Post-Training Probability Manifold Correction via Structured SVD Pruning and Self-Referential Distillation
Fuente:
arXiv
Salvato in:
| Autori principali: | , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866910006719283200 |
|---|---|
| author | Flouro, Aaron R. Chadwick, Shawn P. |
| author_facet | Flouro, Aaron R. Chadwick, Shawn P. |
| contents | Large language models are expensive to deploy. We introduce Sparse Knowledge Distillation (SparseKD), a post-training method that compresses transformer models by combining structured SVD pruning with self-referential knowledge distillation. The key insight is simple: instead of using an external teacher, the model teaches itself by matching its own probability distribution from before compression. This self-referential setup enables surprisingly strong quality recovery after aggressive pruning.
Our experiments reveal an unexpected finding: self-referential distillation alone, applied post-training under an identical objective and fixed calibration dataset, improves model quality by 39% relative to the original converged checkpoint. When combined with structured pruning, SparseKD achieves 15-65% parameter reduction with acceptable quality trade-offs. Kernel profiling shows that speedups arise entirely from reduced dense matrix multiplication in feed-forward layers while attention remains unchanged, making this approach complementary to attention optimizations.
We validate across two model families (0.6B and 3.8B parameters) with multi-seed experiments confirming high reproducibility. SparseKD requires no external super-teacher, no architectural changes, and no custom inference kernels, making it immediately deployable with existing infrastructure. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_00372 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Post-Training Probability Manifold Correction via Structured SVD Pruning and Self-Referential Distillation Flouro, Aaron R. Chadwick, Shawn P. Machine Learning Artificial Intelligence Computation and Language 68T05 I.2.6 Large language models are expensive to deploy. We introduce Sparse Knowledge Distillation (SparseKD), a post-training method that compresses transformer models by combining structured SVD pruning with self-referential knowledge distillation. The key insight is simple: instead of using an external teacher, the model teaches itself by matching its own probability distribution from before compression. This self-referential setup enables surprisingly strong quality recovery after aggressive pruning. Our experiments reveal an unexpected finding: self-referential distillation alone, applied post-training under an identical objective and fixed calibration dataset, improves model quality by 39% relative to the original converged checkpoint. When combined with structured pruning, SparseKD achieves 15-65% parameter reduction with acceptable quality trade-offs. Kernel profiling shows that speedups arise entirely from reduced dense matrix multiplication in feed-forward layers while attention remains unchanged, making this approach complementary to attention optimizations. We validate across two model families (0.6B and 3.8B parameters) with multi-seed experiments confirming high reproducibility. SparseKD requires no external super-teacher, no architectural changes, and no custom inference kernels, making it immediately deployable with existing infrastructure. |
| title | Post-Training Probability Manifold Correction via Structured SVD Pruning and Self-Referential Distillation |
| topic | Machine Learning Artificial Intelligence Computation and Language 68T05 I.2.6 |
| url | https://arxiv.org/abs/2602.00372 |