Hey community,
I’ve been training a Llama model using run_clm.py and came across something I’m unsure about. After training, in the checkpoints folder, I found this zero_to_fp32.py script. My question is, do I actually need to run this on my completed model, or can I just go ahead with the model.safetensors file which seems to hold the bf16 weights?
Appreciate any insights you guys might have!