- Director of language and multimodal models, IBM Research1
Luis Lastras is IBM Research's director of language and multimodal models. Lastras is a co-author of the Granite Switch 4.1 launch post and of the Activated LoRA paper on adapters that can be switched in at inference time without reprocessing a conversation, an approach Granite Switch uses to combine a base Granite model with adapter functions.123
At a glance
- Current role
- Director of language and multimodal models
- Company
- IBM
- Sector
- Company
- Work listed
- 2
- Sources cited
- 5
- Profile reviewed
Generative computing at IBM Research
In April 2025 IBM Research described Lastras, then its director of language technologies, in connection with its work on activated LoRA (aLoRA) adapters, which Lastras co-authored. By June 2026 IBM Research described Lastras as its director of language and multimodal models, in a post introducing Granite Libraries and Project Granite Switch as part of IBM's approach to building LLMs with the modularity of software.431
Granite Switch adds a switching layer to a base Granite model so that adapter functions for tasks such as relevance scoring, query rewriting, hallucination detection and safety decisions can be activated at inference time. Lastras co-wrote the October 2026 post launching Granite Switch 4.1, which adds a decision adapter and speech recognition to the models.12
Information theory
Lastras is a co-author of a paper in the Proceedings of the National Academy of Sciences that extends Shannon's sender-receiver model of communication by giving the receiver a reasoning capability. IBM Research describes the work as a long-running research effort that Lastras pursued with a small team before it grew into a larger collaboration.5
Work and contributions
Granite Switch 4.1 8B Open weights
Co-author, with Kate Soule, Subhajit Chaudhury and Hendrik Strobelt, of the October 2026 Hugging Face post launching the Granite Switch 4.1 models.2
Activated LoRA: Fine-tuned LLMs for Intrinsics Published research
Co-author of the 2025 paper introducing activated LoRA adapters, which IBM describes as giving LLMs specialized capabilities at inference time without the delay of reprocessing the conversation.34
Timeline
- Co-authored the Activated LoRA paper on inference-friendly adapters for LLMs.34
- IBM Research introduced Granite Libraries and Project Granite Switch, quoting Lastras as director.1
- IBM Research reported a PNAS paper, co-authored by Lastras, extending information theory to account for reasoning.5
- Co-authored the Hugging Face post launching Granite Switch 4.1.2
Where they work
- IBM 29 people in the wiki
Related people
- Piero MolinoAlso at IBM
- Kate SouleAlso at IBM
- Marina DanilevskyAlso at IBM
- David D. CoxAlso at IBM
- Ranit AharonovAlso at IBM
- Thomas WolfML frameworks and libraries · Language models
- John GiannandreaLanguage models · ML frameworks and libraries
- Lysandre DebutML frameworks and libraries · Language models
- Yuxin WuML frameworks and libraries · Language models
- Dong YuML frameworks and libraries · Language models
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
5 sources · profile reviewed Oct 22, 2026 · data updated Oct 22, 2026
- How to build AI more like software (IBM Research blog)
- Granite Switch 4.1: Chat, Decisions, and Speech from One Endpoint
- Activated LoRA: Fine-tuned LLMs for Intrinsics (arXiv 2504.12397)
- A new LoRA technology for efficient agentic applications (IBM Research blog)
- What happens when information theory accounts for reasoning? (IBM Research blog)