ferrotensor 0.2.1__tar.gz → 0.4.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/Cargo.lock +11 -11
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/Cargo.toml +1 -1
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/PKG-INFO +1 -1
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/Cargo.toml +1 -1
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/backend/compute.rs +16 -2
- ferrotensor-0.4.0/ferrotensor/src/backend/kernels.rs +787 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/functional.rs +112 -16
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/lib.rs +3 -0
- ferrotensor-0.4.0/ferrotensor/src/random.rs +77 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/tensor/core.rs +4 -2
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/tensor/cpu.rs +150 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/tensor/ops.rs +26 -76
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/tensor/raw.rs +353 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/tensor/tests.rs +193 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/src/lib.rs +1 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/src/tensor.rs +6 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/tests/test_tensor.py +14 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/tests/test_train.py +3 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/python/ferrotensor/__init__.py +2 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/python/ferrotensor/_ferrotensor.pyi +6 -0
- ferrotensor-0.2.1/ferrotensor/src/backend/kernels.rs +0 -405
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/README.md +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/README.md +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/examples/README.md +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/examples/char_transformer.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/examples/data/sonnets.txt +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/examples/gpu_matmul_attention.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/examples/mlp_two_moons.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/examples/quickstart.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/examples/similarity_search.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/backend/mod.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/data.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/device.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/nn/activation.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/nn/attention.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/nn/container.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/nn/dropout.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/nn/embedding.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/nn/linear.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/nn/mod.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/nn/module.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/nn/normalization.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/nn/tests.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/optim/adam.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/optim/mod.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/optim/optimizer.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/optim/scheduler.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/optim/sgd.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/optim/tests.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/serialize.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/tensor/autograd.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/tensor/mod.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor/src/tensor/shape.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/.gitignore +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/Cargo.toml +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/README.md +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/build.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/examples/mlp.py +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/src/device.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/src/error.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/src/functional.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/src/nn.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/src/optim.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/src/serialize.rs +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/tests/conftest.py +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/tests/test_autograd.py +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/tests/test_errors.py +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/tests/test_functional.py +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/tests/test_gpu.py +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/tests/test_nn.py +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/tests/test_optim.py +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/ferrotensor-py/tests/test_serialize.py +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/pyproject.toml +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/python/ferrotensor/nn/__init__.py +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/python/ferrotensor/nn/functional.py +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/python/ferrotensor/optim.py +0 -0
- {ferrotensor-0.2.1 → ferrotensor-0.4.0}/python/ferrotensor/py.typed +0 -0
|
@@ -606,7 +606,7 @@ dependencies = [
|
|
|
606
606
|
|
|
607
607
|
[[package]]
|
|
608
608
|
name = "ferrotensor"
|
|
609
|
-
version = "0.
|
|
609
|
+
version = "0.4.0"
|
|
610
610
|
dependencies = [
|
|
611
611
|
"bytemuck",
|
|
612
612
|
"naga",
|
|
@@ -619,7 +619,7 @@ dependencies = [
|
|
|
619
619
|
|
|
620
620
|
[[package]]
|
|
621
621
|
name = "ferrotensor-autograd"
|
|
622
|
-
version = "0.
|
|
622
|
+
version = "0.4.0"
|
|
623
623
|
dependencies = [
|
|
624
624
|
"ferrotensor-backend",
|
|
625
625
|
"pollster",
|
|
@@ -628,7 +628,7 @@ dependencies = [
|
|
|
628
628
|
|
|
629
629
|
[[package]]
|
|
630
630
|
name = "ferrotensor-backend"
|
|
631
|
-
version = "0.
|
|
631
|
+
version = "0.4.0"
|
|
632
632
|
dependencies = [
|
|
633
633
|
"bytemuck",
|
|
634
634
|
"env_logger",
|
|
@@ -641,7 +641,7 @@ dependencies = [
|
|
|
641
641
|
|
|
642
642
|
[[package]]
|
|
643
643
|
name = "ferrotensor-benchmarks"
|
|
644
|
-
version = "0.
|
|
644
|
+
version = "0.4.0"
|
|
645
645
|
dependencies = [
|
|
646
646
|
"ferrotensor",
|
|
647
647
|
"ferrotensor-backend",
|
|
@@ -652,7 +652,7 @@ dependencies = [
|
|
|
652
652
|
|
|
653
653
|
[[package]]
|
|
654
654
|
name = "ferrotensor-cli"
|
|
655
|
-
version = "0.
|
|
655
|
+
version = "0.4.0"
|
|
656
656
|
dependencies = [
|
|
657
657
|
"clap",
|
|
658
658
|
"env_logger",
|
|
@@ -670,7 +670,7 @@ dependencies = [
|
|
|
670
670
|
|
|
671
671
|
[[package]]
|
|
672
672
|
name = "ferrotensor-graph"
|
|
673
|
-
version = "0.
|
|
673
|
+
version = "0.4.0"
|
|
674
674
|
dependencies = [
|
|
675
675
|
"bytemuck",
|
|
676
676
|
"env_logger",
|
|
@@ -685,7 +685,7 @@ dependencies = [
|
|
|
685
685
|
|
|
686
686
|
[[package]]
|
|
687
687
|
name = "ferrotensor-hub"
|
|
688
|
-
version = "0.
|
|
688
|
+
version = "0.4.0"
|
|
689
689
|
dependencies = [
|
|
690
690
|
"ferrotensor-autograd",
|
|
691
691
|
"ferrotensor-backend",
|
|
@@ -698,7 +698,7 @@ dependencies = [
|
|
|
698
698
|
|
|
699
699
|
[[package]]
|
|
700
700
|
name = "ferrotensor-loader"
|
|
701
|
-
version = "0.
|
|
701
|
+
version = "0.4.0"
|
|
702
702
|
dependencies = [
|
|
703
703
|
"bytemuck",
|
|
704
704
|
"env_logger",
|
|
@@ -716,7 +716,7 @@ dependencies = [
|
|
|
716
716
|
|
|
717
717
|
[[package]]
|
|
718
718
|
name = "ferrotensor-monitor"
|
|
719
|
-
version = "0.
|
|
719
|
+
version = "0.4.0"
|
|
720
720
|
dependencies = [
|
|
721
721
|
"anyhow",
|
|
722
722
|
"chrono",
|
|
@@ -730,7 +730,7 @@ dependencies = [
|
|
|
730
730
|
|
|
731
731
|
[[package]]
|
|
732
732
|
name = "ferrotensor-py"
|
|
733
|
-
version = "0.
|
|
733
|
+
version = "0.4.0"
|
|
734
734
|
dependencies = [
|
|
735
735
|
"ferrotensor",
|
|
736
736
|
"pyo3",
|
|
@@ -739,7 +739,7 @@ dependencies = [
|
|
|
739
739
|
|
|
740
740
|
[[package]]
|
|
741
741
|
name = "ferrotensor-trainer"
|
|
742
|
-
version = "0.
|
|
742
|
+
version = "0.4.0"
|
|
743
743
|
dependencies = [
|
|
744
744
|
"bytemuck",
|
|
745
745
|
"ferrotensor-autograd",
|
|
@@ -9,14 +9,21 @@ use wgpu::{BindGroup, BindGroupLayout, BindingResource, CommandEncoder, ComputeP
|
|
|
9
9
|
use super::KERNELS_WGSL;
|
|
10
10
|
|
|
11
11
|
/// Entry points compiled from [`KERNELS_WGSL`].
|
|
12
|
-
pub(crate) const KERNEL_NAMES: [&str;
|
|
12
|
+
pub(crate) const KERNEL_NAMES: [&str; 14] = [
|
|
13
13
|
"unary",
|
|
14
14
|
"binary",
|
|
15
15
|
"reduce",
|
|
16
|
+
"reduce_all",
|
|
17
|
+
"arg_reduce",
|
|
16
18
|
"softmax",
|
|
17
19
|
"matmul",
|
|
18
20
|
"permute",
|
|
19
21
|
"gather_rows",
|
|
22
|
+
"sort_ids",
|
|
23
|
+
"scatter_rows",
|
|
24
|
+
"copy_block",
|
|
25
|
+
"cross_entropy_rows",
|
|
26
|
+
"cross_entropy_grad",
|
|
20
27
|
];
|
|
21
28
|
|
|
22
29
|
/// Threads per workgroup of the 1-D kernels.
|
|
@@ -129,7 +136,14 @@ impl ComputeEngine {
|
|
|
129
136
|
/// Spills into the y dimension past 65535 workgroups; the kernels rebuild
|
|
130
137
|
/// the flat index as `x + y * nx * 256`.
|
|
131
138
|
pub fn workgroups_1d(n: usize) -> (u32, u32, u32) {
|
|
132
|
-
|
|
139
|
+
Self::workgroup_grid(n.div_ceil(WORKGROUP_1D as usize))
|
|
140
|
+
}
|
|
141
|
+
|
|
142
|
+
/// Grid of at least `groups` workgroups (at least one), spilling into the
|
|
143
|
+
/// y dimension past 65535; the kernels rebuild the workgroup index as
|
|
144
|
+
/// `x + y * nx`.
|
|
145
|
+
pub fn workgroup_grid(groups: usize) -> (u32, u32, u32) {
|
|
146
|
+
let groups = u32::try_from(groups.max(1)).expect("too many workgroups");
|
|
133
147
|
if groups <= MAX_GROUPS_PER_DIM {
|
|
134
148
|
(groups, 1, 1)
|
|
135
149
|
} else {
|