LargeScale
Set DATA_PATH, MULTITASK_DATA_PATH, CHECKPOINT_PATH in condigs/glm-130b/glm-130b.sh and HOST_FILE_PATH in scripts/submit_gpu.sh. Run the following scripts to reproduce GLM-130B's training.
bash scripts/submit_gpu.sh condigs/glm-130b/glm-130b.sh
At least 24 DGX-A100 (40G) is needed to lanuch training. A more detailed README will be released soon.