bumblebee

Fixed

• Fix padded batch items lengthening generation time (#419)

Added

• Support for passing string pairs as input to text classification serving
• Support for partial rotary embedding for Phi3

Changed

• Updated Phoenix examples to use phoenix_playground (#400)

Fixed

• Fixed NLLB model crashing on batched input
• Out-of-bound position ids with left padded generation input

Added

• Notebook about Retrieval-Augmented Generation (RAG) (#353)
• Phi model (#356)
• Gemma model (#358)
• Support for input streaming in Whisper serving (#361)
• Stable Diffusion Con...

Fixed

• Loading generation config with nil attributes
• Generating with :no_repeat_ngram_length when using lower precision

Fixed

• Fixed loading sharded parameters in the safetensors format