Releases: k2-fsa/k2
Releases · k2-fsa/k2
Release v1.15
- Implement linear_fsa_with_self_loops.
- Allow different encoder/decoder dims for RNN-T.
- Recover windows build.
- Bug fixes.
Release v1.14
- Support modified RNN-T loss.
- Support parallel RNN-T decoding.
Release v1.13
Change rnnt_loss
interface.
Release v1.12
- Support RNN-T loss.
Release v1.11
Bugfixes for computing framewise alignments.
v1.10
Mainly bugfixes.
Release v1.9
- Support creating ragged tensor from regular tensor.
- Add levenshtein alignment.
Release v1.8
- Bug fixes
- Change
data
property ofk2.RaggedTensor
tovalues
Release v1.7
- Bug fixes
- Replace
k2.RaggedInt
andk2.RaggedFloat
withk2.RaggedTensor
- Documentation ULR is changed to https://k2-fsa.github.io/k2/
Caution: This version is not backward compatible.
Release v1.4
Mainly bugfixes.