jspr's picture
Create README.md
3700967 verified

Mistral-7B finetuned on a dataset of BTS fanfic at 32k context.

This model uses the alpaca format:

{"instruction": "An interaction between a user providing instructions, and an imaginative assistant providing responses.", "input": "...", "output": "..."}

Note RoPE scaling parameter 8.0, with RoPE scaling type linear