Results for “mmlu”

Launch distributed Megatron-LM training jobs on a SLURM cluster with a minimal sbatch skeleton, environment-variable setup for torch.distributed.run, CUDA_DEVICE_MAX_CONNECTIONS rules, container conventions, monitoring, and per-rank failure diagnosis.

Write, read, and validate HUML documents, converting between YAML/JSON/TOML and HUML for human-readable configuration files.