_args=base_args,
extra_args=extra_buffer_args,
variant="Triton",
),
# TODO: Fix this and re-enable it
# ModelLaunchSettings(
# QWEN35_FP4_MODEL,
# extra_args=base_args + ["--linear-attn-decode-backend", "flashinfer"],
# variant="FlashInfer",
# ),
ModelLaunchSettings(
QWEN35_FP4_MODEL,
extra_args=extra_buffer_args
+ ["--linear-attn-decode-backend", "flashinfer"],
variant="FlashInfer",
),
ModelLaunchSettings(
QWEN35_FP4_MODEL,
extra_args=no_buffer_args
+ ["--linear-attn-decode-backend", "flashinfer"],
variant="FlashInfer-NoBuffer",
),
]

run_combined_tests(
Expand Down
Loading