# CYCLE-INSTRUCT: Fully Seed-Free Instruction Tuning via Dual Self-Training and Cycle Consistency

arXiv: https://arxiv.org/abs/2508.16100 · HTML: https://mlsanity.com/paper/2508.16100/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- Placed in LLM preference optimization methods, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 21 of its 32 references that are in the map.
- First posted on arXiv 22 Aug 2025.

## Placed in

- LLM alignment / RLHF / preference optimisation / [LLM preference optimization methods](https://mlsanity.com/alignment/llm-preference-optimization-methods/)
- Synthetic data and self-improvement for LLMs / [Code and instruction data synthesis](https://mlsanity.com/synth/code-and-instruction-data-synthesis/)

## Builds on (most cited first, 21 of 21)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- [Finetuned Language Models Are Zero-Shot Learners](https://mlsanity.com/paper/2109.01652/) (2021)
- [Self-Instruct: Aligning Language Models with Self-Generated Instructions](https://mlsanity.com/paper/2212.10560/) (2022)
- [Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Them](https://mlsanity.com/paper/2210.09261/) (2022)
- [Multitask Prompted Training Enables Zero-Shot Task Generalization](https://mlsanity.com/paper/2110.08207/) (2021)
- [DROP: A Reading Comprehension Benchmark Requiring Discrete Reasoning Over Paragraphs](https://mlsanity.com/paper/1903.00161/) (2019)
- [Cross-Task Generalization via Natural Language Crowdsourcing Instructions](https://mlsanity.com/paper/2104.08773/) (2021)
- [Towards Building the Federated GPT: Federated Instruction Tuning](https://mlsanity.com/paper/2305.05644/) (2023)
- [Dual Learning for Machine Translation](https://mlsanity.com/paper/1611.00179/) (2016)
- [Phrase-Based & Neural Unsupervised Machine Translation](https://mlsanity.com/paper/1804.07755/) (2018)
- [Edinburgh Neural Machine Translation Systems for WMT 16](https://mlsanity.com/paper/1606.02891/) (2016)
- [CycleGAN, a Master of Steganography](https://mlsanity.com/paper/1712.02950/) (2017)
- [Dynosaur: A Dynamic Growth Paradigm for Instruction-Tuning Data Curation](https://mlsanity.com/paper/2305.14327/) (2023)
- [CRASS: A Novel Data Set and Benchmark to Test Counterfactual Reasoning of Large Language Models](https://mlsanity.com/paper/2112.11941/) (2021)
- [DoG-Instruct: Towards Premium Instruction-Tuning Data via Text-Grounded Instruction Wrapping](https://mlsanity.com/paper/2309.05447/) (2023)
- [CRAFT Your Dataset: Task-Specific Synthetic Dataset Generation Through Corpus Retrieval and Augmentation](https://mlsanity.com/paper/2409.02098/) (2024)
- [REInstruct: Building Instruction Data from Unlabeled Corpus](https://mlsanity.com/paper/2408.10663/) (2024)
- [LANDeRMT: Detecting and Routing Language-Aware Neurons for Selectively Finetuning LLMs to Machine Translation](https://mlsanity.com/paper/2409.19523/) (2024)

## Built on by (most cited first, 6 of 6)

- [Training-Trajectory-Aware Token Selection](https://mlsanity.com/paper/2601.10348/) (2026)
- [When the Strongest Teacher Is Not the Best Teacher: Student-Centric Answer Selection](https://mlsanity.com/paper/2605.26872/) (2026)
- [Self-Evolving Visual Questioner](https://mlsanity.com/paper/2606.13929/) (2026)
- [Improving Lean4 Autoformalization via Cycle Consistency Fine-tuning](https://mlsanity.com/paper/2603.24372/) (2026)
- [Consistency-Driven Co-Evolution for Self-Supervised Cross-Representation Learning](https://mlsanity.com/paper/2608.04926/) (2026)
- [Unsupervised Post-Training of Foundation Models: A Survey](https://mlsanity.com/paper/2608.24982/) (2026)

## Abstract (arXiv)

Instruction tuning is vital for aligning large language models (LLMs) with human intent, but current methods typically rely on costly human-annotated seed data or powerful external teacher models. While instruction back-translation techniques reduce this dependency, they remain fundamentally tethered to an initial seed set, which limits full automation, introduces biases, and can lead to inefficient use of unlabeled corpora. In this paper, we propose Cycle-Instruct, a novel framework that achieves fully seed-free instruction tuning. Inspired by cycle consistency, Cycle-Instruct employs a dual self-training loop where two models-an answer generator and a question generator-are bootstrapped solely from raw, unlabeled text. These models mutually supervise each other by reconstructing original text segments from their counterpart's generated pseudo-labels, effectively learning from the intrinsic structure of the data without any human-provided seeds. We demonstrate Cycle-Instruct's efficacy across four diverse data tracks, including general instruction-following, domain-specific tasks, dialogue logs, and plain text. Our extensive experiments show that Cycle-Instruct not only outperforms seed-driven back-translation baselines but also achieves performance comparable to strongly supervised methods.