Salvato in:
Dettagli Bibliografici
Autori principali: Zausinger, Jonas, Pennig, Lars, Kozina, Anamarija, Sdahl, Sean, Sikora, Julian, Dendorfer, Adrian, Kuznetsov, Timofey, Hagog, Mohamad, Wiedemann, Nina, Chlodny, Kacper, Limbach, Vincent, Ketteler, Anna, Prein, Thorben, Singh, Vishwa Mohan, Danziger, Michael Morris, Born, Jannis
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:https://arxiv.org/abs/2411.02083
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866911108169728000
author Zausinger, Jonas
Pennig, Lars
Kozina, Anamarija
Sdahl, Sean
Sikora, Julian
Dendorfer, Adrian
Kuznetsov, Timofey
Hagog, Mohamad
Wiedemann, Nina
Chlodny, Kacper
Limbach, Vincent
Ketteler, Anna
Prein, Thorben
Singh, Vishwa Mohan
Danziger, Michael Morris
Born, Jannis
author_facet Zausinger, Jonas
Pennig, Lars
Kozina, Anamarija
Sdahl, Sean
Sikora, Julian
Dendorfer, Adrian
Kuznetsov, Timofey
Hagog, Mohamad
Wiedemann, Nina
Chlodny, Kacper
Limbach, Vincent
Ketteler, Anna
Prein, Thorben
Singh, Vishwa Mohan
Danziger, Michael Morris
Born, Jannis
contents While language models have exceptional capabilities at text generation, they lack a natural inductive bias for emitting numbers and thus struggle in tasks involving quantitative reasoning, especially arithmetic. One fundamental limitation is the nature of the cross-entropy (CE) loss, which assumes a nominal scale and thus cannot convey proximity between generated number tokens. In response, we here present a regression-like loss that operates purely on token level. Our proposed Number Token Loss (NTL) comes in two flavors and minimizes either the $L_p$ norm or the Wasserstein distance between the numerical values of the real and predicted number tokens. NTL can easily be added to any language model and extend the CE objective during training without runtime overhead. We evaluate the proposed scheme on various mathematical datasets and find that it consistently improves performance in math-related tasks. In a direct comparison on a regression task, we find that NTL can match the performance of a regression head, despite operating on token level. Finally, we scale NTL up to 3B parameter models and observe improved performance, demonstrating its potential for seamless integration into LLMs. We hope to inspire LLM developers to improve their pretraining objectives and distribute NTL as a minimalistic and lightweight PyPI package $ntloss$: https://github.com/ai4sd/number-token-loss. Development code for full paper reproduction is available separately.
format Preprint
id arxiv_https___arxiv_org_abs_2411_02083
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Regress, Don't Guess -- A Regression-like Loss on Number Tokens for Language Models
Zausinger, Jonas
Pennig, Lars
Kozina, Anamarija
Sdahl, Sean
Sikora, Julian
Dendorfer, Adrian
Kuznetsov, Timofey
Hagog, Mohamad
Wiedemann, Nina
Chlodny, Kacper
Limbach, Vincent
Ketteler, Anna
Prein, Thorben
Singh, Vishwa Mohan
Danziger, Michael Morris
Born, Jannis
Computation and Language
Artificial Intelligence
Computational Engineering, Finance, and Science
Machine Learning
While language models have exceptional capabilities at text generation, they lack a natural inductive bias for emitting numbers and thus struggle in tasks involving quantitative reasoning, especially arithmetic. One fundamental limitation is the nature of the cross-entropy (CE) loss, which assumes a nominal scale and thus cannot convey proximity between generated number tokens. In response, we here present a regression-like loss that operates purely on token level. Our proposed Number Token Loss (NTL) comes in two flavors and minimizes either the $L_p$ norm or the Wasserstein distance between the numerical values of the real and predicted number tokens. NTL can easily be added to any language model and extend the CE objective during training without runtime overhead. We evaluate the proposed scheme on various mathematical datasets and find that it consistently improves performance in math-related tasks. In a direct comparison on a regression task, we find that NTL can match the performance of a regression head, despite operating on token level. Finally, we scale NTL up to 3B parameter models and observe improved performance, demonstrating its potential for seamless integration into LLMs. We hope to inspire LLM developers to improve their pretraining objectives and distribute NTL as a minimalistic and lightweight PyPI package $ntloss$: https://github.com/ai4sd/number-token-loss. Development code for full paper reproduction is available separately.
title Regress, Don't Guess -- A Regression-like Loss on Number Tokens for Language Models
topic Computation and Language
Artificial Intelligence
Computational Engineering, Finance, and Science
Machine Learning
url https://arxiv.org/abs/2411.02083