Skip to content

[NPU] use causal_conv1d_update_v2 for performance #24595

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
sglang-npu-bot merged 3 commits into from
May 12, 2026
Merged

[NPU] use causal_conv1d_update_v2 for performance #24595

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
80a3287
use causal_conv1d_update_v2 instead of torch.ops.npu.causal_conv1d_up…
iridiumine May 7, 2026
3f42d4a
Merge branch 'main' into causal_conv1d_update_v2
iridiumine May 11, 2026
a80676e
Merge branch 'main' into causal_conv1d_update_v2
iridiumine May 12, 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
28 changes: 14 additions & 14 deletions python/sglang/srt/hardware_backend/npu/attention/ascend_gdn_backend.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,6 +8,7 @@
from sgl_kernel_npu.mamba.causal_conv1d import (
causal_conv1d_fn_npu,
causal_conv1d_update_npu,
causal_conv1d_update_v2,
)

from sglang.srt.hardware_backend.npu.attention.ascend_hybrid_linear_attn_backend import (
Expand Down Expand Up @@ -224,9 +225,7 @@ def forward_extend(
else:
has_initial_states = forward_batch.extend_prefix_lens > 0
if is_target_verify:
draft_token_num = forward_batch.spec_info.draft_token_num
num_token_padding = mixed_qkv.shape[0]
batch_size = cache_indices.shape[0]
if (
not self.graph_mode
and forward_batch.num_token_non_padded_cpu != num_token_padding
Expand All @@ -236,23 +235,24 @@ def forward_extend(
b = b[: forward_batch.num_token_non_padded_cpu]
seq_len = forward_batch.num_token_non_padded_cpu

mixed_qkv_reshaped = mixed_qkv.view(batch_size, draft_token_num, -1)
num_accept_tokens = torch.full(
batch_size = cache_indices.shape[0]
draft_token_num = forward_batch.spec_info.draft_token_num
num_accepted_tokens = torch.full(
(batch_size,),
draft_token_num,
dtype=torch.int32,
device=mixed_qkv.device,
)
mixed_qkv = torch.ops.npu.causal_conv1d_update(
mixed_qkv_reshaped,
layer.conv_weights.transpose(0, 1).contiguous(),
conv_states,
cache_indices,
layer.bias,
num_accept_tokens,
None,
layer.activation == "silu",
self.pad_slot_id,
mixed_qkv = causal_conv1d_update_v2(
x=mixed_qkv.view(batch_size, draft_token_num, -1).contiguous(),
conv_state=conv_states.contiguous(),
Comment thread
iridiumine marked this conversation as resolved.
weight=layer.conv_weights.transpose(0, 1).contiguous(),
bias=layer.bias,
activation=layer.activation,
conv_state_indices=cache_indices,
num_accepted_tokens=num_accepted_tokens,
pad_slot_id=-1,
Comment thread
iridiumine marked this conversation as resolved.
validate_data=False,
).view(seq_len, -1)
Comment thread
iridiumine marked this conversation as resolved.
else:
mixed_qkv = mixed_qkv.transpose(0, 1)
Expand Down
Loading