make candle available (#886)

Cargo.toml

@@ -6,6 +6,7 @@ resolver = "2"
 members = [
     "burn",
     "burn-autodiff",
+    "burn-candle",
     "burn-common",
     "burn-compute",
     "burn-core",

README.md

@@ -26,7 +26,7 @@ simplifying the process of experimenting, training, and deploying models.
     [`no_std`](#support-for-no_std) compatibility, ensuring universal platform adaptability 👌
   - [WebGPU](https://github.com/burn-rs/burn/tree/main/burn-wgpu) backend, offering cross-platform,
     browser-inclusive, GPU-based computations 🌐
-  - [Candle](https://github.com/burn-rs/burn/tree/main/burn-candle) backend (alpha) 🕯️
+  - [Candle](https://github.com/burn-rs/burn/tree/main/burn-candle) backend 🕯️
   - [Autodiff](https://github.com/burn-rs/burn/tree/main/burn-autodiff) backend that enables
     differentiability across all backends 🌟
 - [Dataset](https://github.com/burn-rs/burn/tree/main/burn-dataset) crate containing a diverse range

backend-comparison/Cargo.toml

@@ -12,6 +12,7 @@ version = "0.10.0"
 [features]
 default = ["std"]
 std = []
+candle = ["burn/candle"]
 ndarray = ["burn/ndarray"]
 ndarray-blas-accelerate = ["burn/ndarray-blas-accelerate"]
 ndarray-blas-netlib = ["burn/ndarray-blas-netlib"]

backend-comparison/src/lib.rs

@@ -40,5 +40,14 @@ macro_rules! bench_on_backend {
             let device = NdArrayDevice::Cpu;
             bench::<NdArrayBackend>(&device);
         }
+
+        #[cfg(feature = "candle")]
+        {
+            use burn::backend::candle::CandleDevice;
+            use burn::backend::CandleBackend;
+
+            let device = CandleDevice::Cpu;
+            bench::<CandleBackend>(&device);
+        }
     };
 }

burn-core/Cargo.toml

@@ -52,6 +52,8 @@ wgpu = ["burn-wgpu/default"]
 
 tch = ["burn-tch"]
 
+candle = ["burn-candle"]
+
 # Serialization formats
 experimental-named-tensor = ["burn-tensor/experimental-named-tensor"]
 
@@ -72,6 +74,7 @@ burn-ndarray = { path = "../burn-ndarray", version = "0.10.0", optional = true,
 burn-autodiff = { path = "../burn-autodiff", version = "0.10.0", optional = true }
 burn-wgpu = { path = "../burn-wgpu", version = "0.10.0", optional = true }
 burn-tch = { path = "../burn-tch", version = "0.10.0", optional = true }
+burn-candle = { path = "../burn-candle", version = "0.10.0", optional = true }
 
 derive-new = { workspace = true }
 libm = { workspace = true }

burn-core/src/backend.rs

@@ -23,6 +23,18 @@ pub type WgpuBackend<G = wgpu::AutoGraphicsApi, F = f32, I = i32> = wgpu::WgpuBa
 pub type WgpuAutodiffBackend<G = wgpu::AutoGraphicsApi, F = f32, I = i32> =
     crate::autodiff::ADBackendDecorator<WgpuBackend<G, F, I>>;
 
+#[cfg(feature = "candle")]
+/// Candle module.
+pub use burn_candle as candle;
+
+#[cfg(feature = "candle")]
+/// A CandleBackend with a default type of f32/i64.
+pub type CandleBackend = candle::CandleBackend<f32, i64>;
+
+#[cfg(all(feature = "candle", feature = "autodiff"))]
+/// A CandleBackend with autodiffing enabled.
+pub type CandleAutodiffBackend = crate::autodiff::ADBackendDecorator<CandleBackend>;
+
 #[cfg(feature = "tch")]
 /// Tch module.
 pub use burn_tch as tch;

burn-wgpu/README.md

@@ -5,7 +5,7 @@
 [![Current Crates.io Version](https://img.shields.io/crates/v/burn-wgpu.svg)](https://crates.io/crates/burn-wgpu)
 [![license](https://shields.io/badge/license-MIT%2FApache--2.0-blue)](https://github.com/burn-rs/burn-wgpu/blob/master/README.md)
 
-This crate provides a WGPU backend for [Burn](https://github.com/burn-rs/burn) utilizing the
+This crate provides a WGPU backend for [Burn](https://github.com/burn-rs/burn) using the
 [wgpu](https://github.com/gfx-rs/wgpu). 
 
 The backend supports Vulkan, Metal, DirectX11/12, OpenGL, WebGPU.

burn/Cargo.toml

@@ -45,6 +45,7 @@ ndarray-blas-openblas-system = ["burn-core/ndarray-blas-openblas-system"]
 
 wgpu = ["burn-core/wgpu"]
 tch = ["burn-core/tch"]
+candle = ["burn-core/candle"]
 
 # Experimental
 experimental-named-tensor = ["burn-core/experimental-named-tensor"]