numind

/

NuExtract

like 161

Hi, yes the model is a fine-tuned version of phi-3-mini. We performed fine-tuning with 32 bit precision, but you can load the model in 16 bit mode as well:

model = AutoModelForCausalLM.from_pretrained("numind/NuExtract", torch_dtype=torch.bfloat16, trust_remote_code=True)

In fact, we recommend you do this as it leads to negligible performance loss :)

Thanks, makes sense now.