I have a bert-like model that’s trained and exported as fp16 format (safetensors file).
When i deploy the model (following just standard huggingface pipeline), would it leverage the avx512-fp16 instruction sets? if possible, how can i verify that (not so intuitive from the dashboard itself).