0.6.3 - Tools, also Mixtral, Qwen, Nous Hermes 2 Pro, CodeLlama
- In 0.6.3: bugfixes to keycloak (#231), llamaindex container, global tools demos (#221, #229)
- In 0.6.2: switch to 2bit quantizations for 70B+ models for better GPU memory usage
What's Changed in 0.6 series
Major changes:
- Mixtral, Qwen, Nous Hermes 2 Pro, CodeLlama by @lukemarsden in #211
- Feature/ollama runner by @rusenask in #207
--timeout
/ --timeout-seconds
arguments have been removed from runner
- RAG in early preview by @binocarlos in #206
- Keycloak v23 by @chocobar in #216
Minor changes:
- Fix/tool nits vol 2 by @rusenask in #198
- Feature/new session system prompt by @rusenask in #200
- Feature/api examples by @rusenask in #201
- remove empty messages, as seen in production by @lukemarsden in #202
- try clarifying toggle text by @lukemarsden in #203
- highlight inference mode when it's never been clicked by @lukemarsden in #204
- various design tweaks from feedback by @bigadamknight in #205
- remove fragment from account menu to stop console errors by @bigadamknight in #212
- closes #218 by @binocarlos in #219
- Update upgrade instructions to include restart of keycloak by @chocobar in #220
New Contributors
Full Changelog: 0.5.8...0.6.3