Instruction Tuning for Large Language Models: A Survey
Paper
•
2308.10792
•
Published
•
1
LLM Post Training
Note Instruction Tuning 综述
-
Parameter-Efficient Fine-Tuning for Large Models: A Comprehensive Survey
Paper • 2403.14608 • Published -
Efficient Large Language Models: A Survey
Paper • 2312.03863 • Published • 3
ReFT: Reasoning with Reinforced Fine-Tuning
Paper • 2401.08967 • Published • 30Note Reinforced Fine-tuning- bytedance.2024
How Abilities in Large Language Models are Affected by Supervised Fine-tuning Data Composition
Paper • 2310.05492 • Published • 2Note 后训练SFT的数据组合-Qwen 1. 双阶段混合微调（Dual-stage Mixed Fine-tuning，DMT）,多任务学习中灾难性遗忘和性能冲突问题的策略 2. 实验结论: a. 混合比例的选择：DMT的第二阶段需要合理控制专门能力数据和一般能力数据的混合比例。实验中，论文通过调整不同的混合比例（如1/2, 1/4, 1/16等）来探索哪种比例能够最有效地保持模型的专门能力，同时不会对一般能力造成太大影响。 b. 数据量的影响：实验发现，在低资源环境下（即数据量较少时），将不同任务的数据混合在一起进行微调，能够有效地提高多项能力的表现。而在高资源条件下，单独微调效果更佳。因此，DMT策略特别适合解决低资源条件下的多任务微调问题。
-
Finetuned Language Models Are Zero-Shot Learners
Paper • 2109.01652 • Published • 2 -
Language Models are Few-Shot Learners
Paper • 2005.14165 • Published • 12
LLM Critics Help Catch LLM Bugs
Paper • 2407.00215 • PublishedNote OpenAI,使用 RLHF来训练CriticGPT 在代码方面评估模型的输出质量比人评估更好
Training language models to follow instructions with human feedback
Paper • 2203.02155 • Published • 16Note OpenAI: ChatGPT-RLHF
-
A Comprehensive Survey of LLM Alignment Techniques: RLHF, RLAIF, PPO, DPO and More
Paper • 2407.16216 • Published -
Inference Scaling for Long-Context Retrieval Augmented Generation
Paper • 2410.04343 • Published • 9
A Survey on Knowledge Distillation of Large Language Models
Paper • 2402.13116 • Published • 3Note https://github.com/Tebmer/Awesome-Knowledge-Distillation-of-LLMs
-
Distilling System 2 into System 1
Paper • 2407.06023 • Published • 3 -
Knowledge Distillation of Large Language Models
Paper • 2306.08543 • Published • 20