---------

Co-authored-by: Administrator <Administrator@tech8>
Co-authored-by: KexinFeng <fenkexin@amazon.com>

* Implement PtNDArraryEx.multiboxDetection

* MultiboxDetection - code cleanup

* MultiboxDetection - code cleanup

* MultiboxDetection - code cleanup

* MultiboxDetection - code cleanup

* format code

* Fix, add tests, and pass CI

---------

Co-authored-by: Zach Kimberg <kimbergz@amazon.com>

…brary#2796)

This reverts commit 3a90d0a.

This fixes the markdown headers to be h1 so they render correctly in docs.

…y#2818)

…valibrary#2806)

* [api] Added Early stopping configuration (deepjavalibrary#38)

* [api] Added Builder for Early stopping configuration (deepjavalibrary#38)

* Explicitly set NDManager for dataset in EarlyStoppingListenerTest to make the test run on JDK11 in gradle.

This creates an abstraction for combining devices into a single device. The main
use case for now is in DJL Serving TP_parallel. It will allow us to create a
WorkerGroup and a PyPredictor for a set of devices and then track the usage of
devices properly. It could also be used later for multi-gpu training or other
multi-device cases.

…alibrary#2826)

* Updates doc versions to 0.24.0

Also moves android gradle.properties to the new 0.25.0.

* Remove android change

* Updates XGBoost to 2.0.1

* Use devtools 8

* Updates based on new Xgboost JNI API.

---------

Co-authored-by: Frank Liu <frankfliu2000@gmail.com>

Fixes deepjavalibrary#2840

* Added element-wise gauss error function (ERF)

* Added element-wise arctan2

* Format java

* Fixed docs

* added * to other_ptr in Atan2

* Added 2D FFT

* Format java

* Add default fft2

* Convert array to vectors

* Add inverse fft2

* Add better assersion in ifft2 test

* Add really better assersion in ifft2 test

* Move cast bellow ifft2 for unsupported exception

* Format java

* changed dims to axes

* changed dims to axes

* only build triton binaries

* install requests library

* remove script

…javalibrary#2850)

Updates the navigation as a followup to
deepjavalibrary/djl-serving#1316.

)

This property is necessary to avoid conflicts in building multiple flavors which
can be observed in the native properties file not matching the jar for some
engine builds.

…de (deepjavalibrary#2865)

This changes the test from requiring the early stopping at a particular epoch to
ensuring that it stops early due to time constraints. This should help reduce
the chances of being flaky due to race conditions.

* [docs] Move jupyter notebooks to DJL Demo

This moves the notebooks to the demo repositories. It makes it more clearly
follow the trend where the things in the main repo are master build while those
in demos track nightly snapshots or releases.

)

Fixes: deepjavalibrary#2875

…2867)

…library#2871)

…javalibrary#2872)

…eepjavalibrary#2877)

* Support 32-bit toTensor()

This fixes an issue where calling NDArrayEx.toTensor() fails on Apple Silicon due to a lack of support for float64.

* Avoid float64 conversion in Classifications constructor

Don't convert probabilities from float32 to float64, because this causes a failure on Apple Silicon.

Co-authored-by: Frank Liu <frankfliu2000@gmail.com>

…ls as well as pytorch (deepjavalibrary#2880)

* Updated huggingface_models.py to support Safetensors models as well as pytorch
---------

Co-authored-by: Frank Liu <frankfliu2000@gmail.com>

…rary#2881)

…rary#2884)

…library#2882)

…brary#2892)

…y#2894)

* Evaluator support to update multiple accumulators

Improve the performance of EvaluatorTrainingListener by enabling evaluators to update multiple accumulators from the same labels and predictions, rather than needing to recompute values.

* Fix formatting

* Update AbstractCompositeLoss.java

Aims to fix failing test

* Adds llama.cpp engine

* [llama.cpp] Adds CI publish workflow and fixes windows build

---------

Co-authored-by: Frank Liu <frankfliu2000@gmail.com>

* Yelov8 Translator optimization

- Improved post-processing performance up to 40x by reducing expensive native calls
- Additional argument 'maxBox' added to improve post-processing performance by reducing number of considered bounding boxes
- Sunset file fixed,, previous version ignored first 4 rows, so recognized classes were 4 off. Adding 4 rows header fixes the problem. New headers pointing to ultralytics doc pages and original coco dataset page.

* addressed PR comments

- reformatted code
- removed final

* Refactor Yolov8 example

1. Fixes windows line return
2. Furthe reduce NDArray operations
3. Refactor example code
4. Add unittest for Yolov8Detection

---------

Co-authored-by: Frank Liu <frankfliu2000@gmail.com>

)

…library#2905)

* [llama.cpp] Adds unit-test and standardize input parameters

* [llama.cpp] Refactors llama input format

…2913)

…#2920)

Co-authored-by: Aaqib Ansari <mohaan@amazon.com>

This improves upon the creation of MultiDevice in deepjavalibrary#2819 by moving the getDevices
function to the main Device class. This can simplify the usage of something
which is potentially a MultiDevice and make it easier to check for the presence
of a MultiDevice.

* Creates DJL manual engine initialization

fixes deepjavalibrary#2875
reverts deepjavalibrary#2876

This adds new support for DJL manual initialization of engines to support
`DJL_ENGINE_MANUAL_INIT`. Once done, no engines providers will be found or
loaded on startup. Instead, they can be added manually by:

```java
PtEngineProvider provider = new PtEngineProvider();
provider.getEngine(); // Optional, throws exception if the provider can not load
Engine.registerEngine(provider);
Engine.setDefaultEngine(provider.getEngineName()); // Optional, sets as default
```

* Revert "[tensorflow] Revert InstanceHolder for TensorFlow engine (deepjavalibrary#2884)"

This reverts commit 586bb07.

* Revert "[api] Replace double-check singlton with lazy initialization (deepjavalibrary#2826)"

This reverts commit 3927867.

* Make engines initialized

This makes several updates:
- engines will now initialize once per instance of EngineProvider rather than
re-attempt to initialize
- Registering an engine can overwrite the existing one
- All engines now use the synchronized form rather than the static instance
holder. This allows them to have multiple versions and be local to the instance
rather than global (but the instance is saved globally)

* Throws Exception on bad getEngine

* Removes unnecessary check

…javalibrary#2929)

* Suppress serial warning for JDK21

In JDK21, it now throws the serial warning for including potentially
unserializable instance variables. This includes the standard Java data
structures like List, Set, and Map. This changes the JDK21 support from
deepjavalibrary#2903 to suppress the warning rather
than no longer serializing the variables.

* Keep CategoryMask as transient

…alibrary#2951)

Fixes: deepjavalibrary#2949

…2955)

…#2960)

…#2963)