Skip to content

fix: allow fmha_v2_prefill_deepseek on SM121 (DGX Spark) #2559

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

Merged
jimmyzho merged 1 commit into from
Feb 19, 2026
Merged

fix: allow fmha_v2_prefill_deepseek on SM121 (DGX Spark) #2559

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
4 changes: 2 additions & 2 deletions flashinfer/prefill.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3821,8 +3821,8 @@ def fmha_v2_prefill_deepseek(
If return_lse is True, the output will be a tuple of two tensors, the first is the output tensor, the second is the lse tensor.
If return_lse is False, the output will be a single tensor.
"""
if not is_sm120a_supported(query.device):
raise ValueError("fmha_v2_prefill_deepseek is only supported on SM120 GPUs.")
if not (is_sm120a_supported(query.device) or is_sm121a_supported(query.device)):
raise ValueError("fmha_v2_prefill_deepseek is only supported on SM12x GPUs.")
assert query.shape[3] == 192 and key.shape[3] == 192 and value.shape[3] == 128, (
"currently only support deepseek r1 192 query and 128 value"
)
Expand Down
9 changes: 6 additions & 3 deletions tests/attention/test_fmha_v2_prefill_deepseek.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,7 @@

from flashinfer.prefill import fmha_v2_prefill_deepseek
from tests.utils_fp8 import to_float8
from flashinfer.utils import is_sm120a_supported
from flashinfer.utils import is_sm120a_supported, is_sm121a_supported


def attention_ref(
Expand Down Expand Up @@ -57,8 +57,11 @@ def attention_ref(
def test_fmha_v2_prefill_deepseek(
batch_size, num_heads, head_dim_qk, head_dim_v, seq_len, qkv_dtype, o_dtype
):
if not is_sm120a_supported(torch.device("cuda")):
pytest.skip("fmha_v2_prefill_deepseek is only supported on SM120 GPUs.")
if not (
is_sm120a_supported(torch.device("cuda"))
or is_sm121a_supported(torch.device("cuda"))
):
pytest.skip("fmha_v2_prefill_deepseek is only supported on SM12x GPUs.")
torch.manual_seed(42)

def initialize_tensors(batch_size, num_heads, head_dim_qk, head_dim_v, seq_len):
Expand Down
Loading