xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-08-15 17:47:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Remove dead block_quant_to_tensor_quant function (#31294)

Browse Source 
Co-authored-by: yurekami <yurekami@users.noreply.github.com>
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
ゆり 2025-12-25 01:22:48 +08:00 committed by GitHub
parent 5d93089686
commit 506eb0f454
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 1 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
vllm/model_executor/layers/quantization/utils/fp8_utils.py
View File

@ -15,10 +15,7 @@ from vllm import _custom_ops as ops
from vllm._aiter_ops import rocm_aiter_ops from vllm._aiter_ops import rocm_aiter_ops
from vllm.logger import init_logger from vllm.logger import init_logger
from vllm.model_executor.layers.quantization.input_quant_fp8 import QuantFP8 from vllm.model_executor.layers.quantization.input_quant_fp8 import QuantFP8
from vllm.model_executor.layers.quantization.utils.quant_utils import ( from vllm.model_executor.layers.quantization.utils.quant_utils import GroupShape
GroupShape,
group_broadcast,
)
from vllm.model_executor.layers.quantization.utils.w8a8_utils import ( from vllm.model_executor.layers.quantization.utils.w8a8_utils import (
CUTLASS_BLOCK_FP8_SUPPORTED, CUTLASS_BLOCK_FP8_SUPPORTED,
) )
@ -463,21 +460,6 @@ def input_to_float8(
return x_scl_sat.to(dtype).contiguous(), scale.float().reciprocal() return x_scl_sat.to(dtype).contiguous(), scale.float().reciprocal()
def block_quant_to_tensor_quant(
x_q_block: torch.Tensor,
x_s: torch.Tensor,
) -> tuple[torch.Tensor, torch.Tensor]:
"""This function converts block-wise quantization to tensor-wise
quantization. The inputs are block-wise quantization tensor `x_q_block`,
block-wise quantization scale and the block size.
The outputs are tensor-wise quantization tensor and tensor-wise
quantization scale. Note only float8 is supported for now.
"""
x_dq_block = group_broadcast(x_q_block, x_s)
x_q_tensor, scale = input_to_float8(x_dq_block, dtype=x_q_block.dtype)
return x_q_tensor, scale
@triton.jit @triton.jit
def _per_token_group_quant_fp8( def _per_token_group_quant_fp8(
# Pointers to inputs and output # Pointers to inputs and output