مقایسهٔ روش‌ها

	بهینه‌سازی مستقیم ترجیحات	QLoRA
حوزه	یادگیری عمیق	یادگیری عمیق
خانواده	Machine learning	Machine learning
سال پیدایش	2023	2023
پدیدآور	Rafael Rafailov	Tim Dettmers
نوع	Training methodology	Training methodology
منبع بنیادین	Rafailov, R., Sharma, A., Mitchell, E., Manning, C. D., Ermon, S., & Finn, C. (2023). Direct preference optimization: Your language model is secretly a reward model. arXiv preprint arXiv:2305.18290.	Dettmers, T., Pagnoni, A., Holtzman, A., & Contrastive, L. (2023). QLoRA: Efficient finetuning of quantized LLMs. arXiv preprint arXiv:2305.14314.
نام‌های دیگر	DPO, Direct preference	QLoRA, Quantized LoRA
مرتبط	4	4
خلاصه	Direct Preference Optimization (DPO) is a training method introduced by Rafailov et al. in 2023 that aligns language models with human preferences without requiring an explicit reward model. By directly optimizing for preference pairs (better response vs worse response), DPO simplifies the training pipeline compared to reinforcement learning from human feedback (RLHF).	QLoRA is an efficient fine-tuning method introduced by Dettmers et al. in 2023 that enables fine-tuning large language models using quantization and low-rank adaptation. By combining 4-bit quantization with LoRA, QLoRA reduces memory requirements by 75%, enabling fine-tuning of 65B-parameter models on single GPUs.