DecoderLens: Layerwise Interpretation of Encoder-Decoder Transformers
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2023
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866910396922724352 |
|---|---|
| author | Langedijk, Anna Mohebbi, Hosein Sarti, Gabriele Zuidema, Willem Jumelet, Jaap |
| author_facet | Langedijk, Anna Mohebbi, Hosein Sarti, Gabriele Zuidema, Willem Jumelet, Jaap |
| contents | In recent years, many interpretability methods have been proposed to help interpret the internal states of Transformer-models, at different levels of precision and complexity. Here, to analyze encoder-decoder Transformers, we propose a simple, new method: DecoderLens. Inspired by the LogitLens (for decoder-only Transformers), this method involves allowing the decoder to cross-attend representations of intermediate encoder layers instead of using the final encoder output, as is normally done in encoder-decoder models. The method thus maps previously uninterpretable vector representations to human-interpretable sequences of words or symbols. We report results from the DecoderLens applied to models trained on question answering, logical reasoning, speech recognition and machine translation. The DecoderLens reveals several specific subtasks that are solved at low or intermediate layers, shedding new light on the information flow inside the encoder component of this important class of models. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2310_03686 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | DecoderLens: Layerwise Interpretation of Encoder-Decoder Transformers Langedijk, Anna Mohebbi, Hosein Sarti, Gabriele Zuidema, Willem Jumelet, Jaap Computation and Language In recent years, many interpretability methods have been proposed to help interpret the internal states of Transformer-models, at different levels of precision and complexity. Here, to analyze encoder-decoder Transformers, we propose a simple, new method: DecoderLens. Inspired by the LogitLens (for decoder-only Transformers), this method involves allowing the decoder to cross-attend representations of intermediate encoder layers instead of using the final encoder output, as is normally done in encoder-decoder models. The method thus maps previously uninterpretable vector representations to human-interpretable sequences of words or symbols. We report results from the DecoderLens applied to models trained on question answering, logical reasoning, speech recognition and machine translation. The DecoderLens reveals several specific subtasks that are solved at low or intermediate layers, shedding new light on the information flow inside the encoder component of this important class of models. |
| title | DecoderLens: Layerwise Interpretation of Encoder-Decoder Transformers |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2310.03686 |