Skip to content

[Bugfix] Enable PP with AITER+V1 #19822

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
merged 5 commits into from
Jun 20, 2025
Merged

[Bugfix] Enable PP with AITER+V1 #19822

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
56fcb9b
[Bugfix] Enable PP with AITER+V1
qli88 Jun 18, 2025
c0b01ea
pre-commit format changes
qli88 Jun 18, 2025
37777ee
yapf
qli88 Jun 19, 2025
987f7b7
Revert AITER RMSNORM for V1
qli88 Jun 19, 2025
0d0ef73
format updates.
qli88 Jun 19, 2025
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
1 change: 0 additions & 1 deletion vllm/model_executor/layers/layernorm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -45,7 +45,6 @@ def fused_add_rms_norm(

def rocm_aiter_rms_norm(x: torch.Tensor, weight: torch.Tensor,
variance_epsilon: float) -> torch.Tensor:

import aiter as rocm_aiter
if x.dim() > 2:
x_original_shape = x.shape
Expand Down
10 changes: 5 additions & 5 deletions vllm/model_executor/models/qwen2_5_omni_thinker.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -146,11 +146,11 @@ def get_hf_processor(
kwargs["fps"] = fps
processor = self.ctx.get_hf_processor(
Qwen2_5OmniProcessor,
image_processor=self.get_image_processor(
min_pixels=min_pixels,
max_pixels=max_pixels,
size=size,
use_fast=kwargs.get("use_fast", True)),
image_processor=self.get_image_processor(min_pixels=min_pixels,
max_pixels=max_pixels,
size=size,
use_fast=kwargs.get(
"use_fast", True)),
**kwargs,
)
if not hasattr(processor, "audio_token"):
Expand Down
10 changes: 5 additions & 5 deletions vllm/model_executor/models/qwen2_5_vl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -794,11 +794,11 @@ def get_hf_processor(

return self.ctx.get_hf_processor(
Qwen2_5_VLProcessor,
image_processor=self.get_image_processor(
min_pixels=min_pixels,
max_pixels=max_pixels,
size=size,
use_fast=kwargs.get("use_fast", True)),
image_processor=self.get_image_processor(min_pixels=min_pixels,
max_pixels=max_pixels,
size=size,
use_fast=kwargs.get(
"use_fast", True)),
**kwargs,
)

Expand Down
10 changes: 5 additions & 5 deletions vllm/model_executor/models/qwen2_vl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -759,11 +759,11 @@ def get_hf_processor(
) -> Qwen2VLProcessor:
return self.ctx.get_hf_processor(
Qwen2VLProcessor,
image_processor=self.get_image_processor(
min_pixels=min_pixels,
max_pixels=max_pixels,
size=size,
use_fast=kwargs.get("use_fast", True)),
image_processor=self.get_image_processor(min_pixels=min_pixels,
max_pixels=max_pixels,
size=size,
use_fast=kwargs.get(
"use_fast", True)),
**kwargs,
)

Expand Down
13 changes: 3 additions & 10 deletions vllm/v1/attention/backends/mla/rocm_aiter_mla.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -201,16 +201,9 @@ def _forward_decode(

kv_buffer = kv_c_and_k_pe_cache.unsqueeze(2)

if self.num_heads == 16:
# AITER MLA decode kernel only supports
# max_seqlen_q=1 when using 16 heads.
max_seqlen_qo = 1
else:
# AITER MLA decode Kernel handles arbitrary
# max_seqlen_q values when using 128 heads.
assert attn_metadata.prefill is not None
max_seqlen_qo = attn_metadata.prefill.max_query_len

# max_seqlen_qo must be 1 except for MTP
# TODO: Find the best value for MTP
max_seqlen_qo = 1
aiter_mla_decode_fwd(q, kv_buffer, o, self.scale,
attn_metadata.decode.qo_indptr, max_seqlen_qo,
attn_metadata.decode.paged_kv_indptr,
Expand Down