Llama-3-Instruct-8B-SimPO-ExPO-GGUF

Llama-3-Instruct-8B-SimPO-ExPO-GGUF

This is quantized version of chujiezheng/Llama-3-Instruct-8B-SimPO-ExPO created using llama.cpp

Model Description

The extrapolated (ExPO) model based on princeton-nlp/Llama-3-Instruct-8B-SimPO and meta-llama/Meta-Llama-3-8B-Instruct, as in the "Weak-to-Strong Extrapolation Expedites Alignment" paper.

Specifically, we obtain this model by extrapolating (alpha = 0.3) from the weights of the SFT and DPO/RLHF checkpoints, achieving superior alignment with human preference.

This extrapolated model achieves the 40.6% win rate and 45.8% LC win rate on AlpacaEval 2.0, outperforming the original Llama-3-Instruct-8B-SimPO's 40.5% and 44.7%, respectively.

Evaluation Results

Evaluation results on the AlpacaEval 2.0 benchmark (you can find the evaluation outputs on the official GitHub repo):

Evaluation results on the MT-Bench benchmark (you can find the evaluation outputs on the official GitHub repo):