All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- Update candle from 0.4.1 to 0.5.0
- Update candle from 0.3.1 to 0.4.1
Nx.argsort
supports tensors rank = 2 (#66)Nx.dot
supports batched axes (#68)Nx.window_sum
support (#71)Nx.dot
more complete support for M x N tensors operation (#73)
Nx.put_slice
, more complete support (#46)Nx.dot
, support n x m (#51)Nx.take
supports indices rank > 1 (#57)Nx.argsort
basic support, only CPU (#59)Nx.pad
, support tensors of rank > 1 (#61)Nx.pad
supports negative padding (#62)Nx.window_max
supports:same
padding (#64)
- Nx.conv: partially support padding and strides options
- Nx.window_max: initial partial support
- Precompiled binary for ARMv6 gnueabihf
- Support Nx.reverse
- Fixes wrong behavior in Nx.slice when passing Nx.Tensor as start_indices
- Precompiled binaries name fix
- Precompiled binary for CUDA 12.x target.
- Precompiled binaries for a few CPU targets.