Abstract
This paper proposes a message-passing mechanism to address language modelling. A new layer type is introduced that aims to substitute self-attention for unidirectional sequence generation tasks. The system is shown to be competitive with existing methods: Given N tokens, the computational complexity is O(N logN) and the memory complexity is O(N) under reasonable assumptions. In the end, the Dispatcher layer is seen to achieve comparable perplexity to self-attention while being more efficient.- Anthology ID:
- 2022.clasp-1.3
- Volume:
- Proceedings of the 2022 CLASP Conference on (Dis)embodiment
- Month:
- September
- Year:
- 2022
- Address:
- Gothenburg, Sweden
- Editors:
- Simon Dobnik, Julian Grove, Asad Sayeed
- Venue:
- CLASP
- SIG:
- Publisher:
- Association for Computational Linguistics
- Note:
- Pages:
- 24–29
- Language:
- URL:
- https://aclanthology.org/2022.clasp-1.3
- DOI:
- Cite (ACL):
- Alberto Cetoli. 2022. Dispatcher: A Message-Passing Approach to Language Modelling. In Proceedings of the 2022 CLASP Conference on (Dis)embodiment, pages 24–29, Gothenburg, Sweden. Association for Computational Linguistics.
- Cite (Informal):
- Dispatcher: A Message-Passing Approach to Language Modelling (Cetoli, CLASP 2022)
- PDF:
- https://preview.aclanthology.org/nschneid-patch-2/2022.clasp-1.3.pdf
- Code
- fractalego/dispatcher
- Data
- WebText, WikiText-103, WikiText-2