A 4-bits double-quantized version of
ernestoBocini/Phi3-mini-DPO-Tuned.
This is a Phi-3-mini-4k-instruct fine-tuned with SFT and DPO on STEM domains, and finally quantized to a 4 bits precision, to serve the purpose of being an AI university tutor.
1BitsAndBytesConfig(
2 load_in_4bit=True,
3 bnb_4bit_quant_type="nf4",
4 bnb_4bit_use_double_quant=True,
5 bnb_4bit_compute_dtype=torch.bfloat16
6)
