# All is Not Lost: LLM Recovery without Checkpoints

arXiv: https://arxiv.org/abs/2506.15461 · HTML: https://mlsanity.com/paper/2506.15461/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 23 of its 34 references that are in the map.
- First posted on arXiv 18 Jun 2025.

## Builds on (most cited first, 23 of 23)

- [OPT: Open Pre-trained Transformer Language Models](https://mlsanity.com/paper/2205.01068/) (2022)
- [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://mlsanity.com/paper/1909.11942/) (2019)
- [BLOOM: A 176B-Parameter Open-Access Multilingual Language Model](https://mlsanity.com/paper/2211.05100/) (2022)
- [Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time](https://mlsanity.com/paper/2203.05482/) (2022)
- [Linear Mode Connectivity and the Lottery Ticket Hypothesis](https://mlsanity.com/paper/1912.05671/) (2019)
- [TinyStories: How Small Can Language Models Be and Still Speak Coherent English?](https://mlsanity.com/paper/2305.07759/) (2023)
- [LayerSkip: Enabling Early Exit Inference and Self-Speculative Decoding](https://mlsanity.com/paper/2404.16710/) (2024)
- [Understanding Robustness of Transformers for Image Classification](https://mlsanity.com/paper/2103.14586/) (2021)
- [RedPajama: an Open Dataset for Training Large Language Models](https://mlsanity.com/paper/2411.12372/) (2024)
- [Decentralized Training of Foundation Models in Heterogeneous Environments](https://mlsanity.com/paper/2206.01288/) (2022)
- [SWARM Parallelism: Training Large Models Can Be Surprisingly Communication-Efficient](https://mlsanity.com/paper/2301.11913/) (2023)
- [Bamboo: Making Preemptible Instances Resilient for Affordable Training of Large DNNs](https://mlsanity.com/paper/2204.12013/) (2022)
- [Towards Crowdsourced Training of Large Neural Networks using Decentralized Mixture-of-Experts](https://mlsanity.com/paper/2002.04013/) (2020)
- [Stacking Your Transformers: A Closer Look at Model Growth for Efficient LLM Pre-Training](https://mlsanity.com/paper/2405.15319/) (2024)
- [Oobleck: Resilient Distributed Training of Large Models Using Pipeline Templates](https://mlsanity.com/paper/2309.08125/) (2023)
- [Effective Model Sparsification by Scheduled Grow-and-Prune Methods](https://mlsanity.com/paper/2106.09857/) (2021)
- [Every Parameter Matters: Ensuring the Convergence of Federated Learning with Dynamic Heterogeneous Models Reduction](https://mlsanity.com/paper/2310.08670/) (2023)
- [On the Inductive Bias of Stacking Towards Improving Reasoning](https://mlsanity.com/paper/2409.19044/) (2024)
- [ATTNChecker: Highly-Optimized Fault Tolerant Attention for Large Language Model Training](https://mlsanity.com/paper/2410.11720/) (2024)
- [Subspace Networks: Scaling Decentralized Training with Communication-Efficient Model Parallelism](https://mlsanity.com/paper/2506.01260/) (2025)
- [Super Tiny Language Models](https://mlsanity.com/paper/2405.14159/) (2024)
- [Stacking as Accelerated Gradient Descent](https://mlsanity.com/paper/2403.04978/) (2024)
- [SkipPipe: Partial and Reordered Pipelining Framework for Training LLMs in Heterogeneous Networks](https://mlsanity.com/paper/2502.19913/) (2025)

## Built on by (most cited first, 1 of 1)

- [TierCheck: Tiered Checkpointing for Fault Tolerance in Large Language Model Training](https://mlsanity.com/paper/2605.17821/) (2026)

## Abstract (arXiv)

Training LLMs on decentralized nodes or on-spot instances, lowers the training cost and enables model democratization. The inevitable challenge here is the transient churns of nodes due to failures and the operator's scheduling policies, leading to losing parts of the model (some layers). The conventional approaches to recover from failures is to either use checkpointing, where periodically a copy of the entire model is sent to an additional storage, or redundant computation. These approaches yield significant communication and/or computation overhead even in non-failure cases and scale poorly in settings with large models. In this paper we propose CheckFree, an efficient recovery method where a failing stage is substituted by weighted averaging of the closest neighboring stages. In contrast to the state of the art, CheckFree requires no additional computation or storage. However, because of the nature of averaging neighbouring stages, it can only recover failures of intermediate stages. We further extend our method to CheckFree+ with out-of-order pipeline execution to tolerate crashes of the first and last stages. Thanks to out-of-order pipelining, behaviour of the first and last stages are mimicked by their neighboring ones, which allows CheckFree+ to recover them by copying the neighboring stages. To recover the (de-)embedding layers, CheckFree+ copies those layers in the neighboring stages, which requires relatively small storage overhead. We extensively evaluate our method on LLaMa models of model sizes from 124M to 1.5B with varying failure frequencies. In the case of low and medium failure rates (5-10%), CheckFree and CheckFree+ outperform both checkpointing and redundant computation in terms of convergence wall-clock time, achieving up to 12% improvement over redundant computation. Both of our proposals can be ran via our code available at: https://github.com/gensyn-ai/CheckFree