A Post-Training Enhanced Optimization Approach for Small Language Models
This work addresses the challenge of enhancing small language models for more efficient deployment, though it appears incremental as it builds on existing post-training methods.
The paper tackles the problem of optimizing small language models through continuous post-training, proposing a method to construct alignment data using guidance from large models to enhance diversity and accuracy. It demonstrates that this approach significantly improves performance, as validated by experiments with the Qwen2-0.5B-Instruct model across various training setups.
This paper delves into the continuous post-training optimization methods for small language models, and proposes a continuous post-training alignment data construction method for small language models. The core of this method is based on the data guidance of large models, optimizing the diversity and accuracy of alignment data. In addition, to verify the effectiveness of the methods in this paper, we used Qwen2-0.5B-Instruct model as the baseline model for small language models, using the alignment dataset constructed by our proposed method, we trained and compared several groups of experiments, including SFT (Supervised Fine Tuning) post-training experiment and KTO (Kahneman Tversky optimization) post-training experiment, as well as SFT-KTO two-stage post-training experiment and model weight fusion experiment. Finally, we evaluated and analyzed the performance of post-training models, and confirmed that the continuous post-training optimization method proposed by us can significantly improve the performance of small language models.