modify docs-op-plugin-AtomGit

已合并

modify docs #5810

molly123321创建于 5 天前从molly123321/op-plugin： 26.1.0 合入到 Ascend/op-plugin： 26.1.0

modify docs #5810

已合并

molly123321创建于 5 天前

从molly123321/op-plugin 26.1.0合入到Ascend/op-plugin 26.1.0

共 11 个文件变更+19-25

Mdocs/en/custom_APIs/torch_npu-npu/(beta)torch_npu-npu-stress_detect.md+2-2

	@@ -35,12 +35,12 @@ torch_npu.npu.stress_detect(detect_type="aic")
35	- `2`: Online precision detection failed due to a hardware fault.	35	- `2`: Online precision detection failed due to a hardware fault.
36		36
37	- The following error indicates a voltage recovery failure, which requires manual voltage restoration (by referring to section "Comprehensive Check" in CANN Troubleshooting) or a system reboot.	37	- The following error indicates a voltage recovery failure, which requires manual voltage restoration (by referring to section "Comprehensive Check" in CANN Troubleshooting) or a system reboot.
38	-<!-- in [CANN Troubleshooting](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/latest/maintenref/troubleshooting/troubleshooting_0505.html)) or a system reboot. -->
39		38
40	```shell	39	```shell
41	Stress detect error. Error code is 574007. Error message is Voltage recovery failed.	40	Stress detect error. Error code is 574007. Error message is Voltage recovery failed.
42	```	41	```
43	-	42	+
		43	+<!-- in [CANN Troubleshooting](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/latest/maintenref/troubleshooting/troubleshooting_0505.html)) or a system reboot. -->
44	## Constraints	44	## Constraints
45		45
46	- Online precision detection requires modifying the model training script. You are advised to call this API before training starts, after training ends, or between two steps. Reserve 10 GB of memory for the stress test interface.	46	- Online precision detection requires modifying the model training script. You are advised to call this API before training starts, after training ends, or between two steps. Reserve 10 GB of memory for the stress test interface.

Mdocs/en/custom_APIs/torch_npu/torch_npu-npu_alltoallv_quant_gmm.md+1-1

	@@ -133,7 +133,7 @@ torch_npu.npu_alltoallv_quant_gmm(gmm_x, gmm_weight, gmm_x_scale, gmm_weight_sca
133	- When the conditions for resetting the values are met, if `gmm_x_scale`, `mm_x_scale`, and `mm_weight_scale` are all 3D tensors, `gmm_weight_scale` is a 4D tensor, and all data types are `float8_e8m0fnu`, the inferred combination of `[groupSizeM, groupSizeN, groupSizeK]` is `[1, 1, 32]`.	133	- When the conditions for resetting the values are met, if `gmm_x_scale`, `mm_x_scale`, and `mm_weight_scale` are all 3D tensors, `gmm_weight_scale` is a 4D tensor, and all data types are `float8_e8m0fnu`, the inferred combination of `[groupSizeM, groupSizeN, groupSizeK]` is `[1, 1, 32]`.
134	- When all output tensors (`gmm_y`, `mm_y`, and `permute_out`) on a device are empty tensors, `torch.distributed.barrier()` must be called explicitly to ensure that the process on that device synchronizes and waits for other devices to complete communication and computation. If synchronization is not added, `AlltoAllv` communication will be blocked due to process desynchronization.	134	- When all output tensors (`gmm_y`, `mm_y`, and `permute_out`) on a device are empty tensors, `torch.distributed.barrier()` must be called explicitly to ensure that the process on that device synchronizes and waits for other devices to complete communication and computation. If synchronization is not added, `AlltoAllv` communication will be blocked due to process desynchronization.
135		135
136	-- - Detailed input and output data type constraints for each quantization mode are provided in the following tables.	136	+- Detailed input and output data type constraints for each quantization mode are provided in the following tables.
137		137
138	Table 1 Data type constraints for T-T quantization	138	Table 1 Data type constraints for T-T quantization
139		139

Mdocs/en/custom_APIs/torch_npu/torch_npu-npu_grouped_matmul_swiglu_quant_v2.md+2-2

	@@ -70,7 +70,7 @@
70		70
71	$C_{i,act}, gate_{i} = split(C_{i})$	71	$C_{i,act}, gate_{i} = split(C_{i})$
72		72
73	- $S_{i}=Swish(C_{i,act})\odot gate_{i}$，其中$Swish(x)=\frac{x}{1+e^{-x}}$	73	+ $S_{i}=Swish(C_{i,act})\odot gate_{i}$, where $Swish(x)=\frac{x}{1+e^{-x}}$
74	4. Quantize the output results.	74	4. Quantize the output results.
75		75
76	$Q\_scale_{i} = \frac{max(\|S_{i}\|)}{127}$	76	$Q\_scale_{i} = \frac{max(\|S_{i}\|)}{127}$
	@@ -123,7 +123,7 @@
123		123
124	$C_{i,act}, gate_{i} = split(C_{i})$	124	$C_{i,act}, gate_{i} = split(C_{i})$
125		125
126	- $S_{i}=Swish(C_{i,act})\odot gate_{i}$，其中$Swish(x)=\frac{x}{1+e^{-x}}$	126	+ $S_{i}=Swish(C_{i,act})\odot gate_{i}$, where $Swish(x)=\frac{x}{1+e^{-x}}$
127		127
128	2. Quantize the output:	128	2. Quantize the output:
129		129

Mdocs/en/custom_APIs/torch_npu/torch_npu-npu_mla_prolog_v2.md+1-1

	@@ -761,7 +761,7 @@ torch_npu.npu_mla_prolog_v2(token_x, weight_dq, weight_uq_qr, weight_uk, weight_
761	import torch_npu	761	import torch_npu
762	import math	762	import math
763	torch.npu.config.allow_internal_format = True	763	torch.npu.config.allow_internal_format = True
764	- # 生成随机数据, 并发送到npu	764	+ # Generate random data and send it to the NPU
765	B = 8	765	B = 8
766	He = 7168	766	He = 7168
767	Hcq = 1536	767	Hcq = 1536

Mdocs/en/custom_APIs/torch_npu/torch_npu-set_stream_limit.md+1-1

	@@ -53,7 +53,6 @@ No value is returned.
53		53
54	1. The Ascend PyTorch Profiler is used to collect performance data. This includes information about PyTorch layer operators, CANN layer operators, underlying NPU operators, and operator memory footprints.	54	1. The Ascend PyTorch Profiler is used to collect performance data. This includes information about PyTorch layer operators, CANN layer operators, underlying NPU operators, and operator memory footprints.
55	> Note: Ascend PyTorch Profiler is a performance analysis tool developed by CANN for the PyTorch framework. You can add an Ascend PyTorch Profiler interface (`torch_npu.profiler.profile` is recommended) to your PyTorch script to collect specified metrics. During model execution, performance data is collected synchronously. For details about its usage and the resulting files, see Ascend PyTorch Profiler.	55	> Note: Ascend PyTorch Profiler is a performance analysis tool developed by CANN for the PyTorch framework. You can add an Ascend PyTorch Profiler interface (`torch_npu.profiler.profile` is recommended) to your PyTorch script to collect specified metrics. During model execution, performance data is collected synchronously. For details about its usage and the resulting files, see Ascend PyTorch Profiler.
56	- see <a href="https://www.hiascend.com/document/detail/en/mindstudio/latest/TITools/ascend_pytorch_profiler/docs/en/ascend_pytorch_profiler/ascend_pytorch_profiler_user_guide.md">Ascend PyTorch Profiler</a>
57		56
58	```python	57	```python
59	>>> import torch	58	>>> import torch
	@@ -108,3 +107,4 @@ No value is returned.
108		107
109	For single-operator core control, you can check the `kernel_details.csv` file and check the BlockDim column (the number of cores used for operator computation). If the `BlockDim` value is less than or equal to the core limit configured by the user, the core control is successful.	108	For single-operator core control, you can check the `kernel_details.csv` file and check the BlockDim column (the number of cores used for operator computation). If the `BlockDim` value is less than or equal to the core limit configured by the user, the core control is successful.
110	In this example, the `BlockDim` values of the SwiGlu operator are `22` and `23`, which comply with the core control logic.	109	In this example, the `BlockDim` values of the SwiGlu operator are `22` and `23`, which comply with the core control logic.
		110	+ <!-- see <a href="https://www.hiascend.com/document/detail/en/mindstudio/latest/TITools/ascend_pytorch_profiler/docs/en/ascend_pytorch_profiler/ascend_pytorch_profiler_user_guide.md">Ascend PyTorch Profiler</a> -->

Mexamples/cpp_extension_asc/README_en.md+2-3

	@@ -11,9 +11,9 @@ This document describes the complete adaptation development workflow for calling
11	Before getting started, ensure that you have completed the installation of the following environments:	11	Before getting started, ensure that you have completed the installation of the following environments:
12		12
13	1. Install the NPU driver, firmware, and CANN software (Toolkit, ops, and NNAL) by referring to CANN Software Installation.	13	1. Install the NPU driver, firmware, and CANN software (Toolkit, ops, and NNAL) by referring to CANN Software Installation.
14	-<!-- [CANN Software Installation](https://www.hiascend.com/document/detail/en/CANNCommunityStructureEdition/910/softwareinst/instg/instg_0000.html?OS=openEuler&InstallType=netyum). -->
15		14
16	2. Install the PyTorch framework by referring to [Software Installation](https://gitcode.com/Ascend/pytorch/blob/v2.7.1-26.1.0/docs/en/installation_guide/installation_description.md).	15	2. Install the PyTorch framework by referring to [Software Installation](https://gitcode.com/Ascend/pytorch/blob/v2.7.1-26.1.0/docs/en/installation_guide/installation_description.md).
		16	+<!-- [CANN Software Installation](https://www.hiascend.com/document/detail/en/CANNCommunityStructureEdition/910/softwareinst/instg/instg_0000.html?OS=openEuler&InstallType=netyum). -->
17		17
18	### Adaptation File Structure	18	### Adaptation File Structure
19		19
	@@ -190,7 +190,6 @@ The following steps use the `Add` operator as an example.
190	After completing the operator adaptation development, you can call the custom operator through C++ extensions.	190	After completing the operator adaptation development, you can call the custom operator through C++ extensions.
191		191
192	1. Complete the creation, development, compilation, and deployment workflow for the custom operator project. For details, see CANN Ascend C Operator Development.	192	1. Complete the creation, development, compilation, and deployment workflow for the custom operator project. For details, see CANN Ascend C Operator Development.
193	-<!-- see [CANN Ascend C Operator Development](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/latest/programug/Ascendcopdevg/atlas_ascendc_map_10_0002.html) -->
194		193
195	2. Download the code sample.	194	2. Download the code sample.
196		195
	@@ -215,7 +214,7 @@ After completing the operator adaptation development, you can call the custom op
215	Ran xx tests in xx s	214	Ran xx tests in xx s
216	OK	215	OK
217	```	216	```
218	-	217	+<!-- see [CANN Ascend C Operator Development](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/latest/programug/Ascendcopdevg/atlas_ascendc_map_10_0002.html) -->
219	## Reference	218	## Reference
220		219
221	### `Add` Operator	220	### `Add` Operator

Mexamples/cpp_extension_base/README_en.md+2-3

	@@ -11,10 +11,9 @@ This document describes the complete adaptation development workflow for calling
11	Before getting started, ensure that you have completed the installation of the following environments:	11	Before getting started, ensure that you have completed the installation of the following environments:
12		12
13	1. Install the NPU driver, firmware, and CANN software (Toolkit, ops, and NNAL) by referring to CANN Software Installation.	13	1. Install the NPU driver, firmware, and CANN software (Toolkit, ops, and NNAL) by referring to CANN Software Installation.
14	-<!-- [CANN Software Installation](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/910/softwareinst/instg/instg_0000.html?OS=openEuler&InstallType=netyum) -->
15		14
16	2. Install the PyTorch framework by referring to [Installation Guide](https://gitcode.com/Ascend/pytorch/blob/v2.7.1-26.1.0/docs/en/installation_guide/installation_description.md).	15	2. Install the PyTorch framework by referring to [Installation Guide](https://gitcode.com/Ascend/pytorch/blob/v2.7.1-26.1.0/docs/en/installation_guide/installation_description.md).
17	-	16	+<!-- [CANN Software Installation](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/910/softwareinst/instg/instg_0000.html?OS=openEuler&InstallType=netyum) -->
18	### Adaptation File Structure	17	### Adaptation File Structure
19		18
20	```text	19	```text
	@@ -93,7 +92,6 @@ Before getting started, ensure that you have completed the installation of the f
93	After completing the operator adaptation development, you can call the custom operator through C++ extensions.	92	After completing the operator adaptation development, you can call the custom operator through C++ extensions.
94		93
95	1. Complete the creation, development, compilation, and deployment workflow for the custom operator project. For details, see CANN Ascend C Operator Development.	94	1. Complete the creation, development, compilation, and deployment workflow for the custom operator project. For details, see CANN Ascend C Operator Development.
96	-<!-- [CANN Ascend C Operator Development](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/latest/programug/Ascendcopdevg/atlas_ascendc_map_10_0002.html) -->
97		95
98	2. Download the code sample.	96	2. Download the code sample.
99		97
	@@ -118,3 +116,4 @@ After completing the operator adaptation development, you can call the custom op
118	Ran xx tests in xx s	116	Ran xx tests in xx s
119	OK	117	OK
120	```	118	```
		119	+<!-- [CANN Ascend C Operator Development](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/latest/programug/Ascendcopdevg/atlas_ascendc_map_10_0002.html) -->

Mexamples/cpp_extension_full/module/README_en.md+2-3

	@@ -11,10 +11,9 @@ This document describes how to adapt and call a single-operator API through Torc
11	Before getting started, ensure that you have completed the installation of the following environments:	11	Before getting started, ensure that you have completed the installation of the following environments:
12		12
13	1. Install the NPU driver, firmware, and CANN software (Toolkit, ops, and NNAL) by referring to CANN Software Installation.	13	1. Install the NPU driver, firmware, and CANN software (Toolkit, ops, and NNAL) by referring to CANN Software Installation.
14	-<!-- [CANN Software Installation](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/910/softwareinst/instg/instg_0000.html?OS=openEuler&InstallType=netyum) -->
15		14
16	2. Install the PyTorch framework by referring to [Software Installation](https://gitcode.com/Ascend/pytorch/blob/v2.7.1-26.1.0/docs/en/installation_guide/installation_description.md).	15	2. Install the PyTorch framework by referring to [Software Installation](https://gitcode.com/Ascend/pytorch/blob/v2.7.1-26.1.0/docs/en/installation_guide/installation_description.md).
17	-	16	+<!-- [CANN Software Installation](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/910/softwareinst/instg/instg_0000.html?OS=openEuler&InstallType=netyum) -->
18	### Adaptation File Structure	17	### Adaptation File Structure
19		18
20	```text	19	```text
	@@ -143,7 +142,6 @@ Before getting started, ensure that you have completed the installation of the f
143	After completing the operator adaptation development, you can call the custom operator through C++ extensions.	142	After completing the operator adaptation development, you can call the custom operator through C++ extensions.
144		143
145	1. Complete the creation, development, compilation, and deployment workflow for the custom operator project. For details, see CANN Ascend C Operator Development.	144	1. Complete the creation, development, compilation, and deployment workflow for the custom operator project. For details, see CANN Ascend C Operator Development.
146	-<!-- [CANN Ascend C Operator Development](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/latest/programug/Ascendcopdevg/atlas_ascendc_map_10_0002.html) -->
147		145
148	2. Download the code sample.	146	2. Download the code sample.
149		147
	@@ -168,3 +166,4 @@ After completing the operator adaptation development, you can call the custom op
168	Ran xx tests in xx s	166	Ran xx tests in xx s
169	OK	167	OK
170	```	168	```
		169	+<!-- [CANN Ascend C Operator Development](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/latest/programug/Ascendcopdevg/atlas_ascendc_map_10_0002.html) -->

Mexamples/cpp_extension_full/torch_lib_impl/README_en.md+2-3

	@@ -11,10 +11,9 @@ This document describes how to adapt and call a single-operator API through Torc
11	Before getting started, ensure that you have completed the installation of the following environments:	11	Before getting started, ensure that you have completed the installation of the following environments:
12		12
13	1. Install the NPU driver, firmware, and CANN software (Toolkit, ops, and NNAL) by referring to CANN Software Installation.	13	1. Install the NPU driver, firmware, and CANN software (Toolkit, ops, and NNAL) by referring to CANN Software Installation.
14	-<!-- [CANN Software Installation](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/910/softwareinst/instg/instg_0000.html?OS=openEuler&InstallType=netyum) -->
15		14
16	2. Install the PyTorch framework by referring to [Software Installation](https://gitcode.com/Ascend/pytorch/blob/v2.7.1-26.1.0/docs/en/installation_guide/installation_description.md).	15	2. Install the PyTorch framework by referring to [Software Installation](https://gitcode.com/Ascend/pytorch/blob/v2.7.1-26.1.0/docs/en/installation_guide/installation_description.md).
17	-	16	+<!-- [CANN Software Installation](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/910/softwareinst/instg/instg_0000.html?OS=openEuler&InstallType=netyum) -->
18	### Adaptation File Structure	17	### Adaptation File Structure
19		18
20	```text	19	```text
	@@ -138,7 +137,6 @@ Before getting started, ensure that you have completed the installation of the f
138	After completing the operator adaptation development, you can call the custom operator through C++ extensions.	137	After completing the operator adaptation development, you can call the custom operator through C++ extensions.
139		138
140	1. Complete the creation, development, compilation, and deployment workflow for the custom operator project. For details, see CANN Ascend C Operator Development.	139	1. Complete the creation, development, compilation, and deployment workflow for the custom operator project. For details, see CANN Ascend C Operator Development.
141	-<!-- [CANN Ascend C Operator Development](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/latest/programug/Ascendcopdevg/atlas_ascendc_map_10_0002.html) -->
142		140
143	2. Download the code sample.	141	2. Download the code sample.
144		142
	@@ -163,3 +161,4 @@ After completing the operator adaptation development, you can call the custom op
163	Ran xx tests in xx s	161	Ran xx tests in xx s
164	OK	162	OK
165	```	163	```
		164	+<!-- [CANN Ascend C Operator Development](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/latest/programug/Ascendcopdevg/atlas_ascendc_map_10_0002.html) -->

Mexamples/cpp_extension_pybind/README_en.md+2-3

	@@ -11,10 +11,9 @@ This document describes the complete adaptation development workflow for calling
11	Before getting started, ensure that you have completed the installation of the following environments:	11	Before getting started, ensure that you have completed the installation of the following environments:
12		12
13	1. Install the NPU driver, firmware, and CANN software (Toolkit, ops, and NNAL) by referring to CANN Software Installation.	13	1. Install the NPU driver, firmware, and CANN software (Toolkit, ops, and NNAL) by referring to CANN Software Installation.
14	-<!-- [CANN Software Installation](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/910/softwareinst/instg/instg_0000.html?OS=openEuler&InstallType=netyum) -->
15		14
16	2. Install the PyTorch framework by referring to [Installation Guide](https://gitcode.com/Ascend/pytorch/blob/v2.7.1-26.1.0/docs/en/installation_guide/installation_description.md).	15	2. Install the PyTorch framework by referring to [Installation Guide](https://gitcode.com/Ascend/pytorch/blob/v2.7.1-26.1.0/docs/en/installation_guide/installation_description.md).
17	-	16	+<!-- [CANN Software Installation](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/910/softwareinst/instg/instg_0000.html?OS=openEuler&InstallType=netyum) -->
18	### Adaptation File Structure	17	### Adaptation File Structure
19		18
20	```text	19	```text
	@@ -76,7 +75,6 @@ Before getting started, ensure that you have completed the installation of the f
76	After completing the operator adaptation development, you can call the custom operator through C++ extensions.	75	After completing the operator adaptation development, you can call the custom operator through C++ extensions.
77		76
78	1. Complete the creation, development, compilation, and deployment workflow for the custom operator project. For details, see CANN Ascend C Operator Development.	77	1. Complete the creation, development, compilation, and deployment workflow for the custom operator project. For details, see CANN Ascend C Operator Development.
79	-<!-- [CANN Ascend C Operator Development](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/latest/programug/Ascendcopdevg/atlas_ascendc_map_10_0002.html) -->
80		78
81	2. Download the code sample.	79	2. Download the code sample.
82		80
	@@ -101,3 +99,4 @@ After completing the operator adaptation development, you can call the custom op
101	Ran xx tests in xx s	99	Ran xx tests in xx s
102	OK	100	OK
103	```	101	```
		102	+<!-- [CANN Ascend C Operator Development](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/latest/programug/Ascendcopdevg/atlas_ascendc_map_10_0002.html) -->

Mexamples/cpp_extension_structured/README_en.md+2-3

	@@ -11,10 +11,9 @@ This document describes the complete process of custom NPU operator adaptation d
11	Before getting started, ensure that you have completed the installation of the following environments:	11	Before getting started, ensure that you have completed the installation of the following environments:
12		12
13	1. Install the NPU driver, firmware, and CANN software (Toolkit, ops, and NNAL) by referring to CANN Software Installation.	13	1. Install the NPU driver, firmware, and CANN software (Toolkit, ops, and NNAL) by referring to CANN Software Installation.
14	-<!-- [CANN Software Installation](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/910/softwareinst/instg/instg_0000.html?OS=openEuler&InstallType=netyum) -->
15		14
16	2. Install the PyTorch framework by referring to [Installation Guide](https://gitcode.com/Ascend/pytorch/tree/v2.7.1-26.1.0/docs/en/installation_guide/installation_description.md).	15	2. Install the PyTorch framework by referring to [Installation Guide](https://gitcode.com/Ascend/pytorch/tree/v2.7.1-26.1.0/docs/en/installation_guide/installation_description.md).
17	-	16	+<!-- [CANN Software Installation](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/910/softwareinst/instg/instg_0000.html?OS=openEuler&InstallType=netyum) -->
18	### Adaptation File Structure	17	### Adaptation File Structure
19		18
20	```text	19	```text
	@@ -88,7 +87,6 @@ cpp_extension_structured/
88	After completing the operator adaptation development, you can call the custom operator through C++ extensions.	87	After completing the operator adaptation development, you can call the custom operator through C++ extensions.
89		88
90	1. Complete the creation, development, compilation, and deployment workflow for the custom operator project. For details, see CANN Ascend C Operator Development.	89	1. Complete the creation, development, compilation, and deployment workflow for the custom operator project. For details, see CANN Ascend C Operator Development.
91	-<!-- [CANN Ascend C Operator Development](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/latest/programug/Ascendcopdevg/atlas_ascendc_map_10_0002.html) -->
92		90
93	2. Download the code sample.	91	2. Download the code sample.
94		92
	@@ -113,3 +111,4 @@ After completing the operator adaptation development, you can call the custom op
113	Ran xx tests in xx s	111	Ran xx tests in xx s
114	OK	112	OK
115	```	113	```
		114	+<!-- [CANN Ascend C Operator Development](https://www.hiascend.com/document/detail/en/CANNCommunityEdition/latest/programug/Ascendcopdevg/atlas_ascendc_map_10_0002.html) -->