Repository Issue Activity (beta)

#184 Energy Efficiency: 10 Mathematical Techniques for 60-70% AI Energy Reduction (Phi6Simple, FFT-Mix, Phi MoE)

#83 ParallelDroplessMLP initialises self.mlp twice

#180 Compatibility with CUDA 13.0

#179 difference between cuda kernel code and using torch.ops

#178 Issue with Sparse MLP

#170 Standard dmoe script crushes

#94 Unsharding scripts for megablocks models

#155 what devices are supported?

#159 Pytorch 2.5.0 + Megablocks undefined symbol error

#154 Memory cost increase gradually until oom

#166 MOE uses more memory than dense model and is slower

#95 AMP + BF16 failing

#165 bump version for PyTorch 2.6

#134 Running into ValueError when running moe/dmoe scripts

#157 amp_C undefined symbol after installing Megablocks

#164 [rank5]: TypeError: SortOp.forward() takes from 2 to 3 positional arguments but 5 were given When running moe script

#163 Grouped GEMM execution not possible with HW

#161 CUDA Error When Running Single GPU Experiment

#160 Is current megablocks compatible with distributed optimizer in Megatron-LM?

#156 Do you support the bias of mlp？

#153 Does it work with torch.compile?

#107 1-expert worse than dense model

#132 [Do not Merge] Add CI/CD Milestones

#130 Torch ModuleNotFoundError when running pip install megablocks

#61 Question on offsets in figures 5