Autobuild Python/Pytorch module for derivative code - #166

Open

fymue wants to merge 97 commits into

second_derivativefrom

pytorch_interface

Commits on Dec 27, 2022

setup.py generation for automatic creation of a Python module for derivative GAPC code (still need to add pytorch_interface.cc generation)
fymue
committed
91217c1
bugfix
fymue
committed
55fca5f

Commits on Dec 28, 2022

Commits on Dec 30, 2022

modify init function; add functions to convert forward and backward score matrices to pytorch Tensors; add generic_pytorch_interface.cc; can now calculate score matrices in Python for NW
fymue
committed
fa90e7a
need to use torch::kFloat64 instead of torch::kFloat32 to avoid overflow errors when adding C++ double vals to tensor (probably need to change our code to use float so tensor can be float32 for GPU…
fymue
committed
03b8633
fixed setup.py linker flags
fymue
committed
60fbc26

Commits on Dec 31, 2022

Commits on Jan 1, 2023

Commits on Jan 2, 2023

added some comments
fymue
committed
b04a222

Commits on Jan 7, 2023

also clear traces vector in init function for proper resetting of everything if forward gets re-called with different inputs; fix comment
fymue
committed
7e1c18f

Commits on Jan 11, 2023

Commits on Jan 16, 2023

Commits on Mar 16, 2023

Merge branch 'second_derivative' of github.com:jlab/gapc into pytorch_interface
fymue
committed
60f68c1

Commits on Mar 17, 2023

Commits on Mar 19, 2023

Commits on Mar 20, 2023

Commits on Mar 25, 2023

Commits on Mar 26, 2023

Commits on Mar 27, 2023

refactor indexing (was bugged before); add some more tensor ops funcs
fymue
committed
7eb49da
template traces.hh functions and classes
fymue
committed
6afcfe3

Commits on Mar 28, 2023

batched input processing works now; clone tensors returned from get methods to avoid unwanted overwriting; add some funcs that can be called from GAP-L code
fymue
committed
5490cbd

Commits on Apr 1, 2023

Commits on Apr 3, 2023

make index tracing more efficient (cuts runtime in half)
fymue
committed
c794d47
make some funcs a little faster
fymue
committed
63a1416

Commits on Apr 6, 2023

rework batching mechanism to use new Batch type (MUCH faster now)
fymue
committed
d9d6c5b

Commits on Apr 7, 2023

move batch implementation to own file; move some related funcs to src/tensor.hh; add AVX2 SIMD functions
fymue
committed
f47be7e

Commits on Apr 8, 2023

use memory pool to allocate batch objects
fymue
committed
88b4c8d
fix/add some comments
fymue
committed
3adb21d
fix whitespace error
fymue
committed
1c41497

Commits on Apr 10, 2023

Commits on Apr 13, 2023

provide object that can be used to access any input tensor from GAP-L code
fymue
committed
d9fefb9

Commits on Apr 14, 2023

fix std::bad_cast bug for non-batched GAPL declaration; refactors/comments
fymue
committed
d200dce

Commits on Apr 15, 2023

add some copy/move constructors for more explicit/efficient copying/moving
fymue
committed
6377993
forgot return statement
fymue
committed
e87e2bc

Commits on Apr 17, 2023

add macro for max size of index_components obj; fix some comments
fymue
committed
73f71cf

Commits on Apr 19, 2023

add pytorch module tests (don't work 100% yet)
fymue
committed
97b8194

Commits on Apr 21, 2023

get rid of static to allow multiple fw/bw calls in same program (fixes issue of previous comment)
fymue
committed
32463eb
refactor/clean up interface
fymue
committed
9400f51
fully clear vectors in init; some refactors
fymue
committed
25e7f70
add github action for pytorch interface tests
fymue
committed
8af5bce
add -march=native flag for batched inputs to turn on AVX2 (if available)
fymue
committed
0b2ee5c
run action on push to this branch
fymue
committed
54233de
have to install ninja in a separate pip3 call
fymue
committed
e0907d3

Commits on May 9, 2023

add support for second derivative (still need to write some tests)
fymue
committed
574db7b
forgot to adjust grammar
fymue
committed
75b1490

Commits on May 11, 2023

Commits on May 15, 2023

improve memory usage (for 2nd derivative)
fymue
committed
99abbda

Commits on Jun 27, 2023

mark redundant funcs for removal later
fymue
committed
2ea1bc1

Commits on Jun 28, 2023

add minsize/maxsize filters for tensors; warn user if input tensors have differently-sized batch dimensions
fymue
committed
a4af9b5

Commits on Jun 30, 2023

Commits on Jul 4, 2023

test
fymue
committed
f53fa79
add new template expr; allow arbitrarily-sized index_components objects with the help of template args
fymue
committed
9a8001f
linting
fymue
committed
9c805f0
small refactor
fymue
committed
cacc034
linting
fymue
committed
ff9a11a
throw Error if batch-incompatible answer type is provided in algebra when batched input is supposed to be processed; allow users to limit max batch size with environment variable at compilation
fymue
committed
ff7b4fe

Commits on Jul 7, 2023

simply indices initialization
fymue
committed
e02448b

Commits on Jul 8, 2023

update help msg; add option to use malloc instead of memory pool for Batch object allocation
fymue
committed
903805a

Commits on Jul 9, 2023

fix edge case where premature memory allocation can occur
fymue
committed
c6718aa

Commits on Jul 10, 2023

Commits on Jul 11, 2023

add TSLICE0 terminal parser; add GAP-C call to description of compiled package
fymue
committed
c5ffa38

Commits on Jul 18, 2023

Commits on Jul 22, 2023

Commits on Nov 3, 2025

- Re-enabled C++17 support (previously forced C++14 for legacy reasons).
Dominik Huefner
committed
ec1824b

Commits on Nov 4, 2025

Removed unnesecary comments
Dominik Huefner
committed
fe1a330

Commits on Nov 10, 2025

Merge pull request #243 from jlab/fix_pylibs
sjanssen2
authored
6b637cc