Iter2: store my output on the same device as the first input

In order for Iter2 to be usable with a GPU, you needed to do
Iter2::new(...).to_device(vs.device()).shuffle()

Now with this patch, the to_device is not necessary, which
is consistent with my expectations.

However, it has virtually no performance difference, positive
or negative.

CHANGELOG.md

@@ -3,6 +3,10 @@ All notable changes to this project will be documented in this file.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/) and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## Unreleased
+### Changed
+- Create `Iter2`'s output on the same device as its first parameter.
+
 ## v0.10.1 - 2022-12-12
 ### Changed
 - Default `vs.load` to use the Python weight format when the file extension is `.pt` or `.bin`.

src/data.rs

@@ -46,7 +46,7 @@ impl Iter2 {
             batch_index: 0,
             batch_size,
             total_size,
-            device: Device::Cpu,
+            device: xs.device(),
             return_smaller_last_batch: false,
         })
     }