99aedaff63
### What this PR does / why we need it?
**Scope of Changes**:
| File Path |
| :--- |
|` vllm_ascend/quantization/compressed_tensors/compressed_tensors.py`|
|` vllm_ascend/quantization/quant_config.py`|
|` vllm_ascend/quantization/utils.py`|
|` vllm_ascend/quantization/w4a16.py`|
|` vllm_ascend/quantization/w4a4_flatquant_dynamic.py`|
|` vllm_ascend/quantization/w4a8_dynamic.py`|
|` vllm_ascend/quantization/w8a16.py`|
|` vllm_ascend/quantization/w8a8.py`|
|` vllm_ascend/quantization/w8a8_dynamic.py`|
|` vllm_ascend/quantization/w8a8_pdmix.py`|
|` vllm_ascend/quantization/w8a8mxfp8.py`|
|` vllm_ascend/sample/rejection_sampler.py`|
|` vllm_ascend/sample/sampler.py`|
|` vllm_ascend/worker/block_table.py`|
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
2c24bc6996
Signed-off-by: MrZ20 <2609716663@qq.com>
79 lines
2.5 KiB
Python
79 lines
2.5 KiB
Python
#
|
|
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
|
|
# This file is a part of the vllm-ascend project.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
#
|
|
|
|
from typing import Any
|
|
|
|
import torch
|
|
import torch_npu
|
|
|
|
from vllm_ascend.utils import maybe_trans_nz
|
|
|
|
from .base import AscendLinearScheme
|
|
from .registry import register_scheme
|
|
|
|
|
|
@register_scheme("W8A16", "linear")
|
|
class AscendW8A16LinearMethod(AscendLinearScheme):
|
|
"""Linear method for Ascend W8A16.
|
|
|
|
This scheme uses 8-bit quantized weights with 16-bit activations.
|
|
"""
|
|
|
|
def __init__(self) -> None:
|
|
pass
|
|
|
|
def get_weight(
|
|
self,
|
|
input_size: int,
|
|
output_size: int,
|
|
params_dtype: torch.dtype = torch.bfloat16,
|
|
) -> dict[str, Any]:
|
|
params_dict = {"weight": torch.empty(output_size, input_size, dtype=torch.int8)}
|
|
return params_dict
|
|
|
|
def get_perchannel_param(
|
|
self,
|
|
output_size: int,
|
|
params_dtype: torch.dtype,
|
|
) -> dict[str, Any]:
|
|
params_dict = {}
|
|
params_dict["weight_scale"] = torch.empty(output_size, 1, dtype=params_dtype)
|
|
params_dict["weight_offset"] = torch.empty(output_size, 1, dtype=params_dtype)
|
|
return params_dict
|
|
|
|
def apply(
|
|
self,
|
|
layer: torch.nn.Module,
|
|
x: torch.Tensor,
|
|
bias: torch.Tensor | None = None,
|
|
tp_rank: int | None = 0,
|
|
) -> torch.Tensor:
|
|
output = torch_npu.npu_weight_quant_batchmatmul(
|
|
x=x,
|
|
weight=layer.weight,
|
|
antiquant_scale=layer.weight_scale,
|
|
antiquant_offset=layer.weight_offset,
|
|
bias=bias,
|
|
)
|
|
return output
|
|
|
|
def process_weights_after_loading(self, layer):
|
|
layer.weight.data = layer.weight.data.transpose(0, 1).contiguous()
|
|
layer.weight.data = maybe_trans_nz(layer.weight.data)
|
|
layer.weight_scale.data = torch.flatten(layer.weight_scale.data)
|
|
layer.weight_offset.data = torch.flatten(layer.weight_offset.data)
|