- Senior Research Engineer, AllenNLP team, Ai2 (Allen Institute for AI)1278
Akshita Bhagia is a Senior Research Engineer on Ai2's AllenNLP team, where Bhagia builds and evaluates large language models and has recently worked on the FlexOlmo and Olmo projects. Bhagia is a co-author of the original OLMo paper and of the Dolma pretraining corpus paper. In OLMo 2, Bhagia is credited with base-model development, including training and data curation, and in OLMoE with supporting stability investigations and DCLM evaluations.123456
At a glance
- Current role
- Senior Research Engineer, AllenNLP team
- Company
- Ai2
- Sector
- Nonprofit
- Work listed
- 5
- Honors
- 3
- Sources cited
- 13
- Profile reviewed
Career
Akshita Bhagia worked at InFoCusp from 2015 to 2018, first as a Research Programmer and then as Lead Platform Development Engineer, building core infrastructure for a data science platform. During a Master's in Computer Science at the University of Massachusetts Amherst, completed in 2020, Bhagia worked with Mohit Iyyer at the intersection of NLP and digital humanities.81
Bhagia joined the Allen Institute for AI as a Research Engineer in July 2020 and became a Senior Research Engineer in July 2021, on the AllenNLP team. Bhagia's CV lists core contributions to the Olmo family of models in dataset creation, pretraining, evaluation and inference, along with work on FlexOlmo and on language models for cancer data with the Cancer AI Alliance.812
Research
Bhagia is a co-author of the OLMo and Dolma papers (2024) and a core contributor to OLMo 2, where the paper credits Bhagia with base-model development. Bhagia also worked on OLMoE, a fully open mixture-of-experts model, and is second-listed author of FlexOlmo (NeurIPS 2025 spotlight), which lets data owners train parts of a model without sharing their data.345698
Much of Bhagia's work concerns evaluating language models and their data. It includes What's In My Big Data? (ICLR 2024 spotlight), the Paloma benchmark for measuring language-model fit across hundreds of domains, the Catwalk evaluation framework, and a COLM 2025 paper on predicting task performance with compute-efficient model ladders.101112138
Teaching and service
Bhagia co-presented a NeurIPS 2024 tutorial on data preparation, model training and adaptation, and co-organized the 2022 WMT shared task on large-scale machine translation evaluation for African languages. Bhagia has reviewed for NeurIPS, COLING, LREC and ACL Rolling Review.8
Work and contributions
OLMo Open weights
Co-author of the OLMo paper (2024); credited in OLMo 2 with base-model development, and in OLMoE with stability investigations and DCLM evaluations.356
Dolma Open data
Co-author of the Dolma paper, which describes a three-trillion-token English corpus released for research on language-model pretraining.4
FlexOlmo Published research
Second-listed author of the 2025 FlexOlmo paper, a class of language models trained without sharing data, whose parameters and their data can be included or excluded at inference time; Bhagia's homepage lists FlexOlmo among Bhagia's most recent projects.91
Establishing Task Scaling Laws via Compute-Efficient Model Ladders Published research
Co-first author of the 2024 paper, published at COLM 2025, which predicts individual task performance of pretrained language models with task scaling laws and small model ladders.128
What's In My Big Data? Published research
Second-listed author of the WIMBD platform paper, a set of analyses of the content of large text corpora, presented as an ICLR 2024 spotlight.108
Timeline
- Joined InFoCusp as a Research Programmer.8
- Became Lead Platform Development Engineer at InFoCusp.8
- Joined the Allen Institute for AI as a Research Engineer.8
- Became a Senior Research Engineer at the Allen Institute for AI.8
- Released What's In My Big Data?, later an ICLR 2024 spotlight paper.108
- Released the Dolma paper on a three-trillion-token open pretraining corpus as a core contributor.413
- Released the OLMo paper, which received the ACL 2024 Best Theme Paper award.38
- Released the OLMo 2 paper as a core contributor credited with base-model development.58
- Released the FlexOlmo paper, later a NeurIPS 2025 spotlight.98
Other documented roles
Roles the sources document that are not confirmed as current.
- Research Engineer 2 (July 2020 – June 2021), Ai2 (Allen Institute for AI)8
- Research Programmer (2015–2018), then Lead Platform Development Engineer (2018), InFoCusp8
Honors and recognition
Where they work
- Ai2 23 people in the wiki
Related people
- Kyle LoAlso at Ai2
- Dirk GroeneveldAlso at Ai2
- Swabha SwayamdiptaAlso at Ai2
- Valentina PyatkinAlso at Ai2
- Pradeep DasigiAlso at Ai2
- Hyung Won ChungLanguage models · Training systems
- Collin BurnsLanguage models · Evaluation and benchmarks
- Ethan PerezEvaluation and benchmarks · Language models
- Samuel R. BowmanEvaluation and benchmarks · Language models
- Clément FarabetLanguage models · Training systems
Suggest a correction or removal
Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.
References
13 sources · profile reviewed Oct 21, 2026 · data updated Oct 21, 2026
- Akshita Bhagia — homepage
- Team — Ai2
- OLMo: Accelerating the Science of Language Models (arXiv 2402.00838)
- Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Research (arXiv 2402.00159)
- 2 OLMo 2 Furious (arXiv 2501.00656, full text with affiliations and Author Contributions)
- OLMoE: Open Mixture-of-Experts Language Models (arXiv 2409.02060, full text with Author Contributions)
- AkshitaB (Akshita Bhagia) on GitHub
- Akshita Bhagia — CV (PDF)
- FlexOlmo: Open Language Models for Flexible Data Use (arXiv 2507.07024)
- What's In My Big Data? (arXiv 2310.20707)
- Paloma: A Benchmark for Evaluating Language Model Fit (arXiv 2312.10523)
- Establishing Task Scaling Laws via Compute-Efficient Model Ladders (arXiv 2412.04403)
- Akshita Bhagia — publications