ToastyPigeon/Mistral-Nemo-12B-Adventure-QLoRA

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>

Mistral-Nemo-12B-Adventure-QLoRA

Another QLoRA on Mistral Nemo Base, this time with Spring Dragon and Skein data included. ~29M tokens total of text adventure data.

This was trained in completion format where user input is given as > User input. Set > as a stopping string and preface your input with > to use with classic text completion mode.

This method of use is set up as default in Kobold Lite's Adventure mode.

Again, no instruct format was trained into this.

Apply to a Nemo model and use whatever instruct format that model uses - the style (and deadliness) of the LoRA carries over to instruct usage as well.

Training hyperparameters

The following hyperparameters were used during training:

• —learning_rate: 0.00025
• —trainbatchsize: 4
• —evalbatchsize: 1
• —seed: 42
• —optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• —lrschedulertype: cosinewithmin_lr
• —lrschedulerwarmup_steps: 10
• —num_epochs: 1

Framework versions

• —PEFT 0.12.0
• —Transformers 4.45.0.dev0
• —Pytorch 2.3.1+cu121
• —Datasets 2.21.0
• —Tokenizers 0.19.1