Builders & open source
Torsten Scholak
Leads the Foundation Model Lab at ServiceNow Research; co-founder of the SLAM initiative, ServiceNow
- Leads the Foundation Model Lab at ServiceNow Research; co-founder of the SLAM initiative, ServiceNow16
Torsten Scholak leads the Foundation Model Lab at ServiceNow Research and co-founded the ServiceNow Language Model (SLAM) initiative, which develops the Apriel family of open-weight models for enterprise agentic tasks. Scholak was technical co-lead for pre-training and architecture on Apriel-1.5-15B-Thinker, is listed among the leads of Apriel-1.6-15B-Thinker, and led the Super Apriel project. Earlier work includes the PICARD constrained-decoding method for text-to-SQL parsing and the Hasktorch library.12345
At a glance
- Current role
- Leads the Foundation Model Lab at ServiceNow Research; co-founder of the SLAM initiative
- Company
- ServiceNow
- Sector
- Company
- Fields
- Language models, Training systems
- Education
- Toronto
- Work listed
- 3
- Sources cited
- 9
- Profile reviewed
Career
Scholak held a postdoctoral position at the University of Toronto before moving to industry at Unata and then Element AI. Element AI was acquired by ServiceNow in 2021.1
At ServiceNow Research, Scholak leads the Foundation Model Lab and co-founded the ServiceNow Language Model (SLAM) initiative. Scholak's current focus is the Apriel family of open-weight models built for enterprise agentic tasks and efficient inference.16
Research and open-source work
With Nathan Schucher and Dzmitry Bahdanau, Scholak wrote PICARD (2021), a method for constraining autoregressive decoding from language models by parsing incrementally, applied to text-to-SQL parsing. Scholak also co-organized the Deep Learning for Code workshop at ICLR 2022 and 2023 and helped build Hasktorch, a library for GPU-accelerated machine learning in Haskell.51
Scholak was technical co-lead for pre-training and architecture on Apriel-1.5-15B-Thinker (2025), co-authored the release post for Apriel-1.6-15B-Thinker (2025), and led the Super Apriel project (2026), whose paper lists Scholak as a core contributor. Scholak is also a co-author of the StarCoder 2 paper (2024).2347
Work and contributions
Apriel model family Open weights
Technical co-lead for pre-training and architecture on Apriel-1.5-15B-Thinker, and co-founder of the ServiceNow initiative that develops the Apriel models.21
Apriel-1.6-15B-Thinker Open weights
Co-author of the December 2025 release post and listed under the model's leads and management.3
SuperApriel-15B-Instruct Open weights
Led the Super Apriel project, a 15B-parameter supernet whose decoder layers each offer several trained mixer choices, contributing to its design, training and serving code and placement optimization.84
Timeline
- Posted PICARD, on constrained decoding for text-to-SQL parsing, as first author.5
- Co-authored the StarCoder 2 and The Stack v2 paper.7
- Served as technical co-lead for pre-training and architecture on Apriel-1.5-15B-Thinker.92
- Co-authored the release post for Apriel-1.6-15B-Thinker.3
- Led the Super Apriel project, posted as a paper with Scholak among three core contributors.84
Other documented roles
Roles the sources document that are not confirmed as current.
- Worked at Element AI, which ServiceNow acquired in 2021, Element AI1
Where they work
- ServiceNow 6 people in the wiki
Related people
- Sathwik Tejaswi MadhusudhanAlso at ServiceNow
- Raymond LiAlso at ServiceNow
- Oleksiy OstapenkoAlso at ServiceNow
- Shruthan RadhakrishnaAlso at ServiceNow
- Paul SmithAlso at ServiceNow
- Colin RaffelLanguage models · Training systems
- Jakub PachockiLanguage models · Training systems
- Beren MillidgeLanguage models · Training systems
- Liliang RenLanguage models · Training systems
- Furu WeiLanguage 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
9 sources · profile reviewed Oct 22, 2026 · data updated Oct 22, 2026
- Torsten Scholak — personal site
- Apriel-1.5-15B-Thinker: Mid-training is all you need — full text, Contributions and Acknowledgments
- Apriel-1.6-15b-Thinker: Cost-efficient Frontier Multimodal Performance (ServiceNow-AI on Hugging Face)
- Super Apriel: One Checkpoint, Many Speeds — full text (author list and core contributors)
- PICARD: Parsing Incrementally for Constrained Auto-Regressive Decoding from Language Models (arXiv 2109.05093)
- tscholak (Torsten Scholak) on GitHub
- StarCoder 2 and The Stack v2: The Next Generation (arXiv 2402.19173)
- Super Apriel: One Checkpoint, Many Speeds (arXiv 2604.19877)
- Apriel-1.5-15b-Thinker (arXiv 2510.01141)