Tamas Bela Feher

[FEA] ANN bench improvements

1

During porting the ANN benchmarks from RAFT (#130), a list of potential improvements was created: - [ ] clang-tidy: cleanup naming conventions - [ ] cleanup/unify logging (including CPU-only builds)...

[FEA] Support int64_t nearest neighbor index type for all ANN algorithm.

1

Currently most of our the nearest neighbors methods support "int64_t" type for the nearest neighbor indices. The only exception is CAGRA, that returns `uint32_t` indices. For consistency, it would be...

TF-TRT Dynamic Shapes Feature Tracker

7

### Introduction The dynamic shape mode in TF-TRT utilizes TensorRT’s [dynamic shape feature](https://docs.nvidia.com/deeplearning/tensorrt/developer-guide/index.html#work_dynamic_shapes) to improve the conversion rate of networks and handle networks with unknown input shapes efficiently. This issue...

Reduce cagra build binary size

2

During `cagra::build` we use either `ivf_pq` or `nn_descent`. Until now we used these algorithm through the `detail` namespace, which resulted in recompilation. This PR changes to the public interface of...

Document whether the GPU execution of concrete functions are asynchronous

7

## URL(s) with the issue: - https://www.tensorflow.org/guide/eager#performance - https://www.tensorflow.org/guide/intro_to_graphs#seeing_the_speed-up - https://www.tensorflow.org/api_docs/python/tf/experimental/async_scope ## Description of issue (what needs changing): The comment in the example code at https://www.tensorflow.org/guide/eager#performance states: ``` # tf.matmul...

Asset initialization using SavedModel C++ API

2

**System information** - Have I written custom code: yes - OS Platform and Distribution: Ubuntu 20.04 - TensorFlow installed from: source, commit 1dd2070b5f9 - TensorFlow version: 2.8 - Python version:...