This is a big release, We now have working support for RAG inside of RamaLama.
Try out
ramalama rag XYZ.pdf ABC.doc quay.io/NAME/myrag
ramalama run --rag quay.io/NAME/myrag MYMODEL
What's Changed
- Default whisper-server.sh, llama-server.sh to /mnt/models/model.file by @rhatdan in #984
- Improve intel-gpu to work with whisper-server and llama-server by @rhatdan in #986
- whisper.cpp requires ffmpeg by @ericcurtin in #985
- Fix container_build.sh to build all images by @rhatdan in #989
- fix: use expected condition by @benoitf in #992
- [CANN]Fix the bug that openEuler repo does not have ffmpeg-free package, instand of using ffmpeg for openEuler by @leo-pony in #994
- Add docling support version 2 by @rhatdan in #979
- chore: use the reverse condition for models by @benoitf in #995
- FIX: Ollama install with brew for CI by @kush-gupt in #1002
- Add the ability to identify a wider set of Intel GPUs that have enough Execution Units to produce decent results by @cgruver in #996
- Add ramalama client by @ericcurtin in #997
- Fix errors found in RamaLama RAG by @rhatdan in #998
- Turn on verbose logging in llama-server if --debug is on by @ericcurtin in #1001
- Don't use relative paths for destination by @rhatdan in #1003
- Red Hat Konflux update ramalama by @red-hat-konflux in #1005
- Fix errors on python3.9 by @rhatdan in #1007
- Use this container if we detect ROCm accelerator by @ericcurtin in #1008
- Improve UX for ramalama-client by @ericcurtin in #1013
- update docs for Intel GPU support. Clean up code comments by @cgruver in #1011
- Generate quadlets with rag databases by @rhatdan in #1012
- Keep conversation history by @ericcurtin in #1014
- Fix ramalama serve --rag ABC --generate kube by @rhatdan in #1015
- Adds Rag chatbot to ramalama serve and preloads models for doc2rag and rag_framework by @bmahabirbu in #1010
- Rag condition should be and instead of or by @ericcurtin in #1016
- Show model name in API instead of model file path by @bachp in #1009
- Make install script more aesthetically pleasing by @ericcurtin in #1019
- Color each word individually by @ericcurtin in #1017
- Add feature to turn off colored text by @ericcurtin in #1021
- Fix up building of images by @rhatdan in #1023
- Change default ROCM image to rocm-fedora by @rhatdan in #1024
- Run build_rag.sh as root by @rhatdan in #1027
- added hacky method to use 'run' instead of 'serve' for rag by @bmahabirbu in #1026
- More fixes to build scripts by @rhatdan in #1028
- Updated rag to have much better querys at the cost of slight delay by @bmahabirbu in #1029
- More fixes to build scripts by @rhatdan in #1031
- Minor bugfix remove self. from self.prompt by @ericcurtin in #1032
- Added terminal name fixed eof bug and added another model to rag_framework load by @bmahabirbu in #1033
- Update registry.access.redhat.com/ubi9/ubi Docker tag to v9.5-1742918310 by @renovate in #1035
- Typo in the webui by @ericcurtin in #1039
- Fix errors on python3.9 by @marceloleitner in #1038
- More updates for builds by @rhatdan in #1036
New Contributors
- @red-hat-konflux made their first contribution in #1005
- @bachp made their first contribution in #1009
- @marceloleitner made their first contribution in #1038
Full Changelog: v0.6.4...v0.7.0