Skip to content

[NPU][diffusion]: support parallel decoding of qwen-image #20757

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
ping1jing2 merged 17 commits into from
Mar 30, 2026
Merged

[NPU][diffusion]: support parallel decoding of qwen-image #20757

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
17 commits
Select commit Hold shift + click to select a range
1218851
[diffusion]feat: support parallel decoding of qwen-image
Mar 17, 2026
56b0661
Merge branch 'main' into vae-decoding-parallel
Mar 17, 2026
cdb732c
docs: complete the annotations
Mar 17, 2026
ec70b1a
Merge branch 'sgl-project:main' into vae-decoding-parallel
gxxx-hum Mar 17, 2026
1ce3c42
[diffusion]fix: call parallel_tiled_decode via super()
Mar 18, 2026
5b6959f
[Diffusion][QwenImageVAE] Unify decode return type in tiled path
Mar 19, 2026
125c79b
[Diffusion][VAE] Optimize parallel tiled decode
Mar 19, 2026
d37bae0
fix lint and avoid getter
Mar 23, 2026
31d6530
feat: dynamic micro-batch for parallel tiled vae decode
Mar 24, 2026
b4f230b
Move parallel tiled decode changes from base VAE to QwenImage
Mar 25, 2026
6ef1034
restore vae common
Mar 25, 2026
f7fd627
refactor: move parallel_tiled_deocde into base
Mar 29, 2026
2841885
Merge branch 'sgl-project:main' into vae-decoding-parallel
gxxx-hum Mar 29, 2026
d3a6b35
handle ci and review
Mar 29, 2026
488392d
Merge branch 'sgl-project:main' into vae-decoding-parallel
gxxx-hum Mar 30, 2026
8a6eb6d
refactor: abstract parallel tiled decode gather/merge flow
Mar 30, 2026
bcf8973
Merge branch 'vae-decoding-parallel' of https://github.com/dazedsanji…
Mar 30, 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
2 changes: 2 additions & 0 deletions python/sglang/multimodal_gen/configs/models/vaes/qwenimage.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,6 +38,8 @@ class QwenImageVAEConfig(VAEConfig):
use_temporal_tiling: bool = False
use_parallel_tiling: bool = False

use_parallel_decode: bool = False

def get_vae_scale_factor(self):
return 2 ** len(self.arch_config.temperal_downsample)

Expand Down
137 changes: 84 additions & 53 deletions python/sglang/multimodal_gen/runtime/models/dits/qwen_image.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,7 +47,10 @@
apply_flashinfer_rope_qk_inplace,
)
from sglang.multimodal_gen.runtime.models.dits.base import CachableDiT
from sglang.multimodal_gen.runtime.platforms import AttentionBackendEnum
from sglang.multimodal_gen.runtime.platforms import (
AttentionBackendEnum,
current_platform,
)
from sglang.multimodal_gen.runtime.utils.layerwise_offload import OffloadableDiTMixin
from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger

Expand Down Expand Up @@ -826,30 +829,52 @@ def _modulate(

shift, scale, gate = mod_params.chunk(3, dim=-1)
if index is not None:
actual_batch = x.shape[0]
shift0, shift1 = (
shift[:actual_batch],
shift[actual_batch : 2 * actual_batch],
)
scale0, scale1 = (
scale[:actual_batch],
scale[actual_batch : 2 * actual_batch],
)
gate0, gate1 = gate[:actual_batch], gate[actual_batch : 2 * actual_batch]
if not x.is_contiguous():
x = x.contiguous()
if not index.is_contiguous():
index = index.contiguous()
if is_scale_residual:
if not residual_x.is_contiguous():
residual_x = residual_x.contiguous()
if not gate_x.is_contiguous():
gate_x = gate_x.contiguous()
x, residual_out, gate_result = (
fuse_residual_layernorm_scale_shift_gate_select01_kernel(
# ROCm currently fails to compile the select01 Triton kernel, so
# keep using the torch.where fallback there.
if x.is_cuda and not current_platform.is_hip():
Copy link
Copy Markdown
Collaborator

@mickqian mickqian Mar 31, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

if x.is_cuda and not current_platform.is_hip(): this pattern is appearing repeatedly recently, please move the implementation to platform.py, including all the changes recently

Copy link
Copy Markdown
Contributor Author

@gxxx-hum gxxx-hum Mar 31, 2026 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I’ll move it under runtime/platforms and clean up the recent similar cases there as well.

actual_batch = x.shape[0]
shift0, shift1 = (
shift[:actual_batch],
shift[actual_batch : 2 * actual_batch],
)
scale0, scale1 = (
scale[:actual_batch],
scale[actual_batch : 2 * actual_batch],
)
gate0, gate1 = (
gate[:actual_batch],
gate[actual_batch : 2 * actual_batch],
)
if not x.is_contiguous():
x = x.contiguous()
if not index.is_contiguous():
index = index.contiguous()
if is_scale_residual:
if not residual_x.is_contiguous():
residual_x = residual_x.contiguous()
if not gate_x.is_contiguous():
gate_x = gate_x.contiguous()
x, residual_out, gate_result = (
fuse_residual_layernorm_scale_shift_gate_select01_kernel(
x,
residual=residual_x,
residual_gate=gate_x,
weight=getattr(norm_module.norm, "weight", None),
bias=getattr(norm_module.norm, "bias", None),
scale0=scale0.contiguous(),
shift0=shift0.contiguous(),
gate0=gate0.contiguous(),
scale1=scale1.contiguous(),
shift1=shift1.contiguous(),
gate1=gate1.contiguous(),
index=index,
eps=norm_module.eps,
)
)
return x, residual_out, gate_result
else:
x, gate_result = fuse_layernorm_scale_shift_gate_select01_kernel(
x,
residual=residual_x,
residual_gate=gate_x,
weight=getattr(norm_module.norm, "weight", None),
bias=getattr(norm_module.norm, "bias", None),
scale0=scale0.contiguous(),
Expand All @@ -861,39 +886,45 @@ def _modulate(
index=index,
eps=norm_module.eps,
)
)
return x, residual_out, gate_result
return x, gate_result
else:
x, gate_result = fuse_layernorm_scale_shift_gate_select01_kernel(
x,
weight=getattr(norm_module.norm, "weight", None),
bias=getattr(norm_module.norm, "bias", None),
scale0=scale0.contiguous(),
shift0=shift0.contiguous(),
gate0=gate0.contiguous(),
scale1=scale1.contiguous(),
shift1=shift1.contiguous(),
gate1=gate1.contiguous(),
index=index,
eps=norm_module.eps,
actual_batch = x.shape[0]
shift0, shift1 = (
shift[:actual_batch],
shift[actual_batch : 2 * actual_batch],
)
return x, gate_result
scale0, scale1 = (
scale[:actual_batch],
scale[actual_batch : 2 * actual_batch],
)
gate0, gate1 = (
gate[:actual_batch],
gate[actual_batch : 2 * actual_batch],
)
index = index.to(dtype=torch.bool).unsqueeze(-1)
shift_result = torch.where(
index, shift1.unsqueeze(1), shift0.unsqueeze(1)
)
scale_result = torch.where(
index, scale1.unsqueeze(1), scale0.unsqueeze(1)
)
gate_result = torch.where(index, gate1.unsqueeze(1), gate0.unsqueeze(1))
else:
shift_result = shift.unsqueeze(1)
scale_result = scale.unsqueeze(1)
gate_result = gate.unsqueeze(1)
if is_scale_residual:
modulated, residual_out = norm_module(
residual=residual_x,
x=x,
gate=gate_x,
shift=shift_result,
scale=scale_result,
)
return modulated, residual_out, gate_result
else:
modulated = norm_module(x=x, shift=shift_result, scale=scale_result)
return modulated, gate_result
if is_scale_residual:
modulated, residual_out = norm_module(
residual=residual_x,
x=x,
gate=gate_x,
shift=shift_result,
scale=scale_result,
)
return modulated, residual_out, gate_result
else:
modulated = norm_module(x=x, shift=shift_result, scale=scale_result)
return modulated, gate_result

def forward(
self,
Expand Down Expand Up @@ -1127,8 +1158,8 @@ def build_modulate_index(self, img_shapes: tuple[int, int, int], device):
first_size = sample[0][0] * sample[0][1] * sample[0][2]
total_size = sum(s[0] * s[1] * s[2] for s in sample)
if sp_world_size > 1:
first_local_size = _local_seq_len(first_size)
tail_local_size = _local_seq_len(total_size - first_size)
first_local_size = _local_seq_len(first_size, sp_world_size)
tail_local_size = _local_seq_len(total_size - first_size, sp_world_size)
idx = torch.cat(
[
torch.zeros(first_local_size, device=device, dtype=torch.int),
Expand Down
Loading
Loading