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

[Feat]support sequence parallelism by pass for VL models (#5632)

Browse Source
This commit is contained in:
realliujiaxu
2026-02-27 08:27:41 +08:00
committed by GitHub
parent ed175d6d92
commit 5def28dcd3
22 changed files with 460 additions and 101 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm_ascend/ops/linear.py
View File

@@ -40,7 +40,7 @@ from vllm.model_executor.layers.quantization.base_config import QuantizationConf
from vllm.model_executor.utils import set_weight_attrs
from vllm_ascend.ops.linear_op import get_parallel_op, get_replicated_op
from vllm_ascend.utils import enable_sp, maybe_trans_nz
from vllm_ascend.utils import enable_flash_comm_v1, maybe_trans_nz
class AscendUnquantizedLinearMethod(UnquantizedLinearMethod):
@@ -240,7 +240,7 @@ class AscendRowParallelLinear(RowParallelLinear):
disable_tp: bool = False,
):
# TODO(kunpengW-code): Specifying the prefix in linear layers of some models in the vLLM.
if enable_sp():
if enable_flash_comm_v1():
compilation_config = get_current_vllm_config().compilation_config
unique_prefix = prefix
if prefix in compilation_config.static_forward_context: