Skip to content

enable aiter fp8 blockscale quant #7520

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
zhyncs merged 2 commits into from
Jun 25, 2025
Merged

enable aiter fp8 blockscale quant #7520

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
2616035
enable aiter fp8 blockscale quant
valarLip Jun 25, 2025
e324d49
Merge branch 'main' into lip/enable_aiter_fp8quant
HaiShaw Jun 25, 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
9 changes: 5 additions & 4 deletions python/sglang/srt/layers/quantization/fp8_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,7 +42,10 @@
_use_aiter = get_bool_env_var("SGLANG_USE_AITER") and _is_hip

if _use_aiter:
from aiter import gemm_a8w8_blockscale_CK
import aiter
from aiter import gemm_a8w8_blockscale_CK, get_hip_quant

aiter_per1x128_quant = get_hip_quant(aiter.QuantType.per_1x128)

if _is_cuda:
from sgl_kernel import fp8_blockwise_scaled_mm, fp8_scaled_mm
Expand Down Expand Up @@ -271,9 +274,7 @@ def aiter_w8a8_block_fp8_linear(
input_2d = input.view(-1, input.shape[-1])
output_shape = [*input.shape[:-1], weight.shape[0]]

q_input, x_scale = per_token_group_quant_fp8(
input_2d, block_size[1], column_major_scales=False
)
q_input, x_scale = aiter_per1x128_quant(input_2d, quant_dtype=aiter.dtypes.fp8)
output = gemm_a8w8_blockscale_CK(
q_input, weight, x_scale, weight_scale, dtype=input.dtype
)
Expand Down