v0.2.0

Added

• Support for Stable Diffusion v2 (#117)
• CamemBERT model (#110)
• XLM-RoBERTa model (#136)
• Support for configuring backend used by load_model (#140)
• Zero-shot classification serving (#121, #145)
• Support parameter files using Zip64 (#144)
• Notebook exemplifying fine-tuning with Axon (#102)
• Whisper model (#107)
• Speech-to-text serving (#107)
• Support for loading tokenizers with overridden special tokens (#141)
• Question answering serving (#157)
• T5 model (#159)
• Conversational serving (#165)
• DistilBERT model (#172)

Changed

• Cancel download request when the caller terminates (#135)
• Introduced explicit parameter mapping, which changed parameter names in all models (#148)

Fixed

• Initialization of parameters with mismatched shape

v0.1.2

Fixed

• Stable Diffusion to accept a string as the input (#115)

v0.1.1

Added

• Support for a negative prompt in Stable Diffusion input (#109)

Changed

• Improved fill-mask output token to not include whitespace or whitespace placeholder (#106)

Fixed

• Image rendering in the Stable Diffusion notebook (#95)
• Download error when the tmp and cache directories are on different file systems (#98)

v0.1.0

Initial release.