Hanshi Sun

@queensferryme Thanks for your time!

I find that codes below can work. - no line feed - use `\\` not `\` ``` $$\\left\\{\\begin{aligned}PDE \\quad &u_{tt}-a^2u_{xx}=0 \\\B.C.\\quad &u(0,t)=u(l,t)=0\\quad t\\ge 0 \\\I.C.\\quad & u(x,0)=\\varphi(x),u_t(x,0)=\\phi(x)\\quad 0\\le x\\le l...

Thank you for your time! It exactly works. @h-enk

This method seems extremely good and portable. I will use this one!😀 Thanks for your time and nice theme.

@LB-bulb @f-dangel Hello, I have the same problem when reproducing Fishr. I solved the problem by reinstalling `backpack-for-pytorch==1.3.0` (the older version).

> > @LB-bulb @f-dangel Hello, I have the same problem when reproducing Fishr. I solved the problem by reinstalling `backpack-for-pytorch==1.3.0` (the older version). > > @preminstrel Hi, I am also...

Thank you for your interest in our research! To clarify, our algorithm is designed for decoding acceleration, not pre-filling. For the pre-fill phase, we use a straightforward iterative method to...

Hello, thanks for your interest in our work! Here, I can give you an example. Basically, you can define `seq_len` using `tensor` instead of `int`. Then we can use the...

Hello, is that composable with kv cache offloading? But I cannot find its API... @tjruwase Thanks!

I mean, only offload kv cache while keeping whole model weights on GPUs. All the example code looks like for a single GPU.