Magpie-Align/Llama-3-8B-Magpie-Align-v0.3

Magpie-Align/Llama-3-8B-Magpie-Align-v0.3 is an 8 billion parameter language model developed by Magpie-Align, fine-tuned from Meta-Llama-3-8B with an 8192 token context length. This model is aligned using a two-stage pipeline involving Supervised Fine-tuning (SFT) on custom datasets and Direct Preference Optimization (DPO) on the princeton-nlp/llama3-ultrafeedback-armorm dataset. It demonstrates improved performance over the official Llama-3-8B-Instruct model, particularly excelling in alignment benchmarks like Alpaca Eval 2 and Arena Hard, and shows enhanced capability in handling Chinese queries.