biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Update deprecated type hinting in model_executor/layers (#18056)

Browse Source 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2025-05-13 12:17:23 +01:00
committed by GitHub
parent 906f0598fc
commit 6223dd8114
87 changed files with 523 additions and 523 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
vllm/model_executor/layers/quantization/utils/quant_utils.py
View File

@@ -1,7 +1,8 @@
# SPDX-License-Identifier: Apache-2.0
"""This file is used for /tests and /benchmarks"""
from collections.abc import Mapping
from types import MappingProxyType
from typing import List, Mapping, Optional, Tuple
from typing import Optional
import numpy
import torch
@@ -15,7 +16,7 @@ SUPPORTED_GROUP_SIZES = [-1, 32, 64, 128]
# Normalize the group_shape to the full extent for any dims that are -1
def _normalize_quant_group_shape(x: torch.Tensor, group_shape: Tuple[int,
def _normalize_quant_group_shape(x: torch.Tensor, group_shape: tuple[int,
int]):
# -1 means full extent
return (group_shape[0] if group_shape[0] > 0 else x.shape[-2],
@@ -56,9 +57,9 @@ def group_broadcast(t, shape):
# (i.e. per-token-per-group)
def scaled_quantize(
x: torch.Tensor,
group_shape: Tuple[int, int],
group_shape: tuple[int, int],
quant_dtype: torch.dtype,
) -> Tuple[torch.Tensor, torch.Tensor]:
) -> tuple[torch.Tensor, torch.Tensor]:
group_shape = _normalize_quant_group_shape(x, group_shape)
assert quant_dtype.is_floating_point, \
"currently `scaled_quantize` only supports floating point dtypes " \
@@ -97,9 +98,9 @@ def scaled_quantize(
def scaled_dequantize(
x_q: torch.Tensor,
x_s: torch.Tensor,
group_shape: Optional[Tuple[int, int]] = None,
group_shape: Optional[tuple[int, int]] = None,
out_dtype: torch.dtype = torch.float32,
) -> Tuple[torch.Tensor, torch.Tensor]:
) -> tuple[torch.Tensor, torch.Tensor]:
if group_shape is not None:
group_shape = _normalize_quant_group_shape(x_q, group_shape)
@@ -173,8 +174,8 @@ def unpack_quantized_values_into_int32(w_q: torch.Tensor,
def is_layer_skipped(
prefix: str,
ignored_layers: List[str],
fused_mapping: Mapping[str, List[str]] = MappingProxyType({})
ignored_layers: list[str],
fused_mapping: Mapping[str, list[str]] = MappingProxyType({})
) -> bool:
# prefix: model.layers.0.self_attn.q_proj
# proj_name: q_proj