relion icon indicating copy to clipboard operation
relion copied to clipboard

Download model

Open ryao-mdanderson opened this issue 1 year ago • 10 comments

Dear Relion support team,

This is related to a closed issue report #1003. I checked out the work around for blush model download. Thank you.

However, while running relion5 beta, notice the application is trying download some and failed as shown below: ... Downloading: "https://dl.fbaipublicfiles.com/fair-esm/models/esm1b_t33_650M_UR50S.pt" to /risapps/rhel8/relion/5.0/torch-home/hub/checkpoints/esm1b_t33_650M_UR50S.pt (where /risapps/rhel8/relion/5.0/torch-home/ is defined as $torch_home)

Seems download not only for blush, there other models need to taken care as well... Is there an alternative to have all the required files download once manually and set up for use?

our cluster node does not have internet access, therefore it's inconvenient to request firewall open on multiple sites which I don't know ahead of the time for this work.

Thank you very much for your support. Rong Yao

ryao-mdanderson avatar Jan 23 '24 17:01 ryao-mdanderson

Is there an alternative to have all the required files download once manually and set up for use?

The easiest way is to install RELION on a local machine with internet access and copy downloaded models in local $TORCH_HOME to $TORCH_HOME` of your cluster.

If reverse SSH port-forwarding and SOCKS proxying are allowed, you can also use them (see this).

our cluster node does not have internet access

Is the internet access blocked on the head (login) node as well? How did you set up the conda environment then?

biochem-fan avatar Jan 24 '24 01:01 biochem-fan

@biochem-fan , Thank you for your reply.

A local standalone server with full internet access does not have the gpu card. I recall 'make' failed on error make[2]: *** [src/apps/CMakeFiles/relion_gpu_util.dir/build.make:112: therefore I have to build the application on a gpu card. Correct me !

Our cluster login node has approved internet access (firewall open for required site). We have granted to acces zenodo.org to download model. However, user reported the following site also required... Is there any other sites I am not aware? I like the idea one posting suggests, how about put all the download models to zenodo?

Downloading: "https://dl.fbaipublicfiles.com/fair-esm/models/esm1b_t33_650M_UR50S.pt"

Thanks!

ryao-mdanderson avatar Jan 24 '24 02:01 ryao-mdanderson

You don't need a GPU card to build RELION. Try cmake -DCUDA=OFF after deleting your build directory.

biochem-fan avatar Jan 24 '24 03:01 biochem-fan

Hello @biochem-fan

Thank you for your tip for cmake without cuda suppport. I will try.

ryao-mdanderson avatar Jan 24 '24 16:01 ryao-mdanderson

Hello @biochem-fan :

I tried build relion on a standalone server which has internet access. This goal is to download data models.

I failed make process at 58%, however, I am seeing make process accesses two ftp sites: Using src='ftp://ftp.mrc-lmb.cam.ac.uk/pub/scheres/fltk-1.3.5-source.tar.gz' Using src='ftp://ftp.mrc-lmb.cam.ac.uk/pub/dari/class_ranker_0.1.3_torch_1.0.1.pt.tar.gz'

at the end of exited make process, noticed the following contents in my test build [ris_hpc_apps@r1drpswdev3 external]$ ls -lrt total 0 drwxr-xr-x. 3 ris_hpc_apps rists 74 Jan 24 10:35 torch_models drwxr-xr-x. 3 ris_hpc_apps rists 50 Jan 24 10:35 fltk [ris_hpc_apps@r1drpswdev3 external]$ ls -lrt torch_models/ total 2140 -rw-r--r--. 1 ris_hpc_apps rists 2190018 Jan 24 10:35 class_ranker_0.1.3_torch_1.0.1.pt.tar.gz drwxr-xr-x. 2 ris_hpc_apps rists 47 Jan 24 10:35 class_ranker

I do not see the download from Zenodo. I do not find models (e.g. Blush, ModelAngelo and DynaMight). I wonder if the download complete?

I attached my build log here, please advise. Thank you very much! relion5-r1drpswdev3.txt

ryao-mdanderson avatar Jan 24 '24 17:01 ryao-mdanderson

Because you put -j 8, various build processes were running in parallel, so the log is harder to read. But you should notice:

Fl_Gl_Choice.cxx: In function ‘__GLXcontextRec* fl_create_gl_context(Fl_Window*, const Fl_Gl_Choice*)’:
Fl_Gl_Choice.cxx:335:7: error: ‘GLX_CONTEXT_MAJOR_VERSION_ARB’ was not declared in this scope
       GLX_CONTEXT_MAJOR_VERSION_ARB, 3,
       ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Compiling freeglut_stroke_mono_roman.cxx...
Fl_Gl_Choice.cxx:335:7: note: suggested alternative: ‘FL_MAJOR_VERSION’
       GLX_CONTEXT_MAJOR_VERSION_ARB, 3,
       ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
       FL_MAJOR_VERSION
Fl_Gl_Choice.cxx:336:7: error: ‘GLX_CONTEXT_MINOR_VERSION_ARB’ was not declared in this scope
       GLX_CONTEXT_MINOR_VERSION_ARB, 2,
       ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Compiling freeglut_stroke_roman.cxx...
Fl_Gl_Choice.cxx:336:7: note: suggested alternative: ‘GLX_GLXEXT_VERSION’
       GLX_CONTEXT_MINOR_VERSION_ARB, 2,
       ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
       GLX_GLXEXT_VERSION

Because of this, the build process failed before starting the model download step:

[ 58%] Linking CXX static library ../../lib/librelion_lib.a
[ 58%] Built target relion_lib
make: *** [Makefile:136: all] Error 2

What is your OS? I guess you have an extremely old GLX header files, which led to FLTK build failure.

Can you try cmake -DCUDA=OFF -DGUI=OFF to disable all GUI programs. Make sure you delete your old build directory.

biochem-fan avatar Jan 25 '24 03:01 biochem-fan

Thank you very much @biochem-fan for your helps. With cmake -DCUDA=OFF -DGUI=OFF, I am able to complete make at 100%.

Seems this time only downloaded class_ranker. I don't see other models, is this right?

[ris_hpc_apps@r1drpswdev3 torch_models]$ ls -l /var/tmp/relion/external drwxr-xr-x. 3 ris_hpc_apps rists 74 Jan 24 22:49 torch_models

[ris_hpc_apps@r1drpswdev3 class_ranker]$ ls -l /var/tmp/relion/external/torch_models total 2140 drwxr-xr-x. 2 ris_hpc_apps rists 47 Jan 24 22:49 class_ranker -rw-r--r--. 1 ris_hpc_apps rists 2190018 Jan 24 22:49 class_ranker_0.1.3_torch_1.0.1.pt.tar.gz

This is a RHEL 8.7 server. I attached the installation log: relion5-r1drpswdev3.txt

ryao-mdanderson avatar Jan 25 '24 05:01 ryao-mdanderson

You didn't check out the ver5.0 branch. You built RELION 4.0.x, which does not have ModelAngelo, Blush etc.

biochem-fan avatar Jan 25 '24 05:01 biochem-fan

@biochem-fan Thank you very much for your helps. If just for download all models purpose, is the cmake step good enough?

After cmake steps, I am seeing the contents in desired torch-home, is these directories complete?

[ris_hpc_apps@r1drpswdev3 build]$ ls -l /home/ris_hpc_apps/relion/torch-home/hub/checkpoints/ total 7645100 -r-xr-xr-x. 1 ris_hpc_apps rists 3174 Jan 25 08:56 esm1b_t33_650M_UR50S-contact-regression.pt -r-xr-xr-x. 1 ris_hpc_apps rists 7828576466 Jan 25 08:56 esm1b_t33_650M_UR50S.pt drwxr-xr-x. 4 ris_hpc_apps rists 51 Jan 25 09:10 model_angelo_v1.0 drwxr-xr-x. 2 ris_hpc_apps rists 49 Jan 25 08:42 relion_blush drwxr-xr-x. 2 ris_hpc_apps rists 49 Jan 25 08:42 relion_class_ranker

ryao-mdanderson avatar Jan 25 '24 15:01 ryao-mdanderson

The folder looks complete but why don't you run make just to be sure?

biochem-fan avatar Jan 25 '24 23:01 biochem-fan