docs: fix broken relative-link anchors (batch 1)-amct-AtomGit

已关闭

docs: fix broken relative-link anchors (batch 1) #272

不喜欢雨天ༀ创建于 8月25日关闭于 9月22日从2403_87936996/amct： fix/anchor-links-batch1 合入到 cann/amct： master

docs: fix broken relative-link anchors (batch 1) #272

已关闭

不喜欢雨天ༀ创建于 8月25日关闭于 9月22日

从2403_87936996/amct fix/anchor-links-batch1合入到cann/amct master

共 3 个文件变更+9-16

Mdocs/zh/context/ptq_config_param.md+1-2

	@@ -1,6 +1,6 @@
1	# 训练后量化配置参数说明	1	# 训练后量化配置参数说明
2		2
3	-如果通过[create\_quant\_config](../api/create_quant_config.md)接口生成的config.json量化配置文件，推理精度不满足要求，则需要参见该章节不断调整config.json文件中的内容（用户修改JSON文件时，请确保层名唯一），直至精度满足要求，json量化配置文件样例请参见接口中的[调用示例](../api/create_quant_config.md#zh-cn_topic_0240187365_section64231658994)部分。	3	+如果通过[create\_quant\_config](../api/create_quant_config.md)接口生成的config.json量化配置文件，推理精度不满足要求，则需要参见该章节不断调整config.json文件中的内容（用户修改JSON文件时，请确保层名唯一），直至精度满足要求，json量化配置文件样例请参见接口中的[调用示例](../api/create_quant_config.md)部分。
4		4
5	配置文件中参数说明如下：	5	配置文件中参数说明如下：
6		6
	@@ -738,4 +738,3 @@
738	</tr>	738	</tr>
739	</tbody>	739	</tbody>
740	</table>	740	</table>
741	-

Mexamples/models/qwen3.6/Qwen3.6-Moe.md+4-4

	@@ -73,7 +73,7 @@ python -m amct_pytorch.eval \
73	基准测试精度结果：	73	基准测试精度结果：
74	`Wikitext2-ppl=6.2825`	74	`Wikitext2-ppl=6.2825`
75		75
76	-更多参数说明请参见[参数说明](../../../docs/zh/AMCT_Pytorch_LLM.md#31-通用参数)	76	+更多参数说明请参见[参数说明](../../../docs/zh/AMCT_Pytorch_LLM.md#31-%E9%80%9A%E7%94%A8%E5%8F%82%E6%95%B0)
77		77
78	### 直转量化精度评估	78	### 直转量化精度评估
79	根据YAML的bit配置，进行直转量化精度测试，评估与基准精度的差距，当前方案默认为对`quant-target`做全A8W4的int量化：	79	根据YAML的bit配置，进行直转量化精度测试，评估与基准精度的差距，当前方案默认为对`quant-target`做全A8W4的int量化：
	@@ -97,7 +97,7 @@ python -m amct_pytorch.eval \
97	直转量化精度结果：	97	直转量化精度结果：
98	`Wikitext2-ppl=7.0407`	98	`Wikitext2-ppl=7.0407`
99		99
100	-更多参数说明请参见[参数说明](../../../docs/zh/AMCT_Pytorch_LLM.md#32-ppl-评估参数)	100	+更多参数说明请参见[参数说明](../../../docs/zh/AMCT_Pytorch_LLM.md#32-ppl-%E8%AF%84%E4%BC%B0%E5%8F%82%E6%95%B0)
101		101
102	### PTQ数据提取	102	### PTQ数据提取
103	根据不同的量化对象`quant_target`提取相应的 PTQ 校准数据集：	103	根据不同的量化对象`quant_target`提取相应的 PTQ 校准数据集：
	@@ -115,7 +115,7 @@ python -m amct_pytorch.extract_ptq_data \
115	必要传参解释：	115	必要传参解释：
116	- data_dir：提取数据目录	116	- data_dir：提取数据目录
117		117
118	-更多参数说明请参见[参数说明](../../../docs/zh/AMCT_Pytorch_LLM.md#33-数据提取参数)	118	+更多参数说明请参见[参数说明](../../../docs/zh/AMCT_Pytorch_LLM.md#33-%E6%95%B0%E6%8D%AE%E6%8F%90%E5%8F%96%E5%8F%82%E6%95%B0)
119		119
120	### Post-Training Quantization	120	### Post-Training Quantization
121	引入量化算法，对量化过程做优化，以减少量化带来的损失，以autoround为例：	121	引入量化算法，对量化过程做优化，以减少量化带来的损失，以autoround为例：
	@@ -147,7 +147,7 @@ python -m amct_pytorch.ptq \
147	为提升训练效率，我们提供多卡下的训练脚本	147	为提升训练效率，我们提供多卡下的训练脚本
148	多卡环境请参照脚本[ptq_multi_npu](../../ptq_multi_npu.sh)	148	多卡环境请参照脚本[ptq_multi_npu](../../ptq_multi_npu.sh)
149		149
150	-更多参数说明请参见[参数说明](../../../docs/zh/AMCT_Pytorch_LLM.md#35-ptq-参数)	150	+更多参数说明请参见[参数说明](../../../docs/zh/AMCT_Pytorch_LLM.md#35-ptq-%E5%8F%82%E6%95%B0)
151		151
152	### 基于Post-Training Quantization的直转量化精度评估	152	### 基于Post-Training Quantization的直转量化精度评估
153	完成PTQ后，在直转量化精度评估中加入量化算法，与基准测试、无量化算法的直转量化精度比对，验证量化算法的效果：	153	完成PTQ后，在直转量化精度评估中加入量化算法，与基准测试、无量化算法的直转量化精度比对，验证量化算法的效果：

Mexamples/models/qwen3.6/Qwen3.6-Moe_en.md+4-10

	@@ -55,7 +55,6 @@ The one-stop platform has pre-configured deployment runtime environment. When us
55		55
56	```shell	56	```shell
57	python -m amct_pytorch.eval \	57	python -m amct_pytorch.eval \
58	- --trust_remote_code \
59	--model /data/models/Qwen3.6-35B-A3B \	58	--model /data/models/Qwen3.6-35B-A3B \
60	--model_name qwen3_6_moe \	59	--model_name qwen3_6_moe \
61	--seq_len 4096 \	60	--seq_len 4096 \
	@@ -73,13 +72,12 @@ Required parameter explanation:
73	Baseline test accuracy result:	72	Baseline test accuracy result:
74	`Wikitext2-ppl=6.2825`	73	`Wikitext2-ppl=6.2825`
75		74
76	-For more detailed parameter explanation, please refer to [Parameter Description](../../../docs/zh/AMCT_Pytorch_LLM.md#31-通用参数)	75	+For more detailed parameter explanation, please refer to [Parameter Description](../../../docs/zh/AMCT_Pytorch_LLM.md#31-%E9%80%9A%E7%94%A8%E5%8F%82%E6%95%B0)
77		76
78	### Direct Conversion Quantization Accuracy Evaluation	77	### Direct Conversion Quantization Accuracy Evaluation
79	According to YAML bit configuration, perform direct conversion quantization accuracy test, evaluate the gap with baseline accuracy. Current solution defaults to full A8W4 int quantization on `quant-target`:	78	According to YAML bit configuration, perform direct conversion quantization accuracy test, evaluate the gap with baseline accuracy. Current solution defaults to full A8W4 int quantization on `quant-target`:
80	```shell	79	```shell
81	python -m amct_pytorch.eval \	80	python -m amct_pytorch.eval \
82	- --trust_remote_code \
83	--model /data/models/Qwen3.6-35B-A3B \	81	--model /data/models/Qwen3.6-35B-A3B \
84	--model_name qwen3_6_moe \	82	--model_name qwen3_6_moe \
85	--seq_len 4096 \	83	--seq_len 4096 \
	@@ -97,13 +95,12 @@ Required parameter explanation:
97	Direct conversion quantization accuracy result:	95	Direct conversion quantization accuracy result:
98	`Wikitext2-ppl=7.0407`	96	`Wikitext2-ppl=7.0407`
99		97
100	-For more detailed parameter explanation, please refer to [Parameter Description](../../../docs/zh/AMCT_Pytorch_LLM.md#32-ppl-评估参数)	98	+For more detailed parameter explanation, please refer to [Parameter Description](../../../docs/zh/AMCT_Pytorch_LLM.md#32-ppl-%E8%AF%84%E4%BC%B0%E5%8F%82%E6%95%B0)
101		99
102	### PTQ Data Extraction	100	### PTQ Data Extraction
103	Extract corresponding PTQ calibration dataset according to different quantization objects `quant_target`:	101	Extract corresponding PTQ calibration dataset according to different quantization objects `quant_target`:
104	```shell	102	```shell
105	python -m amct_pytorch.extract_ptq_data \	103	python -m amct_pytorch.extract_ptq_data \
106	- --trust_remote_code \
107	--model /data/models/Qwen3.6-35B-A3B \	104	--model /data/models/Qwen3.6-35B-A3B \
108	--model_name qwen3_6_moe \	105	--model_name qwen3_6_moe \
109	--seq_len 4096 \	106	--seq_len 4096 \
	@@ -115,14 +112,13 @@ python -m amct_pytorch.extract_ptq_data \
115	Required parameter explanation:	112	Required parameter explanation:
116	- data_dir: Extracted data directory	113	- data_dir: Extracted data directory
117		114
118	-For more detailed parameter explanation, please refer to [Parameter Description](../../../docs/zh/AMCT_Pytorch_LLM.md#33-数据提取参数)	115	+For more detailed parameter explanation, please refer to [Parameter Description](../../../docs/zh/AMCT_Pytorch_LLM.md#33-%E6%95%B0%E6%8D%AE%E6%8F%90%E5%8F%96%E5%8F%82%E6%95%B0)
119		116
120	### Post-Training Quantization	117	### Post-Training Quantization
121	Introduce quantization algorithm to optimize the quantization process to reduce quantization loss, using autoround as an example:	118	Introduce quantization algorithm to optimize the quantization process to reduce quantization loss, using autoround as an example:
122	#### Single-Card Environment	119	#### Single-Card Environment
123	```shell	120	```shell
124	python -m amct_pytorch.ptq \	121	python -m amct_pytorch.ptq \
125	- --trust_remote_code \
126	--model /data/models/Qwen3.6-35B-A3B \	122	--model /data/models/Qwen3.6-35B-A3B \
127	--model_name qwen3_6_moe \	123	--model_name qwen3_6_moe \
128	--seq_len 4096 \	124	--seq_len 4096 \
	@@ -147,13 +143,12 @@ Required parameter explanation:
147	To improve training efficiency, we provide training scripts under multi-card	143	To improve training efficiency, we provide training scripts under multi-card
148	For multi-card environment, please refer to script [ptq_multi_npu](../../ptq_multi_npu.sh)	144	For multi-card environment, please refer to script [ptq_multi_npu](../../ptq_multi_npu.sh)
149		145
150	-For more detailed parameter explanation, please refer to [Parameter Description](../../../docs/zh/AMCT_Pytorch_LLM.md#35-ptq-参数)	146	+For more detailed parameter explanation, please refer to [Parameter Description](../../../docs/zh/AMCT_Pytorch_LLM.md#35-ptq-%E5%8F%82%E6%95%B0)
151		147
152	### Direct Conversion Quantization Accuracy Evaluation Based on Post-Training Quantization	148	### Direct Conversion Quantization Accuracy Evaluation Based on Post-Training Quantization
153	After completing PTQ, add quantization algorithm in direct conversion quantization accuracy evaluation, compare with baseline test and direct conversion quantization accuracy without quantization algorithm, verify quantization algorithm effectiveness:	149	After completing PTQ, add quantization algorithm in direct conversion quantization accuracy evaluation, compare with baseline test and direct conversion quantization accuracy without quantization algorithm, verify quantization algorithm effectiveness:
154	```shell	150	```shell
155	python -m amct_pytorch.eval \	151	python -m amct_pytorch.eval \
156	- --trust_remote_code \
157	--model /data/models/Qwen3.6-35B-A3B \	152	--model /data/models/Qwen3.6-35B-A3B \
158	--model_name qwen3_6_moe \	153	--model_name qwen3_6_moe \
159	--seq_len 4096 \	154	--seq_len 4096 \
	@@ -173,7 +168,6 @@ Required parameter explanation:
173	After all the preceding steps are complete, export the quantized weights for loading. The weight_map corresponding to the weights is consistent with that on the huggingface official website.	168	After all the preceding steps are complete, export the quantized weights for loading. The weight_map corresponding to the weights is consistent with that on the huggingface official website.
174	```shell	169	```shell
175	python -m amct_pytorch.deploy \	170	python -m amct_pytorch.deploy \
176	- --trust_remote_code \
177	--model /data/models/Qwen3.6-35B-A3B \	171	--model /data/models/Qwen3.6-35B-A3B \
178	--model_name qwen3_6_moe \	172	--model_name qwen3_6_moe \
179	--granularity block \	173	--granularity block \