ArliAI-RPMax-12B-v1.2

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

Saw ArliAI came out with a new version of the MN model and since I quanted it into EXL2 before figured I'd do it again. Though tbh I prefer Stardust v2, it's possible that this new update fixes the issues I had with it previously.

Who knows?

This is the EXL2 4bpw version of this model. Find the original model here.
Find the 8bpw version here.
Find the 6bpw version here.

UPDATE: For those getting gibberish results, it was merged wrongly to base after LORA training. Reuploaded all the files so it should work properly now.

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

ArliAI-RPMax-12B-v1.2 is a variant based on Mistral Nemo 12B Instruct 2407.

This is arguably the most successful RPMax model due to how Mistral is already very uncensored in the first place.

v1.2 update completely removes non-creative/RP examples in the dataset and is also an incremental improvement of the RPMax dataset which dedups the dataset even more and better filtering to cutout irrelevant description text that came from card sharing sites.

Specs

Context Length: 128K
Parameters: 12B

Training Details

Sequence Length: 8192
Training Duration: Approximately 2 days on 2x3090Ti
Epochs: 1 epoch training for minimized repetition sickness
LORA: 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:

Suggested Prompt Format

Mistral Instruct Prompt Format