From a2930eefa2155ae249a3136833a947e9a7597904 Mon Sep 17 00:00:00 2001 From: JavaZero <2487163254@qq.com> Date: Thu, 25 Sep 2025 20:50:50 +0800 Subject: [PATCH] update parallel configuration values in start_tasks.md --- docs/mindformers/docs/source_en/feature/start_tasks.md | 8 ++------ docs/mindformers/docs/source_zh_cn/feature/start_tasks.md | 8 ++------ 2 files changed, 4 insertions(+), 12 deletions(-) diff --git a/docs/mindformers/docs/source_en/feature/start_tasks.md b/docs/mindformers/docs/source_en/feature/start_tasks.md index eb8ddb6c9f..3ebb7f1600 100644 --- a/docs/mindformers/docs/source_en/feature/start_tasks.md +++ b/docs/mindformers/docs/source_en/feature/start_tasks.md @@ -145,12 +145,8 @@ Take Qwen2.5-0.5B as an example to perform 2-node 16-device fine-tuning. ```yaml parallel_config: - data_parallel: 2 - model_parallel: 4 - pipeline_stage: 2 - micro_batch_num: 16 - vocab_emb_dp: True - gradient_aggregation_group: 4 + data_parallel: 16 + ... ``` > If the number of nodes and the number of devices are used to change, `data_parallel`, `model_parallel`, and `pipeline_stage` need to be modified to meet the actual number of running devices. `device_num=data_parallel×model_parallel×pipeline_stage`. Meanwhile, `micro_batch_num >= pipeline_stage`. diff --git a/docs/mindformers/docs/source_zh_cn/feature/start_tasks.md b/docs/mindformers/docs/source_zh_cn/feature/start_tasks.md index f1c8e65b40..82da03c0d8 100644 --- a/docs/mindformers/docs/source_zh_cn/feature/start_tasks.md +++ b/docs/mindformers/docs/source_zh_cn/feature/start_tasks.md @@ -145,12 +145,8 @@ bash scripts/msrun_launcher.sh "run_mindformer.py \ ```yaml parallel_config: - data_parallel: 2 - model_parallel: 4 - pipeline_stage: 2 - micro_batch_num: 16 - vocab_emb_dp: True - gradient_aggregation_group: 4 + data_parallel: 16 + ... ``` > 如使用节点数和卡数改变需要修改`data_parallel`、 `model_parallel`、 `pipeline_stage`满足实际运行的卡数 `device_num=data_parallel×model_parallel×pipeline_stage`，同时满足`micro_batch_num >= pipeline_stage`。 -- Gitee