v1.2.5
What's Changed
- fix choices selections by @urchade in #38
- Choices by @urchade in #50
- add dict format support in _parse_field_spec by @Yuvrajxms09 in #54
- Optimize inference latency in preprocessing pipeline by @maxwbuckley in #75
- From json by @urchade in #65
- Batch post-encoder ops for 22-57% inference speedup on GPU by @maxwbuckley in #76
- Update README.md by @urchade in #84
- Add fp16 quantization and torch.compile for up to 2.42x GPU speedup by @maxwbuckley in #86
- Add support of FlashDeBERTa by @Ingvarstep in #83
- Add max_len, for better memory usage control by @bogdanminko in #79
- Handle non-JSON API responses with typed client errors by @shuofengzhang in #85
- Fix torch.compile graph break in count_embed for 11-44% inference speedup by @maxwbuckley in #88
New Contributors
- @Yuvrajxms09 made their first contribution in #54
- @maxwbuckley made their first contribution in #75
- @Ingvarstep made their first contribution in #83
- @bogdanminko made their first contribution in #79
- @shuofengzhang made their first contribution in #85
Full Changelog: v1.2.3...v1.2.5