Victor Morand

PhD Student @ Sorbonne Université (@mlia_isir)
Towards Language models that know what they know 🧠

[Victor Morand @_VictorMorand_]

🐭 Excited to share MICE (Minimal Interaction Cross-Encoders) - accepted to #EMNLP2026 main !

Cross-encoders are sota for ranking, but often too heavy

We ask a simple question: How many operations can be pruned while maintaining effectiveness?

👉 A lot more than you think 🧵

[Thibault Formal @thibault_formal]

Introducing SPARSEUP: the first model release from @Linkup_platform , and the missing sparse companion of DenseOn and LateOn.

Same backbone, same data, <150M, 56+ on BEIR-13. The model is open-source, use it!

Blog: https://t.co/Eo83WVH1WB

Model: https://t.co/bP4yGemPtM (Apache 2.0)

[Victor Morand @_VictorMorand_]

@thibault_formal @Linkup_platform Huge props for the great work and openness @thibault_formal @Linkup_platform 🎉

[Victor Morand @_VictorMorand_]

@bclavie Brr 🥶

[Victor Morand @_VictorMorand_]

@ManuelFaysse @AIatMeta Ok got it thanks a lot !

[David Bau @davidbau]

If all computer scientists do is create superhuman AI, we will have failed. Our goal must be to make *humans* smarter.

One of my 2026 teaching goals is to teach students to master the difficult art of using AI agents in ways the students get smarter, not dumber.

Lesson ideas?

[Bo @bo_wangbo]

We released pplx-embed-v1 early this year, with Q2D benchmark to evaluate how embedding model performs for web search, not looking at nDCG@10, but Recall@1000.

Q2D-Web further scale it up 7 times. With 70k agentic reformulated queries and 190 million web corpus. We hope the benchmark becomes the modern “MS MARCO”

[Perplexity @perplexity_ai]

We're introducing Q2D-Web (Query2Doc-Web), a benchmark and public leaderboard for evaluating retrieval in agentic RAG systems.

Q2D-Web tests how embedding models perform on large-scale web search using agent-reformulated search queries.

Read more: https://t.co/s476SxkE1L

[Victor Morand @_VictorMorand_]

@AmelieTabatta Thanks a lot 😊

[Victor Morand @_VictorMorand_]

🐭 Excited to share MICE (Minimal Interaction Cross-Encoders) - accepted to #EMNLP2026 main !

Cross-encoders are sota for ranking, but often too heavy

We ask a simple question: How many operations can be pruned while maintaining effectiveness?

👉 A lot more than you think 🧵

[Victor Morand @_VictorMorand_]

@AIQuanting Very fair point ! We're experimenting with that to offer a full retrieval pipeline soon.
For this specific question, yes you can train only the top layers, or even unbound query / doc bottom layers to train only the queries 🙂

[Victor Morand @_VictorMorand_]

@tomaarsen @LightOnIO Indeed its not obvious to fit - I implemented it as a CrossEncoder for now (so no pre-computation by default) but plan to adapt it to fit MultiVectorEncoder with custom Scoring fx !

[Victor Morand @_VictorMorand_]

@paulomouraj Thanks mate 😁

[Victor Morand @_VictorMorand_]

@antoine_chaffin Thanks ☺️ cool stuff coming !

[Victor Morand @_VictorMorand_]

@tomaarsen Thanks a lot ! We plan to release sentence-transformers compatible checkpoints trained with your Ettin-reranker recipe (thanks to the amazing @LightOnIO data) very soon 😁

[Victor Morand @_VictorMorand_]

Thanks to my 🐐 team mate @MathiasVast1, and advisors @BenjaminPiwowa1 and @JosianeMotheFr !

[Victor Morand @_VictorMorand_]

6/6 📄 Read the full paper and check out the code to try it in your IR pipeline:

- 💻 Code: https://t.co/Rf29JWh8Sa
- 📝 Paper: https://t.co/QU6cOQJqWL