goerch

@Igoorx : I took a look into your PR, but I believe we first have to sort out open problems with #2549. Sorry for the delay.

@lgoorx : #2549 is done, I'm now waiting for review of https://github.com/ggerganov/llama.cpp/pull/3252 and am already downloading Vicuna-7B-v1.1.

Currently discussing this at https://github.com/ggerganov/llama.cpp/issues/2820. Maybe close this one and participate over there?

> Not sure if @ggerganov agrees, but I think that the best way to do this may be a simple macro that has all the variables for the 3 tensors...

The changes in `ggml.c` look reasonable to me. Do you want me to touch other files of the code base, too?

> Thanks! Let's rebase on latest `master` so that the diff takes into account the new macros from #292 Should be done. > Btw, there has been recently an update...

https://github.com/dpgaspar/Flask-AppBuilder/pull/2162 shows mostly working tests. I'll try to improve tests some more and check the examples. In the meantime my most urgent question is if we migrate completely to newer...

I'm interested in #7, because it allows simple parallelization with something like ``` psum_kbn(f, X) = singleprec(Folds.mapreduce(f, InitialValues.asmonoid(plus_kbn), X)) psum_kbn(X) = psum_kbn(identity, X) ``` Is there anything wrong with the...

Maybe [relevant](https://stackoverflow.com/questions/25540793/cancellationtoken-with-async-dapper-methods)?

I think we use multi mapping in a similar way. Let me see ``` protected CommandDefinition CompileQuery(Query query, IDbTransaction transaction, CancellationToken token, CommandFlags flags = CommandFlags.Buffered) { var statement =...