Xuye Qin
Xuye Qin
Yeah, the question about the difference between Mars and Dask should be inevitable. First of all, I do want to thank Dask project sincerely for the reason that it leads...
> > The comparison below is based on my knowledge of Dask, anyone can feel free to point out the misunderstood and error I have made, I will thank in...
> Looks like they all got actors and plasma. Some question as @dclong Actors means different for Ray and Mars. I hope we could give a detailed comparison between Mars...
Can we give a isolated example to produce this process? So that we can dive into to profile and optimize.
Seems vineyard cannot pass the ut for this PR, @sighingnow do you have to time to help to see what's going on?
Perhaps not all solvers can be implemented as a distributed algorithm, do you need all solvers or just the one can work?
`class_weight` and `l1_ratio` are not supported by now. The other parts look good to work, we will try to implement the absent algo and optimize the performance.
Thanks @igordertigor for your review, @vinta Could you please merge this PR?
This PR is ready for merge, @rlancemartin Could you please merge it?
This seems introduced by torch==2.3.0 which is specified by vllm v0.4.2. Workaround is downgrading vllm. pip install `vllm