EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

[Misc][V0 Deprecation] Remove Cache Engine Used for V0 Worker (#1878)

Browse Source 
### What this PR does / why we need it?
This PR is a part of
https://github.com/vllm-project/vllm-ascend/issues/1620.

- vLLM version: v0.9.2
- vLLM main:
5895afd780

---------

Signed-off-by: shen-shanshan <467638484@qq.com>
This commit is contained in:
Shanshan Shen
2025-07-19 09:42:32 +08:00
committed by GitHub
parent 53d2ea3789
commit ab68d31a24
2 changed files with 0 additions and 100 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
vllm_ascend/worker/__init__.py
View File

@@ -1,17 +0,0 @@
#
# 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.
#
import vllm_ascend.worker.cache_engine # noqa

83
vllm_ascend/worker/cache_engine.py
View File

@@ -1,83 +0,0 @@
#
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
# This file is a part of the vllm-ascend project.
# Adapted from vllm-project/vllm/vllm/worker/model_runner.py
# Copyright 2023 The vLLM team.
#
# 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, List
import torch
from vllm.utils import is_pin_memory_available
from vllm.worker.cache_engine import CacheEngine
from vllm_ascend.ascend_config import get_ascend_config
def allocate_kv_cache(
self,
num_blocks: int,
device: str,
) -> List[Any]:
"""Allocates KV cache on the specified device."""
kv_cache_shape = self.attn_backend.get_kv_cache_shape(
num_blocks, self.block_size, self.num_kv_heads, self.head_size)
pin_memory = is_pin_memory_available() if device == "cpu" else False
kv_cache: List[Any] = []
ascend_config = get_ascend_config()
if ascend_config.torchair_graph_config.enabled:
# Align entries so they are 256 byte aligned for better performance
# Primarily targets MLA as this typically only ends up having entries
# be 128 byte aligned.
alloc_shape = kv_cache_shape
for _ in range(self.num_attention_layers):
# null block in CpuGpuBlockAllocator requires at least that
# block to be zeroed-out.
# We zero-out everything for simplicity.
layer_kv_cache_nope = torch.zeros(
alloc_shape[:-1] +
(self.model_config.hf_text_config.kv_lora_rank, ),
dtype=self.dtype,
pin_memory=pin_memory,
device=device)
layer_kv_cache_pe = torch.zeros(
alloc_shape[:-1] +
(self.model_config.hf_text_config.qk_rope_head_dim, ),
dtype=self.dtype,
pin_memory=pin_memory,
device=device)
# view back to (TOTAL_PAGES, PAGE_SIZE, entry_shape...) for cases
# when entry_shape is higher than 1D
kv_cache.append((layer_kv_cache_nope, layer_kv_cache_pe))
else:
for _ in range(self.num_attention_layers):
# null block in CpuGpuBlockAllocator requires at least that
# block to be zeroed-out.
# We zero-out everything for simplicity.
layer_kv_cache = torch.zeros(kv_cache_shape,
dtype=self.dtype,
pin_memory=pin_memory,
device=device)
# view back to (TOTAL_PAGES, PAGE_SIZE, entry_shape...) for cases
# when entry_shape is higher than 1D
kv_cache.append(layer_kv_cache)
return kv_cache
CacheEngine._allocate_kv_cache = allocate_kv_cache