Edit model card



After some testing, finetuning and multiple merges of Llama-3 LLM models, here is something a little different.

This model is a MoE of 2x Llama-3 model trained on different RP format.

This repo contains FP16 files of Chatty-2x8B.

The idea

I started with two separate Llama-3-Instruct-8B models, each fine-tuned for specific RP formats.

Here is two simple exemple of how it was trained.

  • Expert 1: This model is trained to handle RP that requires actions and descriptions between asterisks. For example:
    *nods* Yes, I understand.
  • Expert 2: This model is fine-tuned for plain text RP where characters’ dialogues and actions are described straightforwardly. For example:
    Nods. "Yes, I understand."

My initial idea was to make a 11B or bigger Llama-3 model, or just make a 2x8B from existing model, but I got some issues, they were not stable enough, even after DPO and FFT on top my frankenmerge/moe of Llama-3, it was not working well enough to release them.

So I just tried the idea of having 2 different RP format trained on 2 separated Llama-3-Instruct-8B, and it worked pretty well!

The dataset

Based on Lumimaid 8B OAS success I still used the same "balance" between RP and non RP in the dataset, the maximum was 50% non RP data on each side.

RP data was different with some exception, the non RP data was exactly the same, despite that, I can't produce repetition so the double usage of non RP datasets didn't hurt the model in the end.

Prompt template: Llama3






Undi: If you want to support us, you can here.

IkariDev: Visit my retro/neocities style website please kek

Tasks Version Filter n-shot Metric Value Stderr
arc_challenge 1 none 0 acc 0.5469 ± 0.0145
none 0 acc_norm 0.5853 ± 0.0144
arc_easy 1 none 0 acc 0.8308 ± 0.0077
none 0 acc_norm 0.8258 ± 0.0078
gsm8k 3 strict-match 5 exact_match 0.7149 ± 0.0124
flexible-extract 5 exact_match 0.7096 ± 0.0125
hellaswag 1 none 0 acc 0.5945 ± 0.0049
none 0 acc_norm 0.7806 ± 0.0041
piqa 1 none 0 acc 0.7943 ± 0.0094
none 0 acc_norm 0.7998 ± 0.0093
truthfulqa_mc2 2 none 0 acc 0.5097 ± 0.0150
winogrande 1 none 0 acc 0.7356 ± 0.0124
Downloads last month
Model size
13.7B params
Tensor type