The Wisdom of Hindsight Makes Language Models Better Instruction Followers

Tianjun Zhang, Fangchen Liu, Justin Wong, Pieter Abbeel, Joseph E. Gonzalez
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:41414-41428, 2023.

Abstract

Reinforcement learning has seen wide success in finetuning large language models to better align with instructions via human feedback. The so-called algorithm, Reinforcement Learning with Human Feedback (RLHF) demonstrates impressive performance on the GPT series models. However, the underlying reinforcement learning algorithm is complex and requires additional training for reward and value networks. In this paper, we consider an alternative approach: converting feedback to instruction by relabeling the original one and training the model for better alignment in a supervised manner. Such an algorithm doesn’t require any additional parameters except for the original language model and maximally reuses the pretraining pipeline. To achieve this, we formulate instruction alignment problem for language models as a goal-reaching problem in decision making. We propose Hindsight Instruction Relabeling (HIR), a novel algorithm for aligning language models with instructions. The resulting two-stage algorithm shed light to a family of reward-free approaches that utilize the hindsightly relabeled instructions based on feedback. We evaluate the performance of HIR extensively on 12 challenging BigBench reasoning tasks and show that HIR outperforms the baseline algorithms and is comparable to or even surpasses supervised fine-tuning. The implementation of HIR is available at https://github.com/tianjunz/HIR.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-zhang23ab, title = {The Wisdom of Hindsight Makes Language Models Better Instruction Followers}, author = {Zhang, Tianjun and Liu, Fangchen and Wong, Justin and Abbeel, Pieter and Gonzalez, Joseph E.}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {41414--41428}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/zhang23ab/zhang23ab.pdf}, url = {https://proceedings.mlr.press/v202/zhang23ab.html}, abstract = {Reinforcement learning has seen wide success in finetuning large language models to better align with instructions via human feedback. The so-called algorithm, Reinforcement Learning with Human Feedback (RLHF) demonstrates impressive performance on the GPT series models. However, the underlying reinforcement learning algorithm is complex and requires additional training for reward and value networks. In this paper, we consider an alternative approach: converting feedback to instruction by relabeling the original one and training the model for better alignment in a supervised manner. Such an algorithm doesn’t require any additional parameters except for the original language model and maximally reuses the pretraining pipeline. To achieve this, we formulate instruction alignment problem for language models as a goal-reaching problem in decision making. We propose Hindsight Instruction Relabeling (HIR), a novel algorithm for aligning language models with instructions. The resulting two-stage algorithm shed light to a family of reward-free approaches that utilize the hindsightly relabeled instructions based on feedback. We evaluate the performance of HIR extensively on 12 challenging BigBench reasoning tasks and show that HIR outperforms the baseline algorithms and is comparable to or even surpasses supervised fine-tuning. The implementation of HIR is available at https://github.com/tianjunz/HIR.} }
Endnote
%0 Conference Paper %T The Wisdom of Hindsight Makes Language Models Better Instruction Followers %A Tianjun Zhang %A Fangchen Liu %A Justin Wong %A Pieter Abbeel %A Joseph E. Gonzalez %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-zhang23ab %I PMLR %P 41414--41428 %U https://proceedings.mlr.press/v202/zhang23ab.html %V 202 %X Reinforcement learning has seen wide success in finetuning large language models to better align with instructions via human feedback. The so-called algorithm, Reinforcement Learning with Human Feedback (RLHF) demonstrates impressive performance on the GPT series models. However, the underlying reinforcement learning algorithm is complex and requires additional training for reward and value networks. In this paper, we consider an alternative approach: converting feedback to instruction by relabeling the original one and training the model for better alignment in a supervised manner. Such an algorithm doesn’t require any additional parameters except for the original language model and maximally reuses the pretraining pipeline. To achieve this, we formulate instruction alignment problem for language models as a goal-reaching problem in decision making. We propose Hindsight Instruction Relabeling (HIR), a novel algorithm for aligning language models with instructions. The resulting two-stage algorithm shed light to a family of reward-free approaches that utilize the hindsightly relabeled instructions based on feedback. We evaluate the performance of HIR extensively on 12 challenging BigBench reasoning tasks and show that HIR outperforms the baseline algorithms and is comparable to or even surpasses supervised fine-tuning. The implementation of HIR is available at https://github.com/tianjunz/HIR.
APA
Zhang, T., Liu, F., Wong, J., Abbeel, P. & Gonzalez, J.E.. (2023). The Wisdom of Hindsight Makes Language Models Better Instruction Followers. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:41414-41428 Available from https://proceedings.mlr.press/v202/zhang23ab.html.

Related Material