lemonade-sdk/lemonade v8.1.2

on GitHub

Headline

• Many fixes and improvements, on top of the major improvements from v8.1.1:
  • ROCm7 is now available as a llamacpp backend on supported Radeon GPUs @danielholanda
  • Model build for custom NPU and Hybrid LLMs updated for RAI SW 1.5 @iswaryaalex
  • Added gpt-oss-120b-GGUF and gpt-oss-20b-GGUF support to Lemonade Server @danielholanda

What's Changed

• Add host paramater to allow external connections (Solves #85) by @ajh123 in #142
• Add models from the Deepseek R1, Llama3.2, Mistral, Phi3.5, ChatGLM3, and AMD OLMo family to NPU by @Looong01, @henrylearn2rock in #94
• Add lsdev entrypoint by @jeremyfowers in #116

Bug Fixes

• Fix backwards compatibility bug in model_manager.py by @jeremyfowers in #109
• Fix inference engine import by @jeremyfowers in #120
• Enable wmi on python embeddable distribution by @danielholanda in #140
• Fix lemonade-server run on Linux ('OutputDuplicator' is not defined) by @danielholanda in #148
• Move OGA API examples to pre-built OGA model, and remove hf token from test by @jeremyfowers in #156
• Improve HIP ID detection mechanism by @danielholanda in #151

Documentation

• FAQ - New Questions by @vgodsoe in #138
• Some more faq.md edits by @jeremyfowers in #143

New Contributors

• @ajh123 made their first contribution in #142

Full Changelog: v8.1.1...v8.1.2