ArliAI-RPMax-3.8B-v1.1

=====================================

RPMax Series Overview

| 2B | 3.8B | 8B | 9B | 12B | 20B | 22B | 70B |

RPMax is a series of models that are trained on a diverse set of curated creative writing and RP datasets with a focus on variety and deduplication. This model is designed to be highly creative and non-repetitive by making sure no two entries in the dataset have repeated characters or situations, which makes sure the model does not latch on to a certain personality and be capable of understanding and acting appropriately to any characters or situations.

Early tests by users mentioned that these models does not feel like any other RP models, having a different style and generally doesn't feel in-bred.

You can access the model at https://arliai.com and ask questions at https://www.reddit.com/r/ArliAI/

We also have a models ranking page at https://www.arliai.com/models-ranking

Ask questions in our new Discord Server! https://discord.com/invite/t75KbPgwhk

Model Description

Although this is finetuned on the same RPMax v1.1 dataset as usual, this model will not be as good as the 8B, 12B versions of RPMax due to the lower parameter and stricter censoring that Phi 3.5 has.

Or you can always download them and run it yourself if you have the hardware.

Training Details

Sequence Length: 16384
Training Duration: Approximately 1 day on RTX 4090
Epochs: 1 epoch training for minimized repetition sickness
QLORA: 64-rank 128-alpha, resulting in ~2% trainable weights
Learning Rate: 0.00001
Gradient accumulation: Very low 32 for better learning.

Quantization

The model is available in quantized formats:

FP16: https://huggingface.co/ArliAI/ArliAI-RPMax-Phi-3.8B-v1.1
GGUF: https://huggingface.co/ArliAI/ArliAI-RPMax-Phi-3.8B-v1.1-GGUF

Suggested Prompt Format

Phi 3.5 Instruct Prompt Format