Skip to content

[FEAT] [ROCm] [AITER]: Add AITER HIP block quant kernel #21242

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 3 commits into from
Jul 28, 2025
Merged

[FEAT] [ROCm] [AITER]: Add AITER HIP block quant kernel #21242

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
4172797
add aiter block quant
tjtanaa Jul 20, 2025
8132a6e
update import condition
tjtanaa Jul 27, 2025
0ba022c
Merge remote-tracking branch 'origin/main' into aiter-quant
tjtanaa Jul 28, 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
15 changes: 13 additions & 2 deletions vllm/model_executor/layers/quantization/utils/fp8_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -82,6 +82,13 @@ def rocm_aiter_gemm_w8a8_blockscale_fake(
fake_impl=rocm_aiter_gemm_w8a8_blockscale_fake,
dispatch_key=current_platform.dispatch_key,
)
if (envs.VLLM_ROCM_USE_AITER and envs.VLLM_ROCM_USE_AITER_LINEAR
and current_platform.is_fp8_fnuz()):

import aiter as rocm_aiter
from aiter import get_hip_quant

aiter_per1x128_quant = get_hip_quant(rocm_aiter.QuantType.per_1x128)


def dispatch_w8a8_blockscale_func(
Expand Down Expand Up @@ -178,8 +185,12 @@ def apply_w8a8_block_fp8_linear(
block_size, input.dtype)

else:
q_input, x_scale = per_token_group_quant_fp8(
input_2d, block_size[1], column_major_scales=use_cutlass)
if use_aiter_and_is_supported:
q_input, x_scale = aiter_per1x128_quant(
input_2d.contiguous(), quant_dtype=rocm_aiter.dtypes.fp8)
Comment on lines +188 to +190
Copy link
Contributor

Choose a reason for hiding this comment

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

critical

The call to aiter_per1x128_quant is not guarded by a try-except block, which can lead to a crash if the aiter module fails to load. Wrap this call in a try-except block to handle potential import errors gracefully.

try:
            q_input, x_scale = aiter_per1x128_quant(
                input_2d.contiguous(), quant_dtype=rocm_aiter.dtypes.fp8)
        except NameError as e:
            raise ImportError("AITER is not properly installed or configured.") from e

else:
q_input, x_scale = per_token_group_quant_fp8(
input_2d, block_size[1], column_major_scales=use_cutlass)

output = w8a8_blockscale_func(q_input, weight, x_scale, weight_scale,
block_size, input.dtype)
Expand Down