master

ascend-robot[torch-master]修复profiler采集内存流ID数据没有aclrtStreamGetId接口兼容性问题

85d2a49b创建于 2025年12月9日历史提交

文件	最后提交记录	最后更新时间
_afd	!23988 add module afd Merge pull request !23988 from 李宁/master	9 个月前
_inductor	wrap_triton Co-authored-by: 1479518308<cuiduo1@huawei.com> # message auto-generated for no-merge-commit merge: !26741 merge master into master wrap_triton Created-by: cuiduo Commit-by: 1479518308 Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > /kind task **What does this PR do / why do we need it**:Add wrap_triton function adaptation. Add triton operator adaptation modules in the files torch._inductor.lowering.py and torch._inductor.lowering_fx.py. Add triton_kernel_wrapper_mutation to the whitelist in the file torch._inductor.lowering_op_list.py. **Which issue(s) this PR fixes**: <!-- *Automatically closes linked issue when PR is merged. Usage: Fixes #<issue number>, or Fixes (paste link of issue). --> Fixes #Fixes the issue where custom Triton operators under the wrap_triton function fail to function as expected. See merge request: Ascend/pytorch!26741	6 个月前
_logging	[SHMEM] support npu shmem Co-authored-by: wangchao430<wangchao430@huawei.com> # message auto-generated for no-merge-commit merge: !26027 merge v2.99.0_shmem1 into master [SHMEM] support npu shmem Created-by: wangchao430 Commit-by: wangchao430 Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature **What does this PR do / why do we need it**: **Which issue(s) this PR fixes**: <!-- *Automatically closes linked issue when PR is merged. Usage: Fixes #<issue number>, or Fixes (paste link of issue). --> Fixes # **Special notes for your reviewers**: See merge request: Ascend/pytorch!26027	6 个月前
asd	[SilentCheck] fix with DTensor of grad Co-authored-by: wangchao430<wangchao430@huawei.com> # message auto-generated for no-merge-commit merge: merge v2.99.0_checkfix1 into master [SilentCheck] fix with DTensor of grad Created-by: wangchao430 Commit-by: wangchao430 Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature **What does this PR do / why do we need it**: **Which issue(s) this PR fixes**: <!-- *Automatically closes linked issue when PR is merged. Usage: Fixes #<issue number>, or Fixes (paste link of issue). --> Fixes # **Special notes for your reviewers**: See merge request: Ascend/pytorch!25008	8 个月前
contrib	Adapt parameter "device_id" for init_process_group Co-authored-by: l30044004<louyujing@huawei.com> # message auto-generated for no-merge-commit merge: !27487 merge master into master Adapt parameter "device_id" for init_process_group Created-by: louyujing Commit-by: l30044004 Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature **What does this PR do / why do we need it**: **Which issue(s) this PR fixes**: <!-- *Automatically closes linked issue when PR is merged. Usage: Fixes #<issue number>, or Fixes (paste link of issue). --> Fixes # torch.distributed.init_process_group可传入参数device_id，如果传入的device_id包含cuda，在调用torch.distributed.barrier()时会获取process_group中包含cuda的device_id，导致在npu上运行报错； 因此，给torch.distributed.init_process_group添加装饰器，将参数device_id中的cuda替换为npu。 **Special notes for your reviewers**: See merge request: Ascend/pytorch!27487	5 个月前
csrc	[torch-master]修复profiler采集内存流ID数据没有aclrtStreamGetId接口兼容性问题 Co-authored-by: yu-liang-bin<y1416490440@163.com> # message auto-generated for no-merge-commit merge: !27671 merge bug_fix_memory into master [torch-master]修复profiler采集内存流ID数据没有aclrtStreamGetId接口兼容性问题 Created-by: yu_liangbin Commit-by: yu-liang-bin Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature **What does this PR do / why do we need it**: profiler采集内存数据时，会通过aclrtStreamGetId接口采集流ID，如果没有这个接口，会打断训练 **Which issue(s) this PR fixes**: 在调用aclrtStreamGetId接口采集流ID数据前增加aclrtStreamGetId接口检查函数 <!-- *Automatically closes linked issue when PR is merged. Usage: Fixes #<issue number>, or Fixes (paste link of issue). --> Fixes # **Special notes for your reviewers**: See merge request: Ascend/pytorch!27671	5 个月前
distributed	add register sharding for gather and swiglu on master Co-authored-by: zhangguoguang<zhangguoguang2@huawei.com> # message auto-generated for no-merge-commit merge: !27120 merge register_sharding/master into master add register sharding for gather and swiglu on master Created-by: zhangguoguang Commit-by: zhangguoguang Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature **What does this PR do / why do we need it**: **Which issue(s) this PR fixes**: <!-- *Automatically closes linked issue when PR is merged. Usage: Fixes #<issue number>, or Fixes (paste link of issue). --> Fixes # **Special notes for your reviewers**: See merge request: Ascend/pytorch!27120	6 个月前
dynamo	skip npu sync func to master Co-authored-by: lihui<lihui488@huawei.com> # message auto-generated for no-merge-commit merge: !27578 merge master into master skip npu sync func to master Created-by: lihui488 Commit-by: lihui Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature **What does this PR do / why do we need it**: **Which issue(s) this PR fixes**: <!-- *Automatically closes linked issue when PR is merged. Usage: Fixes #<issue number>, or Fixes (paste link of issue). --> Fixes # **Special notes for your reviewers**: See merge request: Ascend/pytorch!27578	5 个月前
jit	!9416 Fixed for the public API. Merge pull request !9416 from yuhaiyan/master-dev2	2 年前
multiprocessing	[IPC]: fix with meta tensor Co-authored-by: wangchao430<wangchao430@huawei.com> # message auto-generated for no-merge-commit merge: !25996 merge v2.99.0_ipcmeta into master [IPC]: fix with meta tensor Created-by: wangchao430 Commit-by: wangchao430 Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature **What does this PR do / why do we need it**: **Which issue(s) this PR fixes**: <!-- *Automatically closes linked issue when PR is merged. Usage: Fixes #<issue number>, or Fixes (paste link of issue). --> Fixes # **Special notes for your reviewers**: See merge request: Ascend/pytorch!25996	7 个月前
npu	[Feature] NPUGraph capture support compiled model Co-authored-by: XDaoHong<xudaohong@huawei.com> # message auto-generated for no-merge-commit merge: !26816 merge master into master [Feature] NPUGraph capture support compiled model Created-by: XDaoHong Commit-by: XDaoHong Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > feature **What does this PR do / why do we need it**: 问题背景： torch.npu.graph(auto_dispatch_capture=True)场景，如果captruetorch.compile后的model，compile会被跳过失效。 问题原因: auto_dispatch_capture=True中会使用class _GraphDispatchMode(torch.utils._python_dispatch.TorchDispatchMode)做update算子的dispatch拦截；_GraphDispatchMode为用户自定义TorchDispatchMode，类方法def is_infra_mode(cls):默认返回False；torch.\_dynamo流程中会校验是否是用了non-infra的dispatch mode，使用的话会跳过dynamo逻辑，详见[pytorch源码](https://github.com/pytorch/pytorch/blob/v2.7.0/torch/_dynamo/convert_frame.py#L1390)。 测试用例： ```python import logging import torch import torch_npu torch._logging.set_logs(dynamo=logging.DEBUG, aot=logging.DEBUG, output_code=True, graph_code=True) class Model(torch.nn.Module): def __init__(self): super(Model, self).__init__() def forward(self, x): y = x + 1 return y model = Model().npu() model = torch.compile(model, fullgraph=True, backend="aot_eager", dynamic=False) static_input = torch.zeros((5, 5), device="npu") g = torch_npu.npu.NPUGraph() with torch.npu.graph(g, auto_dispatch_capture=True): res = model(static_input) x = torch.zeros((5, 5), device="npu") + 1 static_input.copy_(x) g.replay() print(res) ``` 关键日志： ```shell V1120 15:27:06.337000 1308894 site-packages/torch/_dynamo/convert_frame.py:1395] skipping: _wrapped_call_impl (reason: in skipfiles, file: lib/python3.11/site-packages/torch/nn/modules/module.py) V1120 15:27:06.338000 1308894 site-packages/torch/_dynamo/convert_frame.py:1395] skipping: _call_impl (reason: in skipfiles, file: lib/python3.11/site-packages/torch/nn/modules/module.py) V1120 15:27:06.338000 1308894 site-packages/torch/_dynamo/convert_frame.py:1395] skipping: forward (reason: non-infra torch dispatch mode present, this is not supported today in torch.compile, file:eager.py) ``` **Which issue(s) this PR fixes**: <!-- *Automatically closes linked issue when PR is merged. Usage: Fixes #<issue number>, or Fixes (paste link of issue). --> Fixes # **Special notes for your reviewers**: See merge request: Ascend/pytorch!26816	6 个月前
onnx	!22300 Add support for custom dtype Merge pull request !22300 from chuboning/master	11 个月前
optim	!13006 Fixed for the public APIs. Merge pull request !13006 from yuhaiyan/master-dev2	1 年前
profiler	Profiler: adapt to the worker process name change of vllm Co-authored-by: z30043230<zhaiyibo@h-partners.com> # message auto-generated for no-merge-commit merge: !27535 merge main into master Profiler: adapt to the worker process name change of vllm Created-by: zyb_230 Commit-by: z30043230 Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature /kind bug **What does this PR do / why do we need it**: vllm 11.0实际业务进程名字为EngineCore_DPxx然后代码会拦截，最后不会创建共享内存，无法使用msmonitor功能 **Which issue(s) this PR fixes**: <!-- *Automatically closes linked issue when PR is merged. Usage: Fixes #<issue number>, or Fixes (paste link of issue). --> Fixes # 去掉进程名为EngineCore就return False的代码  已和测试一起验证修改没问题 **Special notes for your reviewers**: See merge request: Ascend/pytorch!27535	5 个月前
testing	register strategies for npu custom ops Co-authored-by: jizewei<jizewei@huawei.com> # message auto-generated for no-merge-commit merge: !27012 merge master_register_strategy into master register strategy for npu custom ops Created-by: jizewei Commit-by: jizewei Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature /kind feature **What does this PR do / why do we need it**: register sharding strategies for npu custom ops to support DTensor: 1. npu_add_rms_norm 2. npu_moe_token_permute, npu_moe_token_permute_grad 3. npu_moe_token_unpermute, npu_moe_token_unpermute_grad 4. npu_rotary_mul, npu_rotary_mul_backward **Which issue(s) this PR fixes**: <!-- *Automatically closes linked issue when PR is merged. Usage: Fixes #<issue number>, or Fixes (paste link of issue). --> Fixes # **Special notes for your reviewers**: See merge request: Ascend/pytorch!27012	6 个月前
utils	fix bug of storage.cpu Co-authored-by: huangyunlong2022<huangyunlong4@h-partners.com> # message auto-generated for no-merge-commit merge: !27555 merge 2.10st into master fix bug of storage.cpu Created-by: huangyunlong2022 Commit-by: huangyunlong2022 Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature **What does this PR do / why do we need it**: **Which issue(s) this PR fixes**: <!-- *Automatically closes linked issue when PR is merged. Usage: Fixes #<issue number>, or Fixes (paste link of issue). --> Fixes # **Special notes for your reviewers**: See merge request: Ascend/pytorch!27555	5 个月前
__init__.py	[950] Add 950 support Co-authored-by: chuboning<chuboning1@huawei.com> Co-authored-by: lilongqianxi<lilongqianxi@h-partners.com> Co-authored-by: 路有兵<luyoubing@huawei.com> # message auto-generated for no-merge-commit merge: !26229 merge master into master [950] Add 950 support Created-by: chuboning Commit-by: chuboning;路有兵;lilongqianxi Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> **What type of PR is this?** > Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line: > > /kind bug > /kind task > /kind feature **What does this PR do / why do we need it**: **Which issue(s) this PR fixes**: <!-- *Automatically closes linked issue when PR is merged. Usage: Fixes #<issue number>, or Fixes (paste link of issue). --> Fixes # **Special notes for your reviewers**: See merge request: Ascend/pytorch!26229	6 个月前
acl.json	!24209 回退 'Pull Request !23967 : lazy setdevice' Merge pull request !24209 from 郭光浩/revert-merge-23967-master	9 个月前
acl_default.json	!24312 lazy setdevice Merge pull request !24312 from huangyunlong/2.9ls	8 个月前