From 0ea78ecd95645c9f4c7d75e0656503852df2e1de Mon Sep 17 00:00:00 2001 From: Ziyao Li <36321246+ZiyaoLi@users.noreply.github.com> Date: Thu, 20 Jun 2024 13:44:11 +0800 Subject: [PATCH 1/3] Update README.md --- README.md | 13 +++++++------ 1 file changed, 7 insertions(+), 6 deletions(-) diff --git a/README.md b/README.md index 288e9f0..d62cb4d 100644 --- a/README.md +++ b/README.md @@ -145,20 +145,20 @@ The output directory of running Uni-Fold contain the predicted structures in `*. ## Training Uni-Fold -Training Uni-Fold relies on pre-calculated features of proteins. We provide a demo dataset in the [example data](example_data) folder. A larger dataset will be released soon. +Training Uni-Fold relies on pre-calculated features of proteins. ### Demo case To start with, we provide a demo script to train the monomer/multimer system of Uni-Fold: ```bash -bash train_monomer_demo.sh . +bash train_monomer_demo.sh ./out ``` and ```bash -bash train_multimer_demo.sh . +bash train_multimer_demo.sh ./out ``` This command starts a training process on the [demo data](example_data) included in this repository. Note that this demo script only tests the correctness of package installation and does not reflect any true performances. @@ -166,7 +166,7 @@ This command starts a training process on the [demo data](example_data) included ### Full training dataset download -We thank [ModelScope](https://modelscope.cn/home) and [Volcengine](https://www.volcengine.com) for providing free hosts of the full training dataset. The downloaded dataset could be directly used to train Uni-Fold from-scratch. +We thank [ModelScope](https://modelscope.cn/home) and [Volcengine](https://www.volcengine.com) for providing free hosts of the full training dataset. The downloaded dataset could be directly used to train Uni-Fold from-scratch. Notably, the full dataset is approximately 2TB in size, so make sure the local storage is capable (>3TB recommended). #### Download from ModelScope @@ -175,7 +175,7 @@ Download the dataset from [modelscope](https://modelscope.cn/datasets/DPTech/Uni 1. install modelscope ```bash -pip3 install https://databot-algo.oss-cn-zhangjiakou.aliyuncs.com/maas/modelscope-1.0.0-py3-none-any.whl +pip3 install modelscope ``` 2. download the dataset with python @@ -192,7 +192,8 @@ ds = MsDataset.load(dataset_name='Uni-Fold-Data', namespace='DPTech', split='tra # The data will be located at ${your_data_path}/modelscope/hub/datasets/downloads/DPTech/Uni-Fold-Data/master/* ``` -#### Download from Volcengine + +#### Download from Volcengine (recommended) 1. install rclone From 59f06004e4c9ef3a50b48f7946c4c72d8a8f3e20 Mon Sep 17 00:00:00 2001 From: Ziyao Li <36321246+ZiyaoLi@users.noreply.github.com> Date: Thu, 20 Jun 2024 13:46:20 +0800 Subject: [PATCH 2/3] Update README.md --- README.md | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) diff --git a/README.md b/README.md index d62cb4d..ad84b79 100644 --- a/README.md +++ b/README.md @@ -145,8 +145,6 @@ The output directory of running Uni-Fold contain the predicted structures in `*. ## Training Uni-Fold -Training Uni-Fold relies on pre-calculated features of proteins. - ### Demo case To start with, we provide a demo script to train the monomer/multimer system of Uni-Fold: @@ -161,12 +159,12 @@ and bash train_multimer_demo.sh ./out ``` -This command starts a training process on the [demo data](example_data) included in this repository. Note that this demo script only tests the correctness of package installation and does not reflect any true performances. +These two commands start training processes on the [demo data](example_data) included in this repository. Note that this demo script only tests the correctness of package installation and does not reflect any true performances. ### Full training dataset download -We thank [ModelScope](https://modelscope.cn/home) and [Volcengine](https://www.volcengine.com) for providing free hosts of the full training dataset. The downloaded dataset could be directly used to train Uni-Fold from-scratch. Notably, the full dataset is approximately 2TB in size, so make sure the local storage is capable (>3TB recommended). +Training Uni-Fold relies on pre-calculated features of proteins. We thank [ModelScope](https://modelscope.cn/home) and [Volcengine](https://www.volcengine.com) for providing free hosts of the full training dataset. The downloaded dataset could be directly used to train Uni-Fold from-scratch. Notably, the full dataset is approximately 2TB in size, so make sure the local storage is capable (>3TB recommended). #### Download from ModelScope From 9eafed15c1918f0afe843620d73d6547b982e65c Mon Sep 17 00:00:00 2001 From: Ziyao Li <36321246+ZiyaoLi@users.noreply.github.com> Date: Thu, 20 Jun 2024 14:09:25 +0800 Subject: [PATCH 3/3] Update .gitignore --- .gitignore | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/.gitignore b/.gitignore index e65bc18..8c788e3 100644 --- a/.gitignore +++ b/.gitignore @@ -120,4 +120,5 @@ test/ *.tfevents.* *.sto *.a3m -nogit/ +/nogit/ +/out/