# Reversible Foundations: Training a 120B Sparse MoE through State-Preserving Scaling

arXiv: https://arxiv.org/abs/2606.07404 · HTML: https://mlsanity.com/paper/2606.07404/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in MoE large language models, one of 12 directions in Mixture-of-experts.
- Builds on 36 of its 38 references that are in the map.
- First posted on arXiv 5 Jun 2026.

## Placed in

- Mixture-of-experts / [MoE large language models](https://mlsanity.com/moe/moe-large-language-models/)

## Builds on (most cited first, 30 of 36)

- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer](https://mlsanity.com/paper/1701.06538/) (2017)
- [Gemma 2: Improving Open Language Models at a Practical Size](https://mlsanity.com/paper/2408.00118/) (2024)
- [ZeRO: Memory Optimizations Toward Training Trillion Parameter Models](https://mlsanity.com/paper/1910.02054/) (2019)
- [Intrinsic Dimensionality Explains the Effectiveness of Language Model Fine-Tuning](https://mlsanity.com/paper/2012.13255/) (2020)
- [Linear Transformers Are Secretly Fast Weight Programmers](https://mlsanity.com/paper/2102.11174/) (2021)
- [Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention](https://mlsanity.com/paper/2502.11089/) (2025)
- [Parallelizing Linear Transformers with the Delta Rule over Sequence Length](https://mlsanity.com/paper/2406.06484/) (2024)
- [Better & Faster Large Language Models via Multi-token Prediction](https://mlsanity.com/paper/2404.19737/) (2024)
- [Query-Key Normalization for Transformers](https://mlsanity.com/paper/2010.04245/) (2020)
- [Sparse Upcycling: Training Mixture-of-Experts from Dense Checkpoints](https://mlsanity.com/paper/2212.05055/) (2022)
- [Auxiliary-Loss-Free Load Balancing Strategy for Mixture-of-Experts](https://mlsanity.com/paper/2408.15664/) (2024)
- [ReLoRA: High-Rank Training Through Low-Rank Updates](https://mlsanity.com/paper/2307.05695/) (2023)
- [SOLAR 10.7B: Scaling Large Language Models with Simple yet Effective Depth Up-Scaling](https://mlsanity.com/paper/2312.15166/) (2023)
- [TurboQuant: Online Vector Quantization with Near-optimal Distortion Rate](https://mlsanity.com/paper/2504.19874/) (2025)
- [LLaMA Pro: Progressive LLaMA with Block Expansion](https://mlsanity.com/paper/2401.02415/) (2024)
- [mHC: Manifold-Constrained Hyper-Connections](https://mlsanity.com/paper/2512.24880/) (2025)
- [Hyper-Connections](https://mlsanity.com/paper/2409.19606/) (2024)
- [Upcycling Large Language Models into Mixture of Experts](https://mlsanity.com/paper/2410.07524/) (2024)
- [Approaching Deep Learning through the Spectral Dynamics of Weights](https://mlsanity.com/paper/2408.11804/) (2024)
- [Weight decay induces low-rank attention layers](https://mlsanity.com/paper/2410.23819/) (2024)
- [QDyLoRA: Quantized Dynamic Low-Rank Adaptation for Efficient Large Language Model Tuning](https://mlsanity.com/paper/2402.10462/) (2024)
- [Extending the Context of Pretrained LLMs by Dropping Their Positional Embeddings](https://mlsanity.com/paper/2512.12167/) (2025)
- [Drop-Upcycling: Training Sparse Mixture of Experts with Partial Re-initialization](https://mlsanity.com/paper/2502.19261/) (2025)
- [OPUS: Towards Efficient and Principled Data Selection in Large Language Model Pre-training in Every Iteration](https://mlsanity.com/paper/2602.05400/) (2026)
- [SGD and Weight Decay Secretly Minimize the Rank of Your Neural Network](https://mlsanity.com/paper/2206.05794/) (2022)
- [PeriodicLoRA: Breaking the Low-Rank Bottleneck in LoRA Optimization](https://mlsanity.com/paper/2402.16141/) (2024)
- [Grove MoE: Towards Efficient and Superior MoE LLMs with Adjugate Experts](https://mlsanity.com/paper/2508.07785/) (2025)
- [GeLoRA: Geometric Adaptive Ranks For Efficient LoRA Fine-tuning](https://mlsanity.com/paper/2412.09250/) (2024)

## Abstract (arXiv)

This paper reports on training a hundred-billion-parameter sparse mixture of experts on a single eight-GPU node, end to end. LightningLM 0.1V is a recurrence-backbone language model family grown in four stages from a small dense seed, through a 5B and a 9B mixture of experts, to a 120B model with 460 routed experts under top-12 routing. Each larger model is grown from the trained weights of the smaller one; active parameters rise monotonically from 1.78B at the dense seed to 5.93B at 120B (about 5% of the 118.67B stored). The full lineage runs on single nodes, the larger stages at 8K context, reaching a released training loss of 1.78 at 120B scale. This is a systems and experience report. It is organized around three disciplines. Reversibility: a reversible recurrence stack reconstructs activations in the backward pass instead of storing them, holding activation memory flat as the model grows. State-preserving growth: each expansion (dense to MoE, shallow to deep, few experts to many) is given as a reproducible principle paired with the failure that results from getting it wrong; several failures are silent. Single-node economics: the 120B trains through TQP, a strategy of quantized base expert weights and trained low-rank adapters that carries optimizer state on 2.26B adapter parameters rather than 100B+ resident in routed experts, cutting expert-path optimizer state by a factor of ~45. What is new is the integration of known primitives, not any primitive in isolation: one grown lineage running end to end on a single node, documented at practitioner level, with per-domain held-out loss as evidence that targeted capabilities (multilingual Indic competence, code) were learned by construction. Model family, tokenizer, and training code are released.