驱动：24.1.rc2

CANN:8.0.0.beta1

torch_npu:2.4.0.post2

使用vllm测试部署Qwen2.5-0.5B。报错信息：

[rank0]: RuntimeError: call aclnnPromptFlashAttentionV3 failed, detail:EZ1001: [PID: 1182000] 2025-04-09-15:10:57.206.836 PromptFlashAttention LaunchAicore failed. 

[rank0]:         TraceBack (most recent call last): 

[rank0]:         ascend310p platform do not support batch size(256) more than 128.[FUNC:RunBigKernelTilingWithParams][FILE:prompt_flash_attention_tiling.cpp][LINE:2600] 

[rank0]:         Tiling failed 

[rank0]:         Tiling Failed. 

[rank0]:         Kernel GetWorkspace failed. opType: 5 

[rank0]:         PromptFlashAttention LaunchAicore failed. 

[rank0]: [ERROR] 2025-04-09-15:10:57 (PID:1182000, Device:0, RankID:-1) ERR01100 OPS call acl api failed