chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[NPU][1/N] NPU basic functions refactor and new modelslim quant type (#13359)

Browse Source
This commit is contained in:
Even Zhou
2025-12-04 16:15:31 +08:00
committed by GitHub
parent d6c490192d
commit 894c0dc57c
43 changed files with 2500 additions and 2058 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
python/sglang/srt/managers/schedule_batch.py
View File

@@ -83,11 +83,8 @@ from sglang.srt.sampling.sampling_batch_info import SamplingBatchInfo
from sglang.srt.sampling.sampling_params import SamplingParams
from sglang.srt.server_args import ServerArgs, get_global_server_args
from sglang.srt.utils import flatten_nested_list
from sglang.srt.utils.common import is_npu
from sglang.srt.utils.cuda_ipc_transport_utils import CudaIpcTensorTransportProxy
_is_npu = is_npu()
if TYPE_CHECKING:
from sglang.srt.configs.model_config import ModelConfig
from sglang.srt.speculative.eagle_info import EagleDraftInput
@@ -1152,10 +1149,7 @@ class ScheduleBatch(ScheduleBatchDisaggregationDecodeMixin):
has_grammar: bool = False
# Device
if not _is_npu:
device: str = "cuda"
else:
device: str = "npu"
device: str = "cuda"
# Speculative decoding
spec_algorithm: SpeculativeAlgorithm = None