Repository navigation
What's Changed
- feat: Pool embeddings in parallel by @Pringled in #53
- feat: Add IDs-only fast WordPiece tokenizer by @Jeadie in #54
- fix: move unk order of operations to match Python by @stephantul in #57
- feat: Add IDs-only fast Unigram model by @Karthik777 in #58
- feat: add faster pooling by @stephantul in #59
- feat: add int8 storage by @stephantul in #60
- feat: Quantize f32 models on the fly by @stephantul in #61
- feat: add nanobeir bench by @stephantul in #62
- feat: up batch size by @stephantul in #63
- feat: use wordpiece trie by @stephantul in #64
- feat: add bekko embedding to benchmark by @stephantul in #65
- feat: add optimization for blocks by @stephantul in #66
- feat: Much faster loading by @stephantul in #67
- feat: fast unicode path by @stephantul in #69
- fix: out of bounds issues with added tokens/truncation/padding by @stephantul in #70
- chore: update benchmark in readme by @stephantul in #71
- chore: add trusted publishing release workflow by @Pringled in #72
- chore: Bump version by @Pringled in #73
New Contributors
- @Jeadie made their first contribution in #54
- @stephantul made their first contribution in #57
- @Karthik777 made their first contribution in #58
Full Changelog: v0.3.0...v0.4.0