mirror of
https://github.com/handsomezhuzhu/vllm-npu-plugin.git
synced 2026-02-20 19:50:15 +00:00
大改
This commit is contained in:
0
vllm_npu/lora/__init__.py
Normal file
0
vllm_npu/lora/__init__.py
Normal file
113
vllm_npu/lora/lora_ops.py
Normal file
113
vllm_npu/lora/lora_ops.py
Normal file
@@ -0,0 +1,113 @@
|
||||
#
|
||||
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
|
||||
#
|
||||
# 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.
|
||||
|
||||
import torch
|
||||
|
||||
|
||||
def bgmv_shrink(inputs: torch.Tensor,
|
||||
lora_a_weights: torch.Tensor,
|
||||
output_tensor: torch.Tensor,
|
||||
lora_indices_tensor: torch.Tensor,
|
||||
scaling: float = 1.0):
|
||||
return torch.ops._C_ascend.bgmv_shrink(
|
||||
inputs,
|
||||
lora_a_weights,
|
||||
lora_indices_tensor,
|
||||
output_tensor,
|
||||
scaling,
|
||||
)
|
||||
|
||||
|
||||
def bgmv_expand(inputs: torch.Tensor,
|
||||
lora_b_weights: torch.Tensor,
|
||||
output_tensor: torch.Tensor,
|
||||
lora_indices_tensor: torch.Tensor,
|
||||
add_inputs: bool = True):
|
||||
return torch.ops._C_ascend.bgmv_expand(
|
||||
inputs,
|
||||
lora_b_weights,
|
||||
lora_indices_tensor,
|
||||
output_tensor,
|
||||
0,
|
||||
output_tensor.size(1),
|
||||
)
|
||||
|
||||
|
||||
def bgmv_expand_slice(inputs: torch.Tensor,
|
||||
lora_b_weights: torch.Tensor,
|
||||
output_tensor: torch.Tensor,
|
||||
lora_indices_tensor: torch.Tensor,
|
||||
slice_offset: int,
|
||||
slice_size: int,
|
||||
add_inputs: bool = True):
|
||||
return torch.ops._C_ascend.bgmv_expand(inputs, lora_b_weights,
|
||||
lora_indices_tensor, output_tensor,
|
||||
slice_offset, slice_size)
|
||||
|
||||
|
||||
def sgmv_shrink(
|
||||
inputs: torch.Tensor,
|
||||
lora_a_weights: torch.Tensor,
|
||||
output_tensor: torch.Tensor,
|
||||
b_seq_start_loc: torch.Tensor,
|
||||
seq_len_tensor: torch.Tensor,
|
||||
lora_indices_tensor: torch.Tensor,
|
||||
batches: int,
|
||||
max_seq_length: int,
|
||||
token_nums: int,
|
||||
scaling: float,
|
||||
):
|
||||
return torch.ops._C_ascend.sgmv_shrink(inputs, lora_a_weights,
|
||||
lora_indices_tensor, seq_len_tensor,
|
||||
output_tensor, scaling)
|
||||
|
||||
|
||||
def sgmv_expand(inputs: torch.Tensor,
|
||||
lora_b_weights: torch.Tensor,
|
||||
output_tensor: torch.Tensor,
|
||||
b_seq_start_loc: torch.Tensor,
|
||||
seq_len_tensor: torch.Tensor,
|
||||
lora_indices_tensor: torch.Tensor,
|
||||
batches: int,
|
||||
max_seq_length: int,
|
||||
token_nums: int,
|
||||
add_inputs: bool = False):
|
||||
return torch.ops._C_ascend.sgmv_expand(
|
||||
inputs,
|
||||
lora_b_weights,
|
||||
lora_indices_tensor,
|
||||
seq_len_tensor,
|
||||
output_tensor,
|
||||
0,
|
||||
output_tensor.size(1),
|
||||
)
|
||||
|
||||
|
||||
def sgmv_expand_slice(inputs: torch.Tensor,
|
||||
lora_b_weights: torch.Tensor,
|
||||
output_tensor: torch.Tensor,
|
||||
b_seq_start_loc: torch.Tensor,
|
||||
seq_len_tensor: torch.Tensor,
|
||||
lora_indices_tensor: torch.Tensor,
|
||||
batches: int,
|
||||
max_seq_length: int,
|
||||
token_nums: int,
|
||||
slice_offset: int,
|
||||
slice_size: int,
|
||||
add_inputs: bool = False):
|
||||
return torch.ops._C_ascend.sgmv_expand(inputs, lora_b_weights,
|
||||
lora_indices_tensor, seq_len_tensor,
|
||||
output_tensor, slice_offset,
|
||||
slice_size)
|
||||
356
vllm_npu/lora/punica_npu.py
Normal file
356
vllm_npu/lora/punica_npu.py
Normal file
@@ -0,0 +1,356 @@
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
from typing import Callable, Optional, Tuple, Union
|
||||
|
||||
import torch
|
||||
|
||||
from vllm_npu.utils import is_310p
|
||||
|
||||
if is_310p():
|
||||
from vllm.lora.ops.torch_ops import (bgmv_expand, bgmv_expand_slice,
|
||||
bgmv_shrink, sgmv_expand,
|
||||
sgmv_expand_slice, sgmv_shrink)
|
||||
else:
|
||||
from vllm_npu.lora.lora_ops import (bgmv_expand, bgmv_expand_slice,
|
||||
bgmv_shrink, sgmv_expand,
|
||||
sgmv_expand_slice, sgmv_shrink)
|
||||
|
||||
from vllm.lora.punica_wrapper.punica_base import PunicaWrapperBase
|
||||
|
||||
from vllm_npu.lora.utils import refresh_all_lora_classes
|
||||
|
||||
|
||||
# The platforms that are compatible with the PyTorch-native implementation can
|
||||
# inherit this class
|
||||
class PunicaWrapperNPU(PunicaWrapperBase):
|
||||
"""
|
||||
PunicaWrapperNPU is designed to manage and provide metadata for the punica
|
||||
kernel. The main function is to maintain the state information for
|
||||
Multi-LoRA, and to provide the interface for the pytorch punica ops.
|
||||
"""
|
||||
|
||||
def __init__(self, max_num_batched_tokens: int, max_batches: int,
|
||||
device: Union[torch.device, str], **kwargs):
|
||||
PunicaWrapperBase.__init__(self, max_num_batched_tokens, max_batches,
|
||||
device)
|
||||
refresh_all_lora_classes()
|
||||
|
||||
def _shrink_prefill(
|
||||
self,
|
||||
y: torch.Tensor,
|
||||
x: torch.Tensor,
|
||||
w_t_all: torch.Tensor,
|
||||
scale: float,
|
||||
):
|
||||
#No LoRA request, so return directly
|
||||
if self.no_lora:
|
||||
return
|
||||
sgmv_shrink(
|
||||
x,
|
||||
w_t_all,
|
||||
y,
|
||||
*self.prefill_metadata,
|
||||
scale,
|
||||
)
|
||||
|
||||
def _shrink_decode(
|
||||
self,
|
||||
y: torch.Tensor,
|
||||
x: torch.Tensor,
|
||||
w_t_all: torch.Tensor,
|
||||
scale: float,
|
||||
):
|
||||
bgmv_shrink(x, w_t_all, y, self.token_lora_indices, scale)
|
||||
|
||||
def _expand_prefill(
|
||||
self,
|
||||
y: torch.Tensor,
|
||||
x: torch.Tensor,
|
||||
w_t_all: torch.Tensor,
|
||||
add_inputs: bool,
|
||||
):
|
||||
#No LoRA request, so return directly
|
||||
if self.no_lora:
|
||||
return
|
||||
sgmv_expand(
|
||||
x,
|
||||
w_t_all,
|
||||
y,
|
||||
*self.prefill_metadata,
|
||||
add_inputs,
|
||||
)
|
||||
|
||||
def _expand_decode(
|
||||
self,
|
||||
y: torch.Tensor,
|
||||
x: torch.Tensor,
|
||||
w_t_all: torch.Tensor,
|
||||
add_inputs: bool,
|
||||
):
|
||||
bgmv_expand(x, w_t_all, y, self.token_lora_indices, add_inputs)
|
||||
|
||||
def _expand_slice_prefill(
|
||||
self,
|
||||
y: torch.Tensor,
|
||||
x: torch.Tensor,
|
||||
w_t_all: torch.Tensor,
|
||||
y_offset: int,
|
||||
y_slice_size: int,
|
||||
add_inputs: bool,
|
||||
):
|
||||
#No LoRA request, so return directly
|
||||
if self.no_lora:
|
||||
return
|
||||
sgmv_expand_slice(
|
||||
x,
|
||||
w_t_all,
|
||||
y,
|
||||
*self.prefill_metadata,
|
||||
y_offset,
|
||||
y_slice_size,
|
||||
add_inputs,
|
||||
)
|
||||
|
||||
def _expand_slice_decode(
|
||||
self,
|
||||
y: torch.Tensor,
|
||||
x: torch.Tensor,
|
||||
w_t_all: torch.Tensor,
|
||||
y_offset: int,
|
||||
y_slice_size: int,
|
||||
add_inputs: bool,
|
||||
):
|
||||
bgmv_expand_slice(x, w_t_all, y, self.token_lora_indices, y_offset,
|
||||
y_slice_size, add_inputs)
|
||||
|
||||
def _apply_expand(
|
||||
self,
|
||||
y: torch.Tensor,
|
||||
x: torch.Tensor,
|
||||
w_t_all: torch.Tensor,
|
||||
y_offset: int,
|
||||
y_slice_size: int,
|
||||
add_inputs: bool = True,
|
||||
):
|
||||
"""
|
||||
Perform the ` y[:,y_offset:y_offset+y_slice_size]+=x@w_t_all`
|
||||
computation, which is suitable for the
|
||||
GEMM of lora'b.
|
||||
"""
|
||||
|
||||
expand_slice_fun: Callable = (self._expand_slice_prefill
|
||||
if self.is_prefill else
|
||||
self._expand_slice_decode)
|
||||
expand_slice_fun(y, x, w_t_all, y_offset, y_slice_size, add_inputs)
|
||||
|
||||
def _apply_shrink(self, y: torch.Tensor, x: torch.Tensor,
|
||||
w_t_all: torch.Tensor, scale: float):
|
||||
"""
|
||||
Perform the ` y+=x@w_t_all` computation, which is suitable for the
|
||||
GEMM of lora'a.
|
||||
When `is_prefill is` true, it indicates that it is currently the
|
||||
prefill stage, and the `_shrink_prefill` function should be called.
|
||||
Otherwise, it is the decode stage, and the _shrink_decode function
|
||||
should be called.
|
||||
"""
|
||||
y_org = y
|
||||
y = y.view(-1, y.shape[-1])
|
||||
shrink_fun: Callable = (self._shrink_prefill
|
||||
if self.is_prefill else self._shrink_decode)
|
||||
shrink_fun(y, x, w_t_all, scale)
|
||||
y = y.view_as(y_org)
|
||||
|
||||
def add_shrink(self, y: Union[Tuple[torch.Tensor, ...], torch.Tensor],
|
||||
x: torch.Tensor, lora_a_stacked: Tuple[torch.Tensor, ...],
|
||||
scale: float, **kwargs):
|
||||
"""
|
||||
Performs GEMM for multiple slices of lora_a.
|
||||
When `is_prefill is` true, it indicates that it is currently the
|
||||
prefill stage, and the `_shrink_prefill` function should be called.
|
||||
Otherwise, it is the decode stage, and the _shrink_decode function
|
||||
should be called.
|
||||
|
||||
Semantics:
|
||||
for i in range(len(lora_a_stacked)):
|
||||
y[i] += (x @ lora_a_stacked[i]) * scale
|
||||
|
||||
Args:
|
||||
y (Union[Tuple[torch.Tensor, ...], torch.Tensor]): Output tensors
|
||||
x (torch.Tensor): Input tensor
|
||||
lora_a_stacked (Tuple[torch.Tensor, ...]): lora_a's weights
|
||||
scale (float): Scaling factor for the operation
|
||||
"""
|
||||
|
||||
x = x.view(-1, x.shape[-1])
|
||||
# TODO fuse these kernels
|
||||
for slice_idx in range(len(lora_a_stacked)):
|
||||
self._apply_shrink(y[slice_idx], x, lora_a_stacked[slice_idx],
|
||||
scale)
|
||||
|
||||
def add_expand(self,
|
||||
y: torch.Tensor,
|
||||
x: Union[Tuple[torch.Tensor, ...], torch.Tensor],
|
||||
lora_b_stacked: Tuple[torch.Tensor, ...],
|
||||
lora_bias_stacked: Optional[Tuple[torch.Tensor, ...]],
|
||||
output_slices: Tuple[int, ...],
|
||||
offset_start: int = 0,
|
||||
add_inputs=True,
|
||||
**kwargs) -> None:
|
||||
"""
|
||||
Performs GEMM and bias addition for multiple slices of lora_b.
|
||||
|
||||
Semantics:
|
||||
for i in range(len(lora_b_stacked)):
|
||||
slice = output_slices[i]
|
||||
y[:, offset:offset+slice] += x[i] @ lora_b_stacked[i] +
|
||||
lora_bias_stacked[i]
|
||||
offset += slice
|
||||
|
||||
Args:
|
||||
y (torch.Tensor): Output tensor.
|
||||
x (Union[Tuple[torch.Tensor, ...], torch.Tensor]): Input tensors
|
||||
lora_b_stacked (Tuple[torch.Tensor, ...]): lora_b's weight
|
||||
lora_bias_stacked (Optional[Tuple[torch.Tensor, ...]]):
|
||||
bias's weight
|
||||
output_slices (Tuple[int, ...]): Every slice's size
|
||||
add_inputs (bool): Defaults to True.
|
||||
"""
|
||||
y_org = y
|
||||
y = y.view(-1, y.shape[-1])
|
||||
offset_left = offset_start
|
||||
if lora_bias_stacked is not None:
|
||||
self._apply_bias(self.token_lora_indices, y, output_slices,
|
||||
lora_bias_stacked)
|
||||
for slice_idx in range(len(lora_b_stacked)):
|
||||
self._apply_expand(
|
||||
y,
|
||||
x[slice_idx],
|
||||
lora_b_stacked[slice_idx],
|
||||
offset_left,
|
||||
output_slices[slice_idx],
|
||||
add_inputs=add_inputs,
|
||||
)
|
||||
offset_left += output_slices[slice_idx]
|
||||
y = y.view_as(y_org)
|
||||
|
||||
def add_lora_embedding(self,
|
||||
y: torch.Tensor,
|
||||
x: torch.Tensor,
|
||||
lora_b_stacked: torch.Tensor,
|
||||
add_inputs: bool = True,
|
||||
**kwargs) -> None:
|
||||
"""
|
||||
Applies lora specifically for VocabParallelEmbeddingWithLoRA.
|
||||
|
||||
Semantics:
|
||||
y += x @ lora_b_stacked
|
||||
|
||||
Args:
|
||||
y (torch.Tensor): Output tensor.
|
||||
x (torch.Tensor): Input tensor.
|
||||
lora_b_stacked (torch.Tensor): lora_b's weights.
|
||||
add_inputs (bool): Default to True.
|
||||
"""
|
||||
|
||||
# Embedding layer only need expand op
|
||||
expand_fun: Callable = (self._expand_prefill
|
||||
if self.is_prefill else self._expand_decode)
|
||||
expand_fun(y, x, lora_b_stacked, add_inputs)
|
||||
|
||||
def add_lora_linear(self,
|
||||
y: torch.Tensor,
|
||||
x: torch.Tensor,
|
||||
lora_a_stacked: Tuple[torch.Tensor, ...],
|
||||
lora_b_stacked: Tuple[torch.Tensor, ...],
|
||||
lora_bias_stacked: Optional[Tuple[torch.Tensor, ...]],
|
||||
scale: float,
|
||||
output_slices: Tuple[int, ...],
|
||||
*,
|
||||
buffer: Optional[Tuple[torch.Tensor, ...]] = None,
|
||||
**kwargs) -> None:
|
||||
"""
|
||||
Applicable to linear-related lora.
|
||||
|
||||
Semantics:
|
||||
for i in range(len(lora_a_stacked)):
|
||||
y[i] += (
|
||||
x[i].unsqueeze(0)
|
||||
@ lora_a_stacked[indices[i], layer_idx, :, :]
|
||||
@ lora_b_stacked[indices[i], layer_idx, :, :]
|
||||
* scale
|
||||
).squeeze(0)+lora_bias_stacked[i]
|
||||
|
||||
Args:
|
||||
y (torch.Tensor): Output tensor. Will be changed in-place.
|
||||
x (torch.Tensor): Input tensor
|
||||
lora_a_stacked (Tuple[torch.Tensor, ...]): lora_a's weight.
|
||||
lora_b_stacked (Tuple[torch.Tensor, ...]): lora_b's weight.
|
||||
lora_bias_stacked (Optional[Tuple[torch.Tensor, ...]]): lora's bias.
|
||||
scale (float): Scaling factor.
|
||||
output_slices (Tuple[int, ...]): Every slice's size.
|
||||
buffer (Optional[Tuple[torch.Tensor, ...]]): Defaults to None.
|
||||
"""
|
||||
|
||||
assert len(lora_a_stacked) == len(lora_b_stacked) == len(output_slices)
|
||||
if lora_bias_stacked is not None:
|
||||
assert len(lora_bias_stacked) == len(output_slices)
|
||||
y = self._apply_bias(self.token_lora_indices, y, output_slices,
|
||||
lora_bias_stacked)
|
||||
|
||||
if buffer is None:
|
||||
r = lora_b_stacked[0].size(-1)
|
||||
# We set the buffer to be float32 by default, consistent with the
|
||||
# triton op
|
||||
buffer = tuple(
|
||||
torch.zeros(
|
||||
(x.size(0), r), dtype=torch.float32, device=x.device)
|
||||
for _ in range(len(output_slices)))
|
||||
self.add_shrink(buffer, x, lora_a_stacked, scale, **kwargs)
|
||||
self.add_expand(y,
|
||||
buffer,
|
||||
lora_b_stacked,
|
||||
None,
|
||||
output_slices,
|
||||
add_inputs=True,
|
||||
**kwargs)
|
||||
|
||||
def add_lora_logits(self,
|
||||
y: torch.Tensor,
|
||||
x: torch.Tensor,
|
||||
lora_a_stacked: torch.Tensor,
|
||||
lora_b_stacked: torch.Tensor,
|
||||
scale,
|
||||
*,
|
||||
buffer: Optional[torch.Tensor] = None,
|
||||
**kwargs) -> None:
|
||||
"""
|
||||
Applies lora specifically for LogitsProcessorWithLoRA.
|
||||
|
||||
Semantics:
|
||||
buffer = (x @ lora_a_stacked) * scale
|
||||
y += buffer @ lora_b_stacked
|
||||
|
||||
Args:
|
||||
y (torch.Tensor): Output tensor.
|
||||
x (torch.Tensor): Input tensor.
|
||||
lora_a_stacked (torch.Tensor): lora_a's weights.
|
||||
lora_b_stacked (torch.Tensor):lora_b's weights.
|
||||
scale (float): Scaling factor.
|
||||
buffer (Optional[torch.Tensor]):Default to None.
|
||||
"""
|
||||
y_org = y
|
||||
y = y.view(-1, y.shape[-1])
|
||||
x = x.view(-1, x.shape[-1])
|
||||
r = lora_b_stacked.size(-1)
|
||||
|
||||
if buffer is None:
|
||||
buffer = torch.zeros((x.size(0), r),
|
||||
dtype=torch.float32,
|
||||
device=x.device)
|
||||
|
||||
indices = self.sampler_indices
|
||||
|
||||
bgmv_shrink(x, lora_a_stacked, buffer, indices, scale)
|
||||
bgmv_expand(buffer, lora_b_stacked, y, indices, add_inputs=True)
|
||||
|
||||
y = y.view_as(y_org)
|
||||
110
vllm_npu/lora/utils.py
Normal file
110
vllm_npu/lora/utils.py
Normal file
@@ -0,0 +1,110 @@
|
||||
from typing import Optional
|
||||
|
||||
import vllm
|
||||
from torch import nn
|
||||
from transformers import PretrainedConfig
|
||||
from vllm.config import LoRAConfig
|
||||
from vllm.lora.layers import (ColumnParallelLinearWithLoRA,
|
||||
MergedColumnParallelLinearWithLoRA,
|
||||
MergedQKVParallelLinearWithLoRA,
|
||||
QKVParallelLinearWithLoRA,
|
||||
RowParallelLinearWithLoRA,
|
||||
VocabParallelEmbeddingWithLoRA)
|
||||
from vllm.lora.layers.utils import _not_fully_sharded_can_replace
|
||||
|
||||
from vllm_npu.ops.linear import (AscendColumnParallelLinear,
|
||||
AscendMergedColumnParallelLinear,
|
||||
AscendQKVParallelLinear,
|
||||
AscendRowParallelLinear)
|
||||
from vllm_npu.ops.vocab_parallel_embedding import \
|
||||
AscendVocabParallelEmbedding
|
||||
|
||||
|
||||
class AscendColumnParallelLinearWithLoRA(ColumnParallelLinearWithLoRA):
|
||||
|
||||
@classmethod
|
||||
def can_replace_layer(
|
||||
cls,
|
||||
source_layer: nn.Module,
|
||||
lora_config: LoRAConfig,
|
||||
packed_modules_list: list,
|
||||
model_config: Optional[PretrainedConfig],
|
||||
) -> bool:
|
||||
return type(source_layer) is AscendColumnParallelLinear
|
||||
|
||||
|
||||
class AscendMergedColumnParallelLinearWithLoRA(
|
||||
MergedColumnParallelLinearWithLoRA):
|
||||
|
||||
@classmethod
|
||||
def can_replace_layer(
|
||||
cls,
|
||||
source_layer: nn.Module,
|
||||
lora_config: LoRAConfig,
|
||||
packed_modules_list: list,
|
||||
model_config: Optional[PretrainedConfig],
|
||||
) -> bool:
|
||||
return type(source_layer) is AscendMergedColumnParallelLinear
|
||||
|
||||
|
||||
class AscendRowParallelLinearWithLoRA(RowParallelLinearWithLoRA):
|
||||
|
||||
@classmethod
|
||||
def can_replace_layer(
|
||||
cls,
|
||||
source_layer: nn.Module,
|
||||
lora_config: LoRAConfig,
|
||||
packed_modules_list: list,
|
||||
model_config: Optional[PretrainedConfig],
|
||||
) -> bool:
|
||||
return type(source_layer) is AscendRowParallelLinear
|
||||
|
||||
|
||||
class AscendVocabParallelEmbeddingWithLoRA(VocabParallelEmbeddingWithLoRA):
|
||||
|
||||
@classmethod
|
||||
def can_replace_layer(
|
||||
cls,
|
||||
source_layer: nn.Module,
|
||||
lora_config: LoRAConfig,
|
||||
packed_modules_list: list,
|
||||
model_config: Optional[PretrainedConfig],
|
||||
) -> bool:
|
||||
return type(source_layer) is AscendVocabParallelEmbedding
|
||||
|
||||
|
||||
class AscendQKVParallelLinearWithLoRA(QKVParallelLinearWithLoRA):
|
||||
|
||||
@classmethod
|
||||
@_not_fully_sharded_can_replace
|
||||
def can_replace_layer(cls, source_layer: nn.Module,
|
||||
lora_config: LoRAConfig, packed_modules_list: list,
|
||||
model_config: Optional[PretrainedConfig]) -> bool:
|
||||
return type(source_layer) is AscendQKVParallelLinear and len(
|
||||
packed_modules_list) == 1
|
||||
|
||||
|
||||
class AscendMergedQKVParallelLinearWithLoRA(MergedQKVParallelLinearWithLoRA):
|
||||
|
||||
@classmethod
|
||||
@_not_fully_sharded_can_replace
|
||||
def can_replace_layer(
|
||||
cls,
|
||||
source_layer: nn.Module,
|
||||
lora_config: LoRAConfig,
|
||||
packed_modules_list: list,
|
||||
model_config: Optional[PretrainedConfig],
|
||||
) -> bool:
|
||||
return (type(source_layer) is AscendQKVParallelLinear
|
||||
and len(packed_modules_list) == 3)
|
||||
|
||||
|
||||
def refresh_all_lora_classes():
|
||||
vllm.lora.utils._all_lora_classes.add(AscendColumnParallelLinearWithLoRA)
|
||||
vllm.lora.utils._all_lora_classes.add(
|
||||
AscendMergedColumnParallelLinearWithLoRA)
|
||||
vllm.lora.utils._all_lora_classes.add(AscendRowParallelLinearWithLoRA)
|
||||
vllm.lora.utils._all_lora_classes.add(AscendVocabParallelEmbeddingWithLoRA)
|
||||
vllm.lora.utils._all_lora_classes.add(AscendQKVParallelLinearWithLoRA)
|
||||
vllm.lora.utils._all_lora_classes.add(
|
||||
AscendMergedQKVParallelLinearWithLoRA)
|
||||
Reference in New Issue
Block a user