Releases: ZurichNLP/mbr
Releases · ZurichNLP/mbr
v0.2.0
Breaking changes:
- Rename
MBRGenerationConfig
toMBRConfig
MetricRunner
now returns aMetricOutput
dict instead of the raw tensor of scores.
New features:
- Make the size of the metric cache configurable via
MBRConfig.metric_cache_size
- Allow that the number of references can be larger than the number of samples (if generated separately from the samples)
Other changes:
- Remove
GenerationConfig
as parent class ofMBRConfig
- Use FIFO cache instead of LRU for metric cache
- Extend caching to
CometMetricRunner
Initial release
v0.1.0 Bump version to 0.1.0