ladeak
ladeak
I have the same issue using the phi-2 and llama models through the integration of semantickernel. The values returned from the 'memory' seems to be completely independent to the search...
Why will the update of Bert models help? @AshD could you expand the issue should go away?
Ok, thank you for the suggestion. I have the impression that the community based alexa package has not really been maintainer anymore. Or is that not the case?
Thank you, I will remain on 4.22.9 as suggested, but not sure if I have a good update path forward. Thank you for the investigation, feel free to close the...
Also very interesting: when I compare the perf of the current implementation to something like on a machine that does not support Avx512, I get equivalent or slightly (0.3ns) better...
Is this an area that community members could take a part of and help with the lead of maintainers?
Let me have a look at one of these suggested files in the coming days.
@BrennanConroy , I find 2 `PathNormalizer` types (one in Kestrel, one in Shared). I assume you are referring to the one in Kestrel? (although the code is really similar).
I will continue with `PathNormalizer` in the coming days.
@BrennanConroy do you have other suggestions for converting unsafe code (other than `StringUtilities.TryGetAsciiString`)?