Specific jaxlib version?

[james-vincent]

Expected Behavior

Fully functioning colab_batch with GPU.

Current Behavior

colabfold_batch: error: unrecognized arguments: --use-gpu-relax

Steps to Reproduce (for bugs)

Install as described for local installation: ./pip install --upgrade pip ./pip install "colabfold[alphafold] @ git+https://github.com/sokrypton/ColabFold" ./pip install "jax[cuda111]>=0.3.8,<0.4" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html ./conda install -y cudatoolkit=11.1.1 cudnn=8.4.1.50 ./conda install -y -c conda-forge -c bioconda kalign2=2.04 hhsuite=3.3.0 ./conda install -y -c conda-forge openmm=7.5.1 pdbfixer

ColabFold Output (for bugs)

usage: colabfold_batch [-h] [--stop-at-score STOP_AT_SCORE] [--num-recycle NUM_RECYCLE] [--num-models {1,2,3,4,5}] [--recompile-padding RECOMPILE_PADDING] [--model-order MODEL_ORDER] [--host-url HOST_URL] [--data DATA] [--msa-mode {MMseqs2 UniRef+Environmental),MMseqs2 (UniRef only,single_sequence}] [--model-type {auto,AlphaFold2-ptm,AlphaFold2-multimer}] [--amber] [--templates] [--env] [--cpu] [--rank {auto,plddt,ptmscore,multimer}] [--pair-mode {unpaired,paired,unpaired+paired}] [--recompile-all-models] [--sort-queries-by {none,length,random}] [--save-single-representations] [--save-pair-representations] [--zip] [--overwrite-existing-results] input results colabfold_batch: error: unrecognized arguments: --use-gpu-relax

Context

Resulting installation contains:

cudatoolkit 11.1.1 ha002fc5_10 conda-forge cudnn 8.4.1.50 hed8a83a_0 conda-forge jax 0.3.13 jaxlib 0.3.10

Your Environment

Local install CentOS Linux release 7.9.2009 (Core) NVIDIA-SMI 460.39 Driver Version: 460.39 CUDA Version: 11.2 GeForce GTX 108

[martin-steinegger]

Google has changed the URL. It now:

pip install -q "jax[cuda]>=0.3.8,<0.4" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html

We updated the README

[james-vincent]

Which version of python does this assume? Python version is not detailed in the readme.

James Vincent, PhD Bioinformatics Software Curator Dept. of BCMP, Harvard Medical School — BioGrids.org --

On Jun 22, 2022, 9:32 PM -0400, Martin Steinegger @.***>, wrote:

Google has changed the URL. It now: pip install -q "jax[cuda]>=0.3.8,<0.4" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html We updated the README — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

[milot-mirdita]

We test 3.7 and 3.9 in the CI, so anything between these should work.

[james-vincent]

What produces the original issue:

unrecognized arguments: --use-gpu-relax

Is this really not an available argument for some versions/commits? Or, is this argument not available when a GPU is not detected?

If it is the latter there should be a message about why the argument is not available.

Jim

James Vincent, PhD Bioinformatics Software Curator Dept. of BCMP, Harvard Medical School — BioGrids.org --

On Jun 23, 2022, 7:05 AM -0400, Milot Mirdita @.***>, wrote:

We test 3.7 and 3.9 in the CI, so anything between these should work. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

[YoshitakaMo]

@james-vincent Your colabfold_batch seems not the latest commit. Did you install it freshly or use ./update_linux.sh to update the latest version if you use localcolabfold? --use-gpu-relax arg was added on April 18 (https://github.com/sokrypton/ColabFold/commit/896c5353642498055817d62d00e78084c0ccff4d) .