Merge branch 'jafioti:main' into feature-conv3d

examples/train_math_net/Cargo.toml

@@ -7,10 +7,12 @@ edition = "2021"
 
 [features]
 metal = ["dep:luminal_metal"]
+cuda = ["dep:luminal_cuda"]
 
 [dependencies]
 luminal = {path="../.."}
 luminal_training = {path="../../crates/luminal_training"}
 luminal_nn =  {path="../../crates/luminal_nn"}
 rand = "0.8.5"
-luminal_metal = {path="../../crates/luminal_metal", optional=true}
+luminal_metal = {path="../../crates/luminal_metal", optional=true}
+luminal_cuda = {path="../../crates/luminal_cuda", optional=true}

examples/train_math_net/src/main.rs

@@ -19,8 +19,11 @@ fn main() {
     let mut weights = params(&model);
     let grads = cx.compile(Autograd::new(&weights, loss), ());
     let (mut new_weights, lr) = sgd_on_graph(&mut cx, &weights, &grads);
+    cx.keep_tensors(&new_weights);
+    cx.keep_tensors(&weights);
     lr.set(1e-1);
 
+    #[cfg(all(not(feature = "metal"), not(feature = "cuda")))]
     cx.compile(
         GenericCompiler::default(),
         (
@@ -46,6 +49,19 @@ fn main() {
         ),
     );
 
+    #[cfg(feature = "cuda")]
+    cx.compile(
+        luminal_cuda::CudaCompiler::<f32>::default(),
+        (
+            &mut input,
+            &mut target,
+            &mut loss,
+            &mut output,
+            &mut weights,
+            &mut new_weights,
+        ),
+    );
+
     let mut rng = thread_rng();
     let (mut loss_avg, mut acc_avg) = (ExponentialAverage::new(1.0), ExponentialAverage::new(0.0));
     let mut iter = 0;