[Hardware][openvino] is_openvino --> current_platform.is_openvino (#9716)
This commit is contained in:
@@ -13,12 +13,12 @@ from vllm.config import (CacheConfig, DeviceConfig, LoadConfig, LoRAConfig,
|
||||
from vllm.distributed import (broadcast_tensor_dict,
|
||||
ensure_model_parallel_initialized,
|
||||
init_distributed_environment)
|
||||
from vllm.executor.openvino_executor import is_openvino_cpu
|
||||
from vllm.inputs import INPUT_REGISTRY
|
||||
from vllm.logger import init_logger
|
||||
from vllm.model_executor import set_random_seed
|
||||
from vllm.model_executor.layers.sampler import SamplerOutput
|
||||
from vllm.multimodal import MULTIMODAL_REGISTRY
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.sampling_params import SamplingParams
|
||||
from vllm.sequence import ExecuteModelRequest, SequenceGroupMetadata
|
||||
from vllm.worker.openvino_model_runner import OpenVINOModelRunner
|
||||
@@ -99,7 +99,7 @@ class OpenVINOCacheEngine:
|
||||
num_blocks, self.block_size, self.num_kv_heads, self.head_size)[1:]
|
||||
kv_cache: List[Tuple[ov.Tensor, ov.Tensor]] = []
|
||||
|
||||
if is_openvino_cpu():
|
||||
if current_platform.is_openvino_cpu():
|
||||
for _ in range(self.num_layers):
|
||||
key_blocks = ov.Tensor(self.cache_config.cache_dtype,
|
||||
k_block_shape)
|
||||
@@ -141,7 +141,7 @@ class OpenVINOCacheEngine:
|
||||
if num_blocks == 0:
|
||||
return swap_cache
|
||||
|
||||
assert not is_openvino_cpu(), \
|
||||
assert not current_platform.is_openvino_cpu(), \
|
||||
"CPU device isn't supposed to have swap cache"
|
||||
|
||||
# Update key_cache shape:
|
||||
@@ -285,7 +285,7 @@ class OpenVINOWorker(LoraNotSupportedWorkerBase):
|
||||
cache_block_size = self.get_cache_block_size_bytes()
|
||||
kvcache_space_bytes = self.cache_config.openvino_kvcache_space_bytes
|
||||
|
||||
if is_openvino_cpu():
|
||||
if current_platform.is_openvino_cpu():
|
||||
num_device_blocks = int(kvcache_space_bytes // cache_block_size)
|
||||
num_swap_blocks = 0
|
||||
else:
|
||||
@@ -322,7 +322,7 @@ class OpenVINOWorker(LoraNotSupportedWorkerBase):
|
||||
num_device_blocks = num_gpu_blocks
|
||||
num_swap_blocks = num_cpu_blocks
|
||||
|
||||
if is_openvino_cpu():
|
||||
if current_platform.is_openvino_cpu():
|
||||
assert (num_swap_blocks == 0
|
||||
), f"{type(self)} does not support swappable cache for CPU"
|
||||
|
||||
@@ -366,7 +366,7 @@ class OpenVINOWorker(LoraNotSupportedWorkerBase):
|
||||
assert self.kv_cache is not None
|
||||
|
||||
# Populate the cache to warmup the memory
|
||||
if is_openvino_cpu():
|
||||
if current_platform.is_openvino_cpu():
|
||||
for key_cache, value_cache in self.kv_cache:
|
||||
key_cache.data[:] = 0
|
||||
value_cache.data[:] = 0
|
||||
@@ -414,7 +414,7 @@ class OpenVINOWorker(LoraNotSupportedWorkerBase):
|
||||
blocks_to_swap_in = data["blocks_to_swap_in"]
|
||||
blocks_to_swap_out = data["blocks_to_swap_out"]
|
||||
|
||||
if is_openvino_cpu():
|
||||
if current_platform.is_openvino_cpu():
|
||||
assert len(execute_model_req.blocks_to_swap_in) == 0
|
||||
assert len(execute_model_req.blocks_to_swap_out) == 0
|
||||
else:
|
||||
@@ -466,7 +466,7 @@ class OpenVINOWorker(LoraNotSupportedWorkerBase):
|
||||
def profile_run(self) -> int:
|
||||
ov_device = envs.VLLM_OPENVINO_DEVICE
|
||||
|
||||
assert not is_openvino_cpu(), \
|
||||
assert not current_platform.is_openvino_cpu(), \
|
||||
"CPU device isn't supposed to use profile run."
|
||||
|
||||
import openvino.properties.device as device
|
||||
|
||||
Reference in New Issue
Block a user