已合并
修改文档问题 #3575
limingheng创建于 18 天前
修改文档问题 #3575
已合并
共 3 个文件变更+7-38
| @@ -501,7 +501,7 @@ config.export.experimental.enable_lite_export = True | |||
| 501 | 因此提供一种方式，在torchair中屏蔽原生pytorch中的某些decomposition。 | 501 | 因此提供一种方式，在torchair中屏蔽原生pytorch中的某些decomposition。 |
| 502 | **decomposition实现原理：** | 502 | **decomposition实现原理：** |
| 503 | 503 | ||
| 504 | - 1、pytorch使用装饰器[register_decomposition](https://github.com/pytorch/pytorch/blob/v2.1.0/torch/_decomp/__init__.py#L99)将特定的atenIR拆解为被装饰函数所调用的小算子。 | 504 | + 1、pytorch使用装饰器[register_decomposition](https://github.com/pytorch/pytorch/blob/main/torch/_decomp/__init__.py#L99)将特定的atenIR拆解为被装饰函数所调用的小算子。 |
| 505 | 2、[特殊场景]某些atenIR存在特殊的dispatch key，如DispatchKey.CompositeImplicitAutograd。该dispatch key对应的C++ | 505 | 2、[特殊场景]某些atenIR存在特殊的dispatch key，如DispatchKey.CompositeImplicitAutograd。该dispatch key对应的C++ |
| 506 | 实现，是无法进行dynamic=True时torch.compile模式下的sym符号推导。此时，通过aten.upsample_nearest2d.vec.py_impl装饰器 | 506 | 实现，是无法进行dynamic=True时torch.compile模式下的sym符号推导。此时，通过aten.upsample_nearest2d.vec.py_impl装饰器 |
| 507 | 将该dispatch key注册python实现解决该问题。 | 507 | 将该dispatch key注册python实现解决该问题。 |
| @@ -1,36 +0,0 @@ | |||
| 1 | -# torchair | ||
| 2 | - | ||
| 3 | -## Description | ||
| 4 | - | ||
| 5 | -{**When you're done, you can delete the content in this README and update the file with details for others getting started with your repository**} | ||
| 6 | - | ||
| 7 | -### Software Architecture | ||
| 8 | - | ||
| 9 | -Software architecture description | ||
| 10 | - | ||
| 11 | -### Installation | ||
| 12 | - | ||
| 13 | -1. xxxx | ||
| 14 | -2. xxxx | ||
| 15 | -3. xxxx | ||
| 16 | - | ||
| 17 | -### Instructions | ||
| 18 | - | ||
| 19 | -1. xxxx | ||
| 20 | -2. xxxx | ||
| 21 | -3. xxxx | ||
| 22 | - | ||
| 23 | -### Contribution | ||
| 24 | - | ||
| 25 | -1. Fork the repository | ||
| 26 | -2. Create Feat_xxx branch | ||
| 27 | -3. Commit your code | ||
| 28 | -4. Create Pull Request | ||
| 29 | - | ||
| 30 | -### GitCode Feature | ||
| 31 | - | ||
| 32 | -1. You can use Readme\_XXX.md to support different languages, such as Readme\_en.md, Readme\_zh.md | ||
| 33 | -2. GitCode blog [blog.gitcode.com](https://blog.gitcode.com) | ||
| 34 | -3. Explore open source project [https://gitcode.com/explore](https://gitcode.com/explore) | ||
| 35 | -4. The most valuable open source project [G-Star](https://gitcode.com/G-Star) | ||
| 36 | -5. The manual of GitCode [https://docs.gitcode.com/docs/help/home/](https://docs.gitcode.com/docs/help/home/) | ||
| @@ -8,13 +8,18 @@ class CompilerConfig(NpuBaseConfig): | |||
| 8 | 8 | ||
| 9 | def __init__(self): | 9 | def __init__(self): |
| 10 | self.debug = _DebugConfig() | 10 | self.debug = _DebugConfig() |
| 11 | + self.aoe_config = _AoeConfig() | ||
| 11 | self.export = _ExportConfig() | 12 | self.export = _ExportConfig() |
| 12 | self.dump_config = _DataDumpConfig() | 13 | self.dump_config = _DataDumpConfig() |
| 13 | self.fusion_config = _FusionConfig() | 14 | self.fusion_config = _FusionConfig() |
| 14 | self.experimental_config = _ExperimentalConfig() | 15 | self.experimental_config = _ExperimentalConfig() |
| 15 | self.inference_config = _InferenceConfig() | 16 | self.inference_config = _InferenceConfig() |
| 16 | self.ge_config = _GEConfig() | 17 | self.ge_config = _GEConfig() |
| 17 | - self.mode = OptionValue("max-autotune", ["max-autotune"]) | 18 | + self.aclgraph_config = _AclGraphConfig() |
| 19 | + self.npugraphex_config = _NpuGraphExConfig() | ||
| 20 | + self.mode = OptionValue("max-autotune", ["max-autotune", "reduce-overhead", "npugraph_ex"]) | ||
| 21 | + self.post_grad_custom_pre_pass = CallableValue(None) | ||
| 22 | + self.post_grad_custom_post_pass = CallableValue(None) | ||
| 18 | 23 | ||
| 19 | super(CompilerConfig, self).__init__() | 24 | super(CompilerConfig, self).__init__() |
| 20 | ``` | 25 | ``` |