What's new in 0.7.3 (2023-12-22)
These are the changes in inference v0.7.3.
New features
- FEAT: Support OpenHermes 2.5 by @Bojun-Feng in #776
- FEAT: Support deepseek models by @aresnow1 in #786
- FEAT: Support tool message by @codingl2k1 in #794
- FEAT: Support Mixtral-8x7B-v0.1 models by @Bojun-Feng in #782
- FEAT: Support mistral instruct v0.2 by @aresnow1 in #796
Enhancements
- ENH: Enable streaming on Ctransformer by @Bojun-Feng in #784
- ENH: vllm backend support tool calls by @codingl2k1 in #785
- ENH: qwen switch to llama cpp by @codingl2k1 in #778
- ENH: [UI] register custom embedding model by @ChengjieLi28 in #791
Bug fixes
- BUG: UI Crash on Search when
model_format
andmodel_size
have been selected by @Bojun-Feng in #772 - BUG: When changing
XINFERENCE_HOME
env, the model files are still stored where they were. by @ChengjieLi28 in #777 - BUG: Remove the modelscope import by @aresnow1 in #788
- BUG: when terminating worker by
ctrl+C
, supervisor does not remove worker information by @ChengjieLi28 in #779 - BUG: Xinference does not release custom model name when registering failed by @ChengjieLi28 in #790
Documentation
- DOC: Update readme by @aresnow1 in #743
- DOC: Update FunctionCall.ipynb by @codingl2k1 in #773
Full Changelog: v0.7.2...v0.7.3