你说的算子开源是指哪些算子？

你说的算子开源是指哪些算子？

@yuhaiyan

aclnnInplaceFfnWorkerScheduler、aclnnInplaceAttentionWorkerScheduler这两AI CPU算子；以及aclnnFfnWorkerBatching和aclnnAttentionWorkerCombine这AI Core算子

你说的算子开源是指哪些算子？

@yuhaiyan

aclnnInplaceFfnWorkerScheduler、aclnnInplaceAttentionWorkerScheduler这两AI CPU算子；以及aclnnFfnWorkerBatching和aclnnAttentionWorkerCombine这AI Core算子

@qq_43728482

这些源码的开源情况，可以到cann仓https://gitcode.com/cann/ops-nn/issues 提相关issue。

你说的算子开源是指哪些算子？

@yuhaiyan

aclnnInplaceFfnWorkerScheduler、aclnnInplaceAttentionWorkerScheduler这两AI CPU算子；以及aclnnFfnWorkerBatching和aclnnAttentionWorkerCombine这AI Core算子

@qq_43728482

这些源码的开源情况，可以到cann仓https://gitcode.com/cann/ops-nn/issues 提相关issue。

@yuhaiyan

嗯嗯，行的；但是单测报错能不能先帮忙解决下

上述用例确实有报错，但是你的报错应该是环境问题，你本地单独执行"python -c 'import torch; import torch_npu; torch_npu.npu.synchronize()'" 试下。

因为我加了同步，所以错误报在同步这里，实际应该是算子调用出错了？或者是不是因为cann-ops-8.5.0出包时没有带上这个算子

torch_npu.npu_attention_worker_combine(schedule_context_npu, expert_scales_npu, layer_id_npu, H, token_dtype=torch.half, need_schedule=0)
torch_npu.npu.synchronize()

那可以再执行下上述用例，看是否还报torch_npu.npu.synchronize()错误

那可以再执行下上述用例，看是否还报torch_npu.npu.synchronize()错误

@yuhaiyan

python3 -c 'import torch; import torch_npu; torch_npu.npu.synchronize()
python3 test_npu_attention_worker_combine.py
/usr/local/conda/lib/python3.11/site-packages/torch_npu/utils/storage.py:43: UserWarning: TypedStorage is deprecated. It will be removed in the future and UntypedStorage will be the only storage class. This should only matter to you if you are using storages directly.  To access UntypedStorage directly, use tensor.untyped_storage() instead of tensor.storage()
  if self.device.type != 'cpu':
/workdir/pytorch/third_party/op-plugin/test/test_custom_ops/test_npu_attention_worker_combine.py:216: UserWarning: The given buffer is not writable, and PyTorch does not support non-writable tensors. This means you can write to the underlying (supposedly non-writable) buffer using the tensor. You may want to copy the buffer to protect its data or make it writable before converting it to a tensor. This type of warning will be suppressed for the rest of this program. (Triggered internally at /pytorch/torch/csrc/utils/tensor_new.cpp:1581.)
  schedule_context_tensor = torch.frombuffer(byte_data, dtype=torch.int8)
E
======================================================================
ERROR: test_npu_attention_worker_combine_001 (__main__.TestAttentionWorkerCombine.test_npu_attention_worker_combine_001)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/workdir/pytorch/third_party/op-plugin/test/test_custom_ops/test_npu_attention_worker_combine.py", line 225, in test_npu_attention_worker_combine_001
    torch_npu.npu.synchronize()
  File "/usr/local/conda/lib/python3.11/site-packages/torch_npu/npu/utils.py", line 72, in synchronize
    return torch_npu._C._npu_synchronize()
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: npuSynchronizeDevice:build/CMakeFiles/torch_npu.dir/compiler_depend.ts:564 NPU function error: SUSPECT REMOTE ERROR, error code is 507057
[ERROR] 2026-01-23-18:30:50 (PID:657932, Device:0, RankID:-1) ERR00100 PTA call acl api failed

----------------------------------------------------------------------
Ran 1 test in 3.196s

FAILED (errors=1)

那可以再执行下上述用例，看是否还报torch_npu.npu.synchronize()错误

@yuhaiyan

hello，请问有啥解法吗

我看原来的测试用例没有这个torch_npu.npu.synchronize()，这是你手动添加的吗

我看原来的测试用例没有这个torch_npu.npu.synchronize()，这是你手动添加的吗

@yuhaiyan

是的，如果不加会在self.assertRtolEqual(y_cpu, y_npu)里面报错，我理解就是算子调用有问题；请问您这边不能复现？

问题已解决。https://gitcode.com/Ascend/op-plugin/pull/4140