hychiang

I also read the code ScanNet/SensReader/c++/src/main.cpp. ` //THIS IS A DEMO FUNCTION: HOW TO DECODE .SENS FILES: CHECK IT OUT! (doesn't do anything real though) void processFrame(const ml::SensorData& sd, size_t...

Same question

Shall we also consider the parameters loaded into memory and do mul-add with feature maps? The number should be equal to MemRead.

Hi, I found an [issue](https://github.com/switchablenorms/DeepFashion_Try_On/issues/50) in ACGPN in which they download the mask from [QD-IMD](https://github.com/karfly/qd-imd). Probably, PASTA-GAN uses the random masks from QD-IMD. Best

Hello, we have some initial results to share, but it is still under reviewing. Please see our pre-viewed version at https://hychiang.info/projects/quamba/

Similar issue. How should I workaround? RuntimeError: derivative for aten::hardswish_backward is not implemented

@jackkosaian Thanks! `cutlass::gemm::threadblock::GemmIdentityThreadblockSwizzle` works! It boost the performance of `UniversalGemmStreamK` by around `10x` (0.083712 ms -> 0.007076 ms). Here is the result: ``` GemmSplitKParallel: 0.010189 ms UniversalGemmStreamK: 0.007076 ms ```...

Thanks for pointing out! I will try that!

Hi, I tried newest stable version `0.4.9.1`. I got a slightly better score 66.4%, but still far from the reported 72.6%. I use `humaneval_instruct` and `--apply_chat_template`: ```bash # lm_eval version...

Hello, Here are the training script and log that I reproduced from the repository according to the paper and the author replied to the [issue](https://github.com/snap-research/EfficientFormer/issues/34#issuecomment-1355741852). I run the code with...