license: apache-2.0
InternLM2_5-20B-ArliAI-RPMax-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 models 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
InternLM2_5-20B-ArliAI-RPMax-v1.1 is a variant based on internlm2_5-20b-chat.
Unfortunately somehow InternLM uses so much VRAM I could only train it with 2048 context length.
Training Details
- Sequence Length: 2048
- Training Duration: Approximately 4 days on 2x3090Ti
- 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.
Suggested Prompt Format
ChatML Instruct Format
<|im_start|>system
Provide some context and/or instructions to the model.
<|im_end|>
<|im_start|>user
The user’s message goes here
<|im_end|>
<|im_start|>assistant