Abstract
LLaMA-Annotate is a tool that allows visually inspecting the confidences that a large language model assigns to individual tokens, and the alternative tokens considered for that position. We provide both a simple, non-interactive command-line interface, as well as a more elaborate web application. Besides generally helping to form an intuition about the “thinking” of the LLM, our tool can be used for context-aware spellchecking, or to see how a different prompt or a differently trained LLM can impact the interpretation of a piece of text. The tool can be tried online at https://huggingface.co/spaces/s-t-j/llama-annotate.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Similar content being viewed by others
Notes
- 1.
- 2.
For a demonstration, see https://drive.google.com/file/d/1HEBbuhJL_vrrP8-WpxmdhoUA8GVmcdWo/view?usp=sharing.
- 3.
Code is available at https://version.aalto.fi/gitlab/schulte1/llama-annotate.
- 4.
- 5.
- 6.
References
Brown, T., et al.: Language models are few-shot learners. In: Advances in Neural Information Processing Systems, vol. 33, pp. 1877–1901 (2020)
Lin, B.Y., et al.: The unlocking spell on base LLMs: rethinking alignment via in-context learning. ArXiv preprint (2023)
Meister, C., Pimentel, T., Wiher, G., Cotterell, R.: Locally typical sampling. Trans. Assoc. Comput. Linguist. 11, 102–121 (2023)
Sarti, G., Feldhus, N., Sickert, L., van der Wal, O., Nissim, M., Bisazza, A.: Inseq: an interpretability toolkit for sequence generation models. In: 61st Annual Meeting of the Association for Computational Linguistics, pp. 421–435 (2023)
Touvron, H., et al.: LLaMA 2: open foundation and fine-tuned chat models (2023). arXiv:2307.09288
Vaswani, A., et al.: Attention is all you need. In: Advances in Neural Information Processing Systems, vol. 30 (2017)
Vig, J.: A multiscale visualization of attention in the transformer model. In: 57th Annual Meeting of the Association for Computational Linguistics, pp. 37–42 (2019)
Yun, Z., Chen, Y., Olshausen, B.A., LeCun, Y.: Transformer visualization via dictionary learning: contextualized embedding as a linear superposition of transformer factors. In: Deep Learning Inside Out (DeeLIO) NAACL: The Second Workshop on Knowledge Extraction and Integration for Deep Learning Architectures (2021)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2024 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this paper
Cite this paper
Schultheis, E., John, S.T. (2024). LLaMA-Annotate—Visualizing Token-Level Confidences for LLMs. In: Bifet, A., et al. Machine Learning and Knowledge Discovery in Databases. Research Track and Demo Track. ECML PKDD 2024. Lecture Notes in Computer Science(), vol 14948. Springer, Cham. https://doi.org/10.1007/978-3-031-70371-3_33
Download citation
DOI: https://doi.org/10.1007/978-3-031-70371-3_33
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-70370-6
Online ISBN: 978-3-031-70371-3
eBook Packages: Computer ScienceComputer Science (R0)
