Fix pytorch recorder adapt_linear when using autodiff backend (#1576)

* Fix pytorch recorder adapt_linear when using autodiff backend

* Fix comment typo

crates/burn-import/pytorch-tests/Cargo.toml

@@ -7,6 +7,7 @@ license.workspace = true
 [dev-dependencies]
 burn = { path = "../../burn" }
 burn-ndarray = { path = "../../burn-ndarray" }
+burn-autodiff = { path = "../../burn-autodiff" }
 serde = { workspace = true }
 float-cmp = { workspace = true }
 burn-import = { path = "../", features = ["pytorch"] }

crates/burn-import/pytorch-tests/tests/complex_nested/mod.rs

@@ -12,6 +12,7 @@ use burn::{
         Tensor,
     },
 };
+use burn_autodiff::Autodiff;
 use burn_import::pytorch::{LoadArgs, PyTorchFileRecorder};
 
 #[derive(Module, Debug)]
@@ -164,6 +165,17 @@ fn full_record() {
     model_test(record, 8);
 }
 
+#[test]
+fn full_record_autodiff() {
+    let device = Default::default();
+    let record = PyTorchFileRecorder::<FullPrecisionSettings>::default()
+        .load("tests/complex_nested/complex_nested.pt".into(), &device)
+        .expect("Should decode state successfully");
+
+    let device = Default::default();
+    let _model = Net::<Autodiff<TestBackend>>::init(&device).load_record(record);
+}
+
 #[test]
 fn half_record() {
     let device = Default::default();

crates/burn-import/src/pytorch/adapter.rs

@@ -36,6 +36,8 @@ impl<PS: PrecisionSettings, B: Backend> BurnModuleAdapter for PyTorchAdapter<PS,
             .try_into_record::<_, PS, DefaultAdapter, B>(&B::Device::default())
             .expect("Failed to deserialize weight");
 
+        // Do not capture transpose op when using autodiff backend
+        let weight = weight.set_require_grad(false);
         // Transpose the weight tensor.
         let weight_transposed = Param::from_tensor(weight.val().transpose());