OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866912470858203136 |
|---|---|
| author | Liu, Jiacheng Blanton, Taylor Elazar, Yanai Min, Sewon Chen, YenSung Chheda-Kothary, Arnavi Tran, Huy Bischoff, Byron Marsh, Eric Schmitz, Michael Trier, Cassidy Sarnat, Aaron James, Jenna Borchardt, Jon Kuehl, Bailey Cheng, Evie Farley, Karen Sreeram, Sruthi Anderson, Taira Albright, David Schoenick, Carissa Soldaini, Luca Groeneveld, Dirk Pang, Rock Yuren Koh, Pang Wei Smith, Noah A. Lebrecht, Sophie Choi, Yejin Hajishirzi, Hannaneh Farhadi, Ali Dodge, Jesse |
| author_facet | Liu, Jiacheng Blanton, Taylor Elazar, Yanai Min, Sewon Chen, YenSung Chheda-Kothary, Arnavi Tran, Huy Bischoff, Byron Marsh, Eric Schmitz, Michael Trier, Cassidy Sarnat, Aaron James, Jenna Borchardt, Jon Kuehl, Bailey Cheng, Evie Farley, Karen Sreeram, Sruthi Anderson, Taira Albright, David Schoenick, Carissa Soldaini, Luca Groeneveld, Dirk Pang, Rock Yuren Koh, Pang Wei Smith, Noah A. Lebrecht, Sophie Choi, Yejin Hajishirzi, Hannaneh Farhadi, Ali Dodge, Jesse |
| contents | We present OLMoTrace, the first system that traces the outputs of language models back to their full, multi-trillion-token training data in real time. OLMoTrace finds and shows verbatim matches between segments of language model output and documents in the training text corpora. Powered by an extended version of infini-gram (Liu et al., 2024), our system returns tracing results within a few seconds. OLMoTrace can help users understand the behavior of language models through the lens of their training data. We showcase how it can be used to explore fact checking, hallucination, and the creativity of language models. OLMoTrace is publicly available and fully open-source. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_07096 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens Liu, Jiacheng Blanton, Taylor Elazar, Yanai Min, Sewon Chen, YenSung Chheda-Kothary, Arnavi Tran, Huy Bischoff, Byron Marsh, Eric Schmitz, Michael Trier, Cassidy Sarnat, Aaron James, Jenna Borchardt, Jon Kuehl, Bailey Cheng, Evie Farley, Karen Sreeram, Sruthi Anderson, Taira Albright, David Schoenick, Carissa Soldaini, Luca Groeneveld, Dirk Pang, Rock Yuren Koh, Pang Wei Smith, Noah A. Lebrecht, Sophie Choi, Yejin Hajishirzi, Hannaneh Farhadi, Ali Dodge, Jesse Computation and Language We present OLMoTrace, the first system that traces the outputs of language models back to their full, multi-trillion-token training data in real time. OLMoTrace finds and shows verbatim matches between segments of language model output and documents in the training text corpora. Powered by an extended version of infini-gram (Liu et al., 2024), our system returns tracing results within a few seconds. OLMoTrace can help users understand the behavior of language models through the lens of their training data. We showcase how it can be used to explore fact checking, hallucination, and the creativity of language models. OLMoTrace is publicly available and fully open-source. |
| title | OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2504.07096 |