Portrait of Eliezer Shlomi
INSIGHT Lab Ben-Gurion University of the Negev Technion – Israel Institute of Technology

Eliezer Shlomi

I make language models faster, both when they write and when they plan.

I'm a PhD student at the INSIGHT Lab, Ben-Gurion University of the Negev, advised by Prof. Chaim Baskin. I did my BSc and MSc at the Technion. My research is on fast inference for language models, most recently speculative decoding for diffusion language models, and on planning with language models through learned transitions in embedding space.

I'm always happy to talk about efficient inference, diffusion language models, planning and research collaborations.

Highlights

Research

Large language models are expensive to run: they generate step by step, and planners built on them call the model at every step. My work cuts the number of calls to the large model.

Inference

Fast decoding for diffusion language models

A tiny drafter predicts where the model commits next, and the model verifies several future steps in one pass, with its exact output distribution.

Planning

Planning in embedding space

A small network learns the transition function on top of frozen LLM embeddings, so a planner can search without calling an LLM at every step.

News

Publications

Google Scholar

On the SPOT: Enhancing Speculative Decoding for Diffusion Language Models via Commit-Order Prediction

Eliezer Shlomi*, Eliron Rahimi*, Avi Mendelson, Chaim Baskin

arXiv 2026Co-first author

SPOT predicts where a diffusion language model commits next, and the model verifies a whole tree of drafts in one pass: up to 6.8× fewer model passes and 3.2× the throughput of standard decoding, exact at any temperature.

BibTeX
@article{shlomi2026spot,
  title         = {On the {SPOT}: Enhancing Speculative Decoding for Diffusion Language Models via Commit-Order Prediction},
  author        = {Shlomi, Eliezer and Rahimi, Eliron and Mendelson, Avi and Baskin, Chaim},
  journal       = {arXiv preprint arXiv:2610.22543},
  year          = {2026},
  doi           = {10.48550/arXiv.2610.22543},
  eprint        = {2610.22543},
  archivePrefix = {arXiv},
  primaryClass  = {cs.LG},
  url           = {https://arxiv.org/abs/2610.22543}
}

Textual Planning with Explicit Latent Transitions

Eliezer Shlomi*, Ido Levy*, Eilam Shapira, Michael Katz, Guy Uziel, Segev Shlomov, Nir Mashkif, Roi Reichart, Sarah Keren

arXiv 2026Co-first author

EmbedPlan learns a transition model on top of frozen LLM text embeddings. It predicts the next state in about 0.17 ms instead of about 1.9 s for an LLM call, and ranks the true next state in the top 5 of 128 candidates 99.7% of the time on held-out transitions of seen problems.

BibTeX
@article{shlomi2026textual,
  title   = {Textual Planning with Explicit Latent Transitions},
  author  = {Shlomi, Eliezer and Levy, Ido and Shapira, Eilam and Katz, Michael and Uziel, Guy
             and Shlomov, Segev and Mashkif, Nir and Reichart, Roi and Keren, Sarah},
  journal = {arXiv preprint arXiv:2602.04557},
  year    = {2026}
}

Transition Function Prediction in AI Planning Using LLMs

Eliezer Shlomi, Guy Azran, Eilam Shapira, Omer Nahum, Guy Uziel, Michael Katz, Ateret Anaby Tavor, Roi Reichart, Sarah Keren

LM4Plan (AAAI-2025)

States and actions are embedded in a structured feature space and a shallow neural network serves as the transition function, so planning runs in the latent space without frequent LLM calls.

BibTeX
@inproceedings{shlomi2025transition,
  title     = {Transition Function Prediction in {AI} Planning Using {LLMs}},
  author    = {Shlomi, Eliezer and Azran, Guy and Shapira, Eilam and Nahum, Omer and Uziel, Guy
               and Katz, Michael and Anaby Tavor, Ateret and Reichart, Roi and Keren, Sarah},
  booktitle = {LM4Plan (AAAI-2025)},
  year      = {2025},
  url       = {https://openreview.net/forum?id=SPRYlWKaPj}
}

* Equal contribution.

Experience and education

Get in touch

I'm always happy to talk about efficient inference, diffusion language models, planning and research collaborations.

eliezershlomi0x@gmail.com