OASIS: Online Activation Subspace Learning for Memory-Efficient Training
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866910119449591808 |
|---|---|
| author | Choudhary, Sakshi Saxena, Utkarsh Roy, Kaushik |
| author_facet | Choudhary, Sakshi Saxena, Utkarsh Roy, Kaushik |
| contents | Training large language models (LLMs) is constrained by memory requirements, with activations accounting for a substantial fraction of the total footprint. Existing approaches reduce memory using low-rank weight parameterizations or low-rank gradient subspaces for optimizer states, while activation memory is addressed through architectural modifications or compression schemes based on periodically updated projections. We propose OASIS, an online activation subspace learning algorithm for memory-efficient training that tracks and continuously updates a low-dimensional activation subspace during training. Intermediate activations are projected onto this evolving subspace, reducing memory without modifying forward-pass computations. The evolving activation subspace induces low-rank gradient representations, enabling both gradients and optimizer states to be maintained directly in this subspace, while a projection-aware optimizer consistently transports optimizer states across subspace updates for stable training. Across various finetuning and pretraining tasks, OASIS achieves up to $2\times$ lower peak memory than full fine-tuning while matching its performance and outperforming prior low-rank methods. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_09406 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | OASIS: Online Activation Subspace Learning for Memory-Efficient Training Choudhary, Sakshi Saxena, Utkarsh Roy, Kaushik Machine Learning Training large language models (LLMs) is constrained by memory requirements, with activations accounting for a substantial fraction of the total footprint. Existing approaches reduce memory using low-rank weight parameterizations or low-rank gradient subspaces for optimizer states, while activation memory is addressed through architectural modifications or compression schemes based on periodically updated projections. We propose OASIS, an online activation subspace learning algorithm for memory-efficient training that tracks and continuously updates a low-dimensional activation subspace during training. Intermediate activations are projected onto this evolving subspace, reducing memory without modifying forward-pass computations. The evolving activation subspace induces low-rank gradient representations, enabling both gradients and optimizer states to be maintained directly in this subspace, while a projection-aware optimizer consistently transports optimizer states across subspace updates for stable training. Across various finetuning and pretraining tasks, OASIS achieves up to $2\times$ lower peak memory than full fine-tuning while matching its performance and outperforming prior low-rank methods. |
| title | OASIS: Online Activation Subspace Learning for Memory-Efficient Training |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2604.09406 |