Release: v0.1.0 - #10

Merged

olivier-peltre merged 51 commits into

mainfrom

Aug 14, 2026

Commits on Aug 14, 2026

refactor(options): harmonize TensorProduct option names
olivier-peltre
committed
2250abf
chore: rename to CUDATensorProductParams
olivier-peltre
committed
ea6a870
test: implement renames
olivier-peltre
committed
f4e25cb
docs: rename options
olivier-peltre
committed
936ee11
docs: document missing options
olivier-peltre
committed
12e7033
Transpose y in TRAILING_CHANNELS MAP tensor-product kernel (#328)

m-braganca
authored and
olivier-peltre
committed
10943dc
fix: static y_parity must be np.ndarray
olivier-peltre
committed
e0b0c52
refactor(core.TensorProduct): make implementation methods private
olivier-peltre
committed
fdaaf40
docs(test): update stale refs to private pathways
olivier-peltre
committed
42a7a03
perf(conv-fwd): transpose x ones so all access avoid sublane indexing

armandpicard
authored and
olivier-peltre
committed
77de3b4
fix(cuda:tp): coef pading

armandpicard
authored and
olivier-peltre
committed
ab68cf8
fix(cuda:conv): bwd coef pad

armandpicard
authored and
olivier-peltre
committed
1611f1b
simplify

armandpicard
authored and
olivier-peltre
committed
d36e897
add changelog entry

armandpicard
authored and
olivier-peltre
committed
002d16c
avoid jax initiliation on test collection

armandpicard
authored and
olivier-peltre
committed
58aa104
patch(config): detect JAX_PLATFORMS=cpu
olivier-peltre
committed
e489678
cleanup: docstring and platform default consts
olivier-peltre
committed
c081e0f
patch(config): update defaults with JAX_PLATFORMS='cpu,...' too
olivier-peltre
committed
4198e7f
fix: Array.blocks() must wrap yielded (mul, ir) in a Space, not a bare Irrep

armandpicard
authored and
olivier-peltre
committed
6098f76
fix: Array.__getitem__ must not silently reindex the feature axis

armandpicard
authored and
olivier-peltre
committed
54b2c05
fix: Array.__add__/__sub__ must check operands are alike

armandpicard
authored and
olivier-peltre
committed
e365fa4
fix: Array.__rmul__ must not crash on lower-rank broadcast scalars

armandpicard
authored and
olivier-peltre
committed
fdb68bc
fix: Array.__repr__ missing f-string prefix on layout suffix

armandpicard
authored and
olivier-peltre
committed
8860013
fix: Array.__getitem__ must count axes consumed by boolean index arrays

armandpicard
authored and
olivier-peltre
committed
6695c49
fix: Array.__getitem__ must not crash on plain numpy-array indices

armandpicard
authored and
olivier-peltre
committed
4164713
fix: Array._check_alike must not hardcode 'add' in __sub__'s error

armandpicard
authored and
olivier-peltre
committed
4ae669a
docs: document Array.__getitem__ usage

armandpicard
authored and
olivier-peltre
committed
b4532d1
precommit

armandpicard
authored and
olivier-peltre
committed
993d185
perf(conv-fwd): skip the channel pad when channels divides num_lanes (#369)

m-braganca
authored and
olivier-peltre
committed
8b79907
patch(ops.convolution): mask dummy edges with INT_MAX
olivier-peltre
committed
a230c55
patch(core.Convolution): accept optional node_mask
olivier-peltre
committed
895ca69
patch(lib/convolution): safely skip OOB as padding edges
olivier-peltre
committed
5dd0d53
fix(ops.convolution): vmap support, skipping padding edges
olivier-peltre
committed
106769b
docs: node_mask applies to trailing nodes only
olivier-peltre
committed
400d080
docs(Graph): clarify mask_edges docstring, padding => trailing
olivier-peltre
committed
d0804dc
test(convolution): parameterize masked vmap tests over graph ordering
olivier-peltre
committed
4ac43c5
fix(convolution): normalise the layout option in the constructor (#367)

m-braganca
authored and
olivier-peltre
committed
efc2171
feat(cuda): float16 and float64 value dtypes

a-pichard
authored and
olivier-peltre
committed
7c308f2
fix(ops): narrow the convolution adjacency to int32 at the FFI boundary

a-pichard
authored and
olivier-peltre
committed
494b92c
make Harmonics match e3nn

armandpicard
authored and
olivier-peltre
committed
39a65ce
fix jvp test on tpu.

armandpicard
authored and
olivier-peltre
committed
9b44d93
changelog

armandpicard
authored and
olivier-peltre
committed
ffd1d2c
rm debug notebook

armandpicard
authored and
olivier-peltre
committed
ef93612
refactor: use numpy to manipulate polynomials

armandpicard
authored and
olivier-peltre
committed
26d172d
feat: support normalization in Harmonics

armandpicard
authored and
olivier-peltre
committed
17a678a
docs: add streaming message-passing animation (rendered in docs build) (#335)

m-braganca
authored and
olivier-peltre
committed
231ea02
docs: update CHANGELOG
olivier-peltre
committed
de29fc8
build: bump version to v0.1.0
olivier-peltre
committed
20551ef
build: uv lock
olivier-peltre
committed
aef19e1
docs: update readme
olivier-peltre
committed
55e1258
docs: README.md
olivier-peltre
committed
e7ce436