LLM To Create Training Data For Domain Generalization

Three key points

✔️ A synthetic data generation method using LLM is proposed to improve the domain generalization performance of the NLI task.
✔️ The synthetic data generated by the proposed method is effective when used for training, especially when the model size is small (where fast inference is important).
✔️ We also show that training synthetic data on sufficiently large models improves within-distribution performance.

A Synthetic Data Approach for Domain Generalization of NLI Models
written by Mohammad Javad Hosseini , Andrey Petrov , AlexFabrikant , Annie Louis
(Submitted on 19 Feb 2024 (v1), last revised 28 Jun 2024 (this version, v2))
Subjects : Computation and Language (cs.CL)

CODE:

 

• 
• 
• 

Categories related to this article

• Natural Language Processing
• Dataset Synthesis With LLM
• Dataset
• Large Language Models
• Datasets