swap batch size for gradient accumulation steps to decouple from num gpu

Update wandb_log_model on cerebras_1_3B_alpaca.yml

4bit quantized support (wip)

deepspeed doesn't work with flash-attn, and the gpu savings w flash attn are better than the deepspeed headaches

more logging, wandb fixes

improve prepared dataset loading, fix inference

config chooser, update readme instructions, device config, llama flash attention, debug out the labels, fix config key checks, other bugfixes