Open aravinda1879 opened 1 year ago
Hi, there. How many residues are there in your complex? Since you are using segment based optimization, I think the OOM error occurs during the AF2 inference.
Yes, you are correct. Following is the full error. My complex got 2200 residues. I was trying a small Ab with an Antigen to see its predictions. I could trim-down the Antigen a little more. But I wonder if there is an alternative for this? BTW, thanks for sharing the code for testing, it seems promising!
flat_sizes = jax.tree_flatten(in_sizes)[0]
0%| | 0/4 [00:00<?, ?it/s]
0%| | 0/4 [00:11<?, ?it/s]
Traceback (most recent call last):
File "/software/ColabDock/main.py", line 158, in
Currently, I think your proposed solution is the only way. Feel free to post comments if you have further questions.
Hi, While the code works perfectly, for larger systems I am keep getting the following error. jaxlib.xla_extension.XlaRuntimeError: RESOURCE_EXHAUSTED: Failed to allocate request for 81.72GiB (87742154752B) on device ordinal 0 I do have access to ~300 Gb. I even tried adding following to the main.py to circumvent this and still the same error. os.environ["XLA_PYTHON_CLIENT_PREALLOCATE"] = "false" os.environ["XLA_PYTHON_CLIENT_MEM_FRACTION"]=".70" os.environ["XLA_PYTHON_CLIENT_ALLOCATOR"]="platform"
I tried using 'crop_len': 100 to 300, and still the same error. Any thoughts or solutions? Thanks!