metadata

language:
  - en
license: cc-by-nc-4.0
tags:
  - text-generation-inference
  - transformers
  - unsloth
  - llama
  - trl
  - sft
  - Roleplay
  - roleplay
base_model: Sao10K/Fimbulvetr-11B-v2

About this model

TSF content

Beta for V2 of https://huggingface.co/Alsebay/Narumashi-11B-v0.9 (wrong typo but I'm too lazy to fix), but have only 32 rank and 32 lora rank, which cause the model didn't learn well all dataset information, it just know basis information. Anyways, it good if your have a chinese, japanese prompt to trigger TSF content. Maybe not smart, I haven't test yet.

Finetuned from model : Sao10K/Fimbulvetr-11B-v2 . Thank Sao10K a lot :)

I have text and found that Sao10K/Fimbulvetr-11B-v2 could unlock as 8K context length (maybe logic will go down a bit?), so I leave it alone to reduce RAM and VRAM. That mean you can use as 8k context length although this model say only 4k.

GGUF version? here is it.

Dataset

All chinese novels dataset

Dataset(all are novels):
60% skinsuit
25% possession
5% transform(shapeshift)
10% other

Thank Unsloth for good finetuning tool. This mistral model was trained 2x faster with Unsloth and Huggingface's TRL library.