Tuning Language Models by Mixture-of-Depths Ensemble
Fuente:
arXiv
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916442593558528 |
|---|---|
| author | Luo, Haoyan Specia, Lucia |
| author_facet | Luo, Haoyan Specia, Lucia |
| contents | Transformer-based Large Language Models (LLMs) traditionally rely on final-layer loss for training and final-layer representations for predictions, potentially overlooking the predictive power embedded in intermediate layers. Surprisingly, we find that focusing training efforts on these intermediate layers can yield training losses comparable to those of final layers, with complementary test-time performance. We introduce a novel tuning framework, Mixture-of-Depths (MoD), which trains late layers as ensembles contributing to the final logits through learned routing weights. With the auxiliary distillation loss and additional normalization modules, we ensure that the outputs of the late layers adapt to language modeling. Our MoD framework, which can be integrated with any existing tuning method, shows consistent improvement on various language modelling tasks. Furthermore, by replacing traditional trainable modules with MoD, our approach achieves similar performance with significantly fewer trainable parameters, demonstrating the potential of leveraging predictive power from intermediate representations during training. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2410_13077 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Tuning Language Models by Mixture-of-Depths Ensemble Luo, Haoyan Specia, Lucia Computation and Language Artificial Intelligence Transformer-based Large Language Models (LLMs) traditionally rely on final-layer loss for training and final-layer representations for predictions, potentially overlooking the predictive power embedded in intermediate layers. Surprisingly, we find that focusing training efforts on these intermediate layers can yield training losses comparable to those of final layers, with complementary test-time performance. We introduce a novel tuning framework, Mixture-of-Depths (MoD), which trains late layers as ensembles contributing to the final logits through learned routing weights. With the auxiliary distillation loss and additional normalization modules, we ensure that the outputs of the late layers adapt to language modeling. Our MoD framework, which can be integrated with any existing tuning method, shows consistent improvement on various language modelling tasks. Furthermore, by replacing traditional trainable modules with MoD, our approach achieves similar performance with significantly fewer trainable parameters, demonstrating the potential of leveraging predictive power from intermediate representations during training. |
| title | Tuning Language Models by Mixture-of-Depths Ensemble |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2410.13077 |