Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Match escn_exportable with escn main #866

Merged
merged 3 commits into from
Sep 20, 2024
Merged

Match escn_exportable with escn main #866

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
dc9f40d
fix l/m; make rescaling optional
misko Sep 19, 2024
4ff238c
Merge branch 'main' into escn_exportable_matching
misko Sep 19, 2024
dcd2501
remove mapping reduced from tests, hoping we wont need to register bu…
misko Sep 20, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 8 additions & 14 deletions src/fairchem/core/models/escn/escn_exportable.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -79,6 +79,7 @@ def __init__(
resolution: int | None = None,
compile: bool = False,
export: bool = False,
rescale_grid: bool = False,
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

unless we plan to test this, lets remove this as an option, i still prefer to remove the rescale code and keep it simpler

) -> None:
super().__init__()

Expand All @@ -103,6 +104,7 @@ def __init__(
self.distance_function = distance_function
self.compile = compile
self.export = export
self.rescale_grid = rescale_grid

# non-linear activation function used throughout the network
self.act = nn.SiLU()
Expand Down Expand Up @@ -152,12 +154,11 @@ def __init__(
# Initialize the transformations between spherical and grid representations
self.SO3_grid = nn.ModuleDict()
self.SO3_grid["lmax_lmax"] = SO3_Grid(
self.lmax, self.lmax, resolution=resolution
self.lmax, self.lmax, resolution=resolution, rescale=self.rescale_grid
)
self.SO3_grid["lmax_mmax"] = SO3_Grid(
self.lmax, self.mmax, resolution=resolution
self.lmax, self.mmax, resolution=resolution, rescale=self.rescale_grid
)
self.mappingReduced = CoefficientMapping([self.lmax], [self.mmax])

# Initialize the blocks for each layer of the GNN
self.layer_blocks = nn.ModuleList()
Expand All @@ -173,7 +174,6 @@ def __init__(
self.max_num_elements,
self.SO3_grid,
self.act,
self.mappingReduced,
)
self.layer_blocks.append(block)

Expand Down Expand Up @@ -435,7 +435,6 @@ def __init__(
max_num_elements: int,
SO3_grid: SO3_Grid,
act,
mappingReduced,
) -> None:
super().__init__()
self.layer_idx = layer_idx
Expand All @@ -444,7 +443,6 @@ def __init__(
self.mmax = mmax
self.sphere_channels = sphere_channels
self.SO3_grid = SO3_grid
self.mappingReduced = mappingReduced

# Message block
self.message_block = MessageBlock(
Expand All @@ -458,7 +456,6 @@ def __init__(
max_num_elements,
self.SO3_grid,
self.act,
self.mappingReduced,
)

# Non-linear point-wise comvolution for the aggregated messages
Expand Down Expand Up @@ -547,7 +544,6 @@ def __init__(
max_num_elements: int,
SO3_grid: SO3_Grid,
act,
mappingReduced,
) -> None:
super().__init__()
self.layer_idx = layer_idx
Expand All @@ -558,8 +554,9 @@ def __init__(
self.lmax = lmax
self.mmax = mmax
self.edge_channels = edge_channels
self.mappingReduced = mappingReduced
self.out_mask = self.mappingReduced.coefficient_idx(self.lmax, self.mmax)
self.out_mask = CoefficientMapping([self.lmax], [self.lmax]).coefficient_idx(
self.lmax, self.mmax
)

# Create edge scalar (invariant to rotations) features
self.edge_block = EdgeBlock(
Expand All @@ -577,7 +574,6 @@ def __init__(
self.lmax,
self.mmax,
self.act,
self.mappingReduced,
)
self.so2_block_target = SO2Block(
self.sphere_channels,
Expand All @@ -586,7 +582,6 @@ def __init__(
self.lmax,
self.mmax,
self.act,
self.mappingReduced,
)

def forward(
Expand Down Expand Up @@ -666,15 +661,14 @@ def __init__(
lmax: int,
mmax: int,
act,
mappingReduced,
) -> None:
super().__init__()
self.sphere_channels = sphere_channels
self.hidden_channels = hidden_channels
self.lmax = lmax
self.mmax = mmax
self.act = act
self.mappingReduced = mappingReduced
self.mappingReduced = CoefficientMapping([self.lmax], [self.mmax])

num_channels_m0 = (self.lmax + 1) * self.sphere_channels

Expand Down
1 change: 0 additions & 1 deletion src/fairchem/core/models/escn/so3.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -280,7 +280,6 @@ def _grid_act(self, SO3_grid, act, mappingReduced) -> None:
from_grid_mat = SO3_grid[self.lmax_list[i]][
self.mmax_list[i]
].get_from_grid_mat(self.device)

x_grid = torch.einsum("bai,zic->zbac", to_grid_mat, x_res)
x_grid = act(x_grid)
x_res = torch.einsum("bai,zbac->zic", from_grid_mat, x_grid)
Expand Down
5 changes: 3 additions & 2 deletions src/fairchem/core/models/escn/so3_exportable.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -250,6 +250,7 @@ def __init__(
mmax: int,
normalization: str = "integral",
resolution: int | None = None,
rescale: bool = False,
):
super().__init__()

Expand All @@ -276,7 +277,7 @@ def __init__(
)
to_grid_mat = torch.einsum("mbi, am -> bai", to_grid.shb, to_grid.sha).detach()
# rescale based on mmax
if lmax != mmax:
if rescale and lmax != mmax:
for lval in range(lmax + 1):
if lval <= mmax:
continue
Expand All @@ -300,7 +301,7 @@ def __init__(
"am, mbi -> bai", from_grid.sha, from_grid.shb
).detach()
# rescale based on mmax
if lmax != mmax:
if rescale and lmax != mmax:
for lval in range(lmax + 1):
if lval <= mmax:
continue
Expand Down
15 changes: 4 additions & 11 deletions tests/core/models/test_escn_compiles.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,7 +26,6 @@
from fairchem.core.datasets import data_list_collater
from fairchem.core.models.escn import escn_exportable
from fairchem.core.models.escn.so3_exportable import (
CoefficientMapping,
SO3_Grid,
)
from fairchem.core.models.scn.smearing import GaussianSmearing
Expand Down Expand Up @@ -70,8 +69,8 @@ def load_model(type: str, compile=False, export=False):
cutoff=CUTOFF,
max_num_elements=MAX_ELEMENTS,
num_layers=8,
lmax_list=[4],
mmax_list=[2],
lmax_list=[6],
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

maybe leave a note here to mention some additional things we can add to make this test more sensitive: add backwards, add better weight initialization etc

mmax_list=[0],
sphere_channels=128,
hidden_channels=256,
edge_channels=128,
Expand All @@ -87,8 +86,8 @@ def load_model(type: str, compile=False, export=False):
cutoff=CUTOFF,
max_num_elements=MAX_ELEMENTS,
num_layers=8,
lmax=4,
mmax=2,
lmax=6,
mmax=0,
sphere_channels=128,
hidden_channels=256,
edge_channels=128,
Expand Down Expand Up @@ -210,7 +209,6 @@ def test_escn_so2_conv_exports_and_compiles(self, tol=1e-5) -> None:
}

lmax, mmax = 4, 2
mappingReduced = escn_exportable.CoefficientMapping([lmax], [mmax])
shpere_channels = 128
edge_channels = 128
args = (torch.rand(680, 19, shpere_channels), torch.rand(680, edge_channels))
Expand All @@ -222,7 +220,6 @@ def test_escn_so2_conv_exports_and_compiles(self, tol=1e-5) -> None:
lmax=lmax,
mmax=mmax,
act=torch.nn.SiLU(),
mappingReduced=mappingReduced,
)
prog = export(so2, args=args, dynamic_shapes=dynamic_shapes1)
export_out = prog.module()(*args)
Expand All @@ -244,7 +241,6 @@ def test_escn_message_block_exports_and_compiles(self, tol=1e-5) -> None:
SO3_grid = torch.nn.ModuleDict()
SO3_grid["lmax_lmax"] = SO3_Grid(lmax, lmax)
SO3_grid["lmax_mmax"] = SO3_Grid(lmax, mmax)
mappingReduced = CoefficientMapping([lmax], [mmax])
message_block = escn_exportable.MessageBlock(
layer_idx=0,
sphere_channels=sphere_channels,
Expand All @@ -256,7 +252,6 @@ def test_escn_message_block_exports_and_compiles(self, tol=1e-5) -> None:
max_num_elements=90,
SO3_grid=SO3_grid,
act=torch.nn.SiLU(),
mappingReduced=mappingReduced,
)

# generate inputs
Expand Down Expand Up @@ -297,7 +292,6 @@ def test_escn_layer_block_exports_and_compiles(self, tol=1e-5) -> None:
SO3_grid = torch.nn.ModuleDict()
SO3_grid["lmax_lmax"] = SO3_Grid(lmax, lmax)
SO3_grid["lmax_mmax"] = SO3_Grid(lmax, mmax)
mappingReduced = CoefficientMapping([lmax], [mmax])
layer_block = escn_exportable.LayerBlock(
layer_idx=0,
sphere_channels=sphere_channels,
Expand All @@ -309,7 +303,6 @@ def test_escn_layer_block_exports_and_compiles(self, tol=1e-5) -> None:
max_num_elements=90,
SO3_grid=SO3_grid,
act=torch.nn.SiLU(),
mappingReduced=mappingReduced,
)

# generate inputs
Expand Down