Closed awf closed 10 months ago
Notes (second code block)
Code fails at tar -xvf ./data/gdb11.tgz --directory ./gdb/
. The folder gdb/
doesn't exist and tar
is not allowed to create (running locally on 3b76ef51c33362120d0ac26b488d04626e08fe60). The line below tries to load from /data/ , that is, sortgdb.sort_gdb looks at gdb_filename="./data/gdb11_size09.smi". Changing to tar -xvf ./data/gdb11.tgz --directory ./data/
seem to fix the issue. (when debugging using -nc
with wget circumvents excessive re-downloading)
Let me know if for some reason I'm at the wrong commit/branch (local code looks identical to cell block in ReviewNB)
Managed to reproduce the jax.jit tracing causing time to increase from ~100ms to 5000ms.
Issue is resolved for me if I don't re-create dcargs
for every conformer. @awf
if conformer_num == 0: dcargs = namedtuple('dcargs', dcargs_names)(*(args.__dict__[a] for a in dcargs_names))
Notes (second code block)
Code fails at
tar -xvf ./data/gdb11.tgz --directory ./gdb/
. The foldergdb/
doesn't exist ...Let me know if for some reason I'm at the wrong commit/branch (local code looks identical to cell block in ReviewNB)
Fixed, thanks.
Issue is resolved for me if I don't re-create
dcargs
for every conformer. @awfif conformer_num == 0: dcargs = namedtuple('dcargs', dcargs_names)(*(args.__dict__[a] for a in dcargs_names))
Thanks for the scoping, which led me to understand that it's not the re-creation of dcargs, but of its type, that is namedtuple('dcargs', dcargs_names)
Check out this pull request on
See visual diffs & provide feedback on Jupyter Notebooks.
Powered by ReviewNB
The existing DFT notebook simply calls the
density_functional_theory.py
version usingos.system
. This change moves more of the python logic into the notebook, movingdensity_functional_theory.py
topyscf_ipu/dft.py
.