v0.15.0

What's new in 0.15.0 (2024-09-06)

These are the changes in inference v0.15.0.

New features

• FEAT: cosyvoice model support streaming reply by @wuminghui-coder in #2192
• FEAT: support qwen2-vl-instruct by @Minamiyama in #2205

Enhancements

• ENH: include openai-whisper into thirdparty by @qinxuye in #2232
• ENH: MiniCPM-V-2.6 Supports continuous batching with transformers engine by @ChengjieLi28 in #2238
• ENH: unpad for image2image/inpainting model by @wxiwnd in #2229
• ENH: Refine request log and add optional request_id by @frostyplanet in #2173
• REF: Use chat_template for LLM instead of prompt_style by @ChengjieLi28 in #2193

Bug fixes

• BUG: Fix docker image startup issue due to entrypoint by @ChengjieLi28 in #2207
• BUG: fix init xinference fail when custom path is fault by @amumu96 in #2208
• BUG: use default_uid to replace uid of actors which may override the xoscar actor's uid property by @qinxuye in #2214
• BUG: fix rerank max length by @qinxuye in #2219
• BUG: logger bug of function using generator decoration by @wxiwnd in #2215
• BUG: fix rerank calculation of tokens number by @qinxuye in #2228
• BUG: fix embedding token calculation & optimize memory by @qinxuye in #2221

Documentation

• DOC: Modify the installation documentation to change single quotes to double quotes for Windows compatibility. by @nikelius in #2211

Others

• Revert "EHN: clean cache for VL models (#2163)" by @qinxuye in #2230
• CHORE: Docker image is only pushed to aliyun when releasing version by @ChengjieLi28 in #2216
• CHORE: Compatible with openai >= 1.40 by @ChengjieLi28 in #2231

New Contributors

• @nikelius made their first contribution in #2211
• @wuminghui-coder made their first contribution in #2192

Full Changelog: v0.14.4...v0.15.0