bump candle to 0.3.1 and conv_transpose_1d (#977)

burn-candle/Cargo.toml

@@ -18,8 +18,8 @@ derive-new = { workspace = true }
 burn-tensor = { path = "../burn-tensor", version = "0.11.0", default-features = false }
 half = { workspace = true }
 
-# TODO remove pinned version ("=") once candle-core is updated to 0.3.1
-candle-core = { version = "=0.3.0" }
+candle-core = { version = "0.3.1" }
+
 
 [dev-dependencies]
 burn-autodiff = { path = "../burn-autodiff", version = "0.11.0", default-features = false, features = [

burn-candle/src/backend.rs

@@ -50,6 +50,7 @@ impl From<candle_core::Device> for CandleDevice {
         match device.location() {
             DeviceLocation::Cpu => CandleDevice::Cpu,
             DeviceLocation::Cuda { gpu_id } => CandleDevice::Cuda(gpu_id),
+            DeviceLocation::Metal => panic!("Metal unsupported"),
         }
     }
 }

burn-candle/src/ops/module.rs

@@ -83,7 +83,26 @@ impl<F: FloatCandleElement, I: IntCandleElement> ModuleOps<Self> for Candle<F, I
         bias: Option<FloatTensor<Self, 1>>,
         options: ConvTransposeOptions<1>,
     ) -> FloatTensor<Self, 3> {
-        panic!("Candle does not support conv_transpose1d")
+        assert!(
+            options.groups == 1,
+            "Candle does not support groups in transposed convolutions"
+        );
+        let conv_transpose = x
+            .tensor
+            .conv_transpose1d(
+                &weight.tensor,
+                options.padding[0],
+                options.padding_out[0],
+                options.stride[0],
+                options.dilation[0],
+            )
+            .unwrap();
+        CandleTensor::new(match bias {
+            Some(bias) => conv_transpose
+                .broadcast_add(&bias.tensor.unsqueeze(0).unwrap().unsqueeze(2).unwrap())
+                .unwrap(),
+            None => conv_transpose,
+        })
     }
 
     fn conv_transpose2d(