[new feature] train finishing ETA! Good First Issue Good for newcomers
#242 opened Jan 26, 2022 by stas00
2
Contribute to bigscience-workshop/Megatron-DeepSpeed
Make your first contribution to this repository by tackling one of the issues listed below.
Each issue displayed here is a "good first issue," selected for its relative approachability for first-time contributors.
Good first issues
Make sure deepspeed powered models are equivalent with their non deepspeed version Good First Issue Good for newcomers
#226 opened Jan 6, 2022 by thomasw21
2
[deepspeed pipe] expand the partitioning method to support weights Good First Issue Good for newcomers
#186 opened Nov 9, 2021 by stas00
2
Parallelize Meg CUDA Kernel build system Good First Issue Good for newcomers Good Difficult Issue For complex tasks
#174 opened Oct 29, 2021 by stas00
[Tensorboard] Log text prediction in evaluation Good First Issue Good for newcomers
#163 opened Oct 27, 2021 by thomasw21
14
extract and log grad norm for individual layers Good First Issue Good for newcomers
#149 opened Oct 21, 2021 by stas00
10
clone HF's `GPT2` to create `GPTMeg` with a few tiny changes. Good First Issue Good for newcomers Good Second Issue For harder tasks
#138 opened Oct 16, 2021 by stas00
6
Plot scaling laws of our baseline models Good First Issue Good for newcomers arch&scale Architecture and Scaling Modeling Group
#125 opened Oct 5, 2021 by slippylolo
2
adding consistency calculations/checks at init time Good First Issue Good for newcomers Good Second Issue For harder tasks
#124 opened Oct 4, 2021 by stas00
11
Need model size dumped at init Good First Issue Good for newcomers
#123 opened Oct 4, 2021 by stas00
5