Douglas Hanley

Reading Multiple Blocks

10

I'm encountering a slight issue when using the library: I'm only getting the first block of bz2 files. In my case, the block size is 900k, and when `read`ing off...

Expose icon sizes for adjustment by user (on different themes)

3

This is a really small one-liner, so I'll just describe it. I can make a pull request if you'd like. Currently the hamburger menu icon in the toolbar looks blurry,...

WIP: Parallel generation implemenation

3

This is a basic and relatively simple implementation of parallel generation, both streaming and non-streaming, as considered in #771. It sticks mostly to using the existing high level API functions,...

Add support for XLMRoberta embedding models

8

This adds support for XLMRoberta embedding models, such as `BAAI/bge-m3`. The inference is done entirely through regular BERT, and tokenization uses the new T5 Unigram work. There is some modification...

Fix multi-sequence embeddings

1

Fixes multi-sequence (batch) embeddings by handling `n_seq_max` and `kv_unified` flags. See discussion in #2051.