OOM for rf3

[Admire7494]

The command is rf3 fold inputs='/data2/wmq/foundry-production/P2_6ACA_rf.json' ckpt_path='/data2/wmq/.foundry/checkpoints/rf3_foundry_01_24_latest_remapped.ckpt' out_dir=logs/inference_outs/demo/P2_6ACA diffusion_batch_size=1

The json file is
{
"name": "P2_6ACA",
"components": [
{
"seq": A lenth of 1000 aa protein,
"msa_path": "/data2/wmq/foundry-production/P2.a3m",
"chain_id": "A"
},
{
"ccd_code": "AMP",
"chain_id": "B"
},
{
"ccd_code": "NDP",
"chain_id": "C"
},
{
"smiles": "C(CCC(=O)O)CCN",
"chain_id": "E"
}
]
}

The error is
torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 4.35 GiB. GPU 0 has a total capacity of 23.65 GiB of which 2.64 GiB is free. Including non-PyTorch memory, this process has 21.00 GiB memory in use. Of the allocated memory 17.12 GiB is allocated by PyTorch, and 2.60 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)

Any suggestion for this large sequence?
Thanks a lot.

[Admire7494]

How to use all GPUs in my machine to predict?

The command is rf3 fold inputs='/data2/wmq/foundry-production/P2_6ACA_rf.json' ckpt_path='/data2/wmq/.foundry/checkpoints/rf3_foundry_01_24_latest_remapped.ckpt' out_dir=logs/inference_outs/demo/P2_6ACA diffusion_batch_size=1

The json file is { "name": "P2_6ACA", "components": [ { "seq": A lenth of 1000 aa protein, "msa_path": "/data2/wmq/foundry-production/P2.a3m", "chain_id": "A" }, { "ccd_code": "AMP", "chain_id": "B" }, { "ccd_code": "NDP", "chain_id": "C" }, { "smiles": "C(CCC(=O)O)CCN", "chain_id": "E" } ] }

The error is torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 4.35 GiB. GPU 0 has a total capacity of 23.65 GiB of which 2.64 GiB is free. Including non-PyTorch memory, this process has 21.00 GiB memory in use. Of the allocated memory 17.12 GiB is allocated by PyTorch, and 2.60 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)

Any suggestion for this large sequence? Thanks a lot.

Besides, how to use all my GPUs at once?

[r-krishna]

We do not currently support parallelizing the model across GPUs. I've noted this down and we can think about it in a future release. I'd love for an open source contribution of this functionality.