Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[2310.09518] Instruction Tuning with Human Curriculum #536

Open
1 task
irthomasthomas opened this issue Feb 15, 2024 · 0 comments
Open
1 task

[2310.09518] Instruction Tuning with Human Curriculum #536

irthomasthomas opened this issue Feb 15, 2024 · 0 comments
Labels
Algorithms Sorting, Learning or Classifying. All algorithms go here. llm Large Language Models llm-benchmarks testing and benchmarking large language models llm-experiments experiments with large language models New-Label Choose this option if the existing labels are insufficient to describe the content accurately Papers Research papers Research personal research notes for a topic

Comments

@irthomasthomas
Copy link
Owner

[2310.09518] Instruction Tuning with Human Curriculum

DESCRIPTION: Computer Science > Computation and Language

[Submitted on 14 Oct 2023 (v1), last revised 13 Feb 2024 (this version, v2)]

Instruction Tuning with Human Curriculum

Bruce W. Lee, Hyunsoo Cho, Kang Min Yoo

In building instruction-tuned large language models (LLMs), the importance of a deep understanding of human knowledge can be often overlooked by the importance of instruction diversification. This research proposes a novel approach to instruction tuning by integrating a structured cognitive learning methodology that takes inspiration from the systematic progression and cognitively stimulating nature of human education through two key steps.

First, our synthetic instruction data generation pipeline, designed with some references to human educational frameworks, is enriched with meta-data detailing topics and cognitive rigor for each instruction. Specifically, our generation framework is infused with questions of varying levels of rigorousness, inspired by Bloom's Taxonomy, a classic educational model for structured curriculum learning.

Second, during instruction tuning, we curate instructions such that questions are presented in an increasingly complex manner utilizing the information on question complexity and cognitive rigorousness produced by our data generation pipeline. Our human-inspired curriculum learning yields significant performance enhancements compared to uniform sampling or round-robin, improving MMLU by 3.06 on LLaMA 2.

We conduct extensive experiments and find that the benefits of our approach are consistently observed in eight other benchmarks. We hope that our work will shed light on the post-training learning process of LLMs and its similarity with their human counterpart.

URL: https://arxiv.org/abs/2310.09518

Suggested labels

{'label-name': 'human-learning', 'label-description': 'Focuses on human-inspired curriculum learning and its application to instruction tuning for large language models.', 'confidence': 96.48}

@irthomasthomas irthomasthomas added llm Large Language Models llm-experiments experiments with large language models Research personal research notes for a topic Algorithms Sorting, Learning or Classifying. All algorithms go here. Papers Research papers llm-benchmarks testing and benchmarking large language models New-Label Choose this option if the existing labels are insufficient to describe the content accurately labels Feb 15, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Algorithms Sorting, Learning or Classifying. All algorithms go here. llm Large Language Models llm-benchmarks testing and benchmarking large language models llm-experiments experiments with large language models New-Label Choose this option if the existing labels are insufficient to describe the content accurately Papers Research papers Research personal research notes for a topic
Projects
None yet
Development

No branches or pull requests

1 participant