update outputs path so that we can mount workspace to /workspace/data (#1623)

fix(examples): remove is_*_derived as it's parsed automatically (#1297)

set fp16 to false if bf16, update bf16: auto in example YAMLs (#1122) [skip ci]

Add shifted sparse attention (#973) [skip-ci]

new evals_per_epoch and saves_per_epoch to make things cleaner (#944)

Feat(wandb): Refactor to be more flexible (#767)

don't compile deepspeed or bitsandbytes from source (#837)

fix eval_steps to be a sane default (#797)

simplify by removing duplicate base_model_config (#772)

prepared dataset caching, other misc fixes (#665)

Fix Codellama examples (#582)

recommend padding when using sample packing (#531)

Feat(cfg): Add code-llama configs for all sizes (#479)