Macé Quentin (@MaceQuent1) profile picture

Cifre PhD student @ ILLUIN Technology & MICS & LIG | Passionate about NLP research. Work on multimodal information retrieval & reliability of LLM based systems.

Joined November 2020
I'm thrilled to finally announce the release of ViDoRe V3! 🚀 This is our new multimodal retrieval benchmark, built from the ground up for enterprise relevant RAG. We've moving beyond synthetic data to tackle the applied challenges we face every day. (1/N)
📢 ViDoRe V3, our new multimodal retrieval benchmark for enterprise use cases, is finally here! It focuses on real-world applied RAG scenarios using high-quality human-verified data. huggingface.co/blog/QuentinJ… 🧵(1/N)
3
3
4
533
Macé Quentin retweeted
🧩 Can you tell a real book from AI writing? We built a fun little game called fauxlio.app Two excerpts: one's a classic, one's AI, and your goal is to pick the real one. 5 new rounds a day, EN/FR, free, no sign-up. Tell @VictorTXing and me what you think! 🫶
2
2
2
6
599
Another Late Interaction win :)
Should you waste tokens having your coding agent use grep, or hand the search to a tiny local model? We release ColGrep-Agent-2B! A code-search agent that runs locally, searches with ColGrep, reads files, & returns locations + code to your main agent -> 34% fewer API tokens
4
72
Macé Quentin retweeted
Our latest work at @illuintech is out! 🕺 DAEDALUS is a training-free pipeline that generates priors on any new environment, that an agent can use in-context at test time. No tasks/verifier/human-written guidelines required! Details and fav take-aways in the thread below 👇 (1/n)
1
3
11
362
Congrats ! another step towards self-improving agentic system,s done in a very elegant and rigorous way ! Check their work out 👇 Particularly fond of the blogpost page :O
🏛️ LLM agents in a new environment repeat the same mistakes. Meet DAEDALUS: agent memory bootstrapped from tasks the agent writes itself, with no training tasks or human-written guidelines. It makes 7 agents from various model families succeed more often, in fewer turns 🧵⬇️
1
8
101
Macé Quentin retweeted
👀 Meet NeoMME: a family of 260M and 800M Multimodal-Native Multilingual efficient Encoders One bidirectional Transformer processes text tokens and raw image patches, with no pretrained vision tower, text encoder, or decoder. (1/N 🧵) arxiv.org/abs/2609.01657
11
30
9
171
23,041
Macé Quentin retweeted
🚨I've just released Sentence Transformers v6.0! MultiVectorEncoder joins the family: ColBERT-style late interaction models are now a first-class model type, for training, inference & interpretation, alongside dense, sparse & reranker models. Big thread 🧵
15
70
25
408
145,335
Macé Quentin retweeted
🚨 Do LLMs need to store everything they read in memory? To reduce KV cache size and improve decoding speeds, we propose Self-Pruned KV attention, a mechanism where the model learns to decide which KVs to write in the persistent KV cache, discarding all the rest! @AIatMeta🧵
9
45
8
199
23,555
Macé Quentin retweeted
ViDoRe V3 has been accepted to ACL 2026! Much more relevant however are the many practitioners I met in recent days that praised the benchmark for it's quality and clean signal. It's not easy making a good retrieval benchmark (non-saturated, practical, wide domain coverage) but it helps having done 2 imperfect earlier iterations, understanding what is needed, and having 12k hours of human annotation (thanks @NVIDIAAI ) + 6 months of hard work from notably @MaceQuent1 @antonio_loison @antoine_edy) amongst others. From a scientific perspective, I believe we converged to a nice recipe that scales but remains challenging by mixing careful taxonomy design (@omrani_bilel), synthetic pre-annotations, and humans in many parts of the loop. This is by no means a cheap project - and going forward, non-trivial benchmarks will probably be more and more expensive. Gone are the days where using a LLM to annotate data creates a sufficient proxy for model improvement (ViDoRe V1). Some people asked me when ViDoRe V4 would be out. I would assume this should not come from us - we are reaching the limits of data annotation, what we need now are fully real queries and real documents from real users of VDR (and more generally IR) models. This comes with data privacy issues and is tough but I believe it should be the north star - too many datasets in IR are completely toy and optimizing them actually hurts the model. In all cases, ViDoRe V3 has a lot more to offer than just VDR (agents, RAG, etc) and should have at least a few months of non-saturation ahead of it!
7
18
2
70
3,726
Macé Quentin retweeted
Replying to @mlpc123
@mlpc123, @MaceQuent1 and I will be at the Late Interaction Workshop at #ECIR2026 on Thursday 🤗 See you in Delft 🇳🇱
Working Notes on Late Interaction Dynamics: Analyzing Targeted Behaviors of Late Interaction Models Analyzes length bias and similarity distributions in late interaction retrieval, showing causal multi-vector models suffer from monotonic length bias. 📝 arxiv.org/abs/2603.26259
1
4
1
14
3,311
Very proud of this paper, we lead many more experiments since the first release. I think we made a pretty complete analysis of what is currently possible with the benchmark ! Thanks again to everyone involved !
Two months ago, we released ViDoRe V3 benchmark. Now, the full paper is out! 📄arxiv.org/abs/2601.08620 Here is a recap of the benchmark along with a comprehensive breakdown of our findings on multimodal RAG. 🧵 (1/N)
2
5
96
Macé Quentin retweeted
Two months ago, we released ViDoRe V3 benchmark. Now, the full paper is out! 📄arxiv.org/abs/2601.08620 Here is a recap of the benchmark along with a comprehensive breakdown of our findings on multimodal RAG. 🧵 (1/N)
1
9
2
16
721
I'd like to thank everyone involved in the project from the (many) annotators to each contributors. In particular, shout out to @antoine_edy @GautierViaud @VictorTXing @ManuelFaysse . Also a big thanks to the whole NVIDIA team who helped us and contributed to this project (N/N)
1
51
Adding to @antonio_loison's post i'll say that this is truly a new step towards complex evaluation of RAG components. From VQA to complex RAG pipelines evaluation including grounding, our annotations should enable the community to build up on many new exciting things ! (2/N)
30
New paper alert! 📄 "ModernVBERT: Towards Smaller Visual Document Retrievers" We introduce a new small, yet very powerful, vision language retriever that matches performance with models 10x its size on ViDoRe ! More details in @pteiletche's post
Introducing ModernVBERT: a vision-language encoder that matches the performance of models 10× its size on visual document retrieval tasks! 👁️ Read more in the thread👇 (1/N)
2
1
3
274
And of course thanks @illuintech for enabling us to work on such amazing projects
34
Thanks a lot for everyone that worked on this, it's been a real pleasure to work with all of you guys ❤️ @ManuelFaysse @pteiletche @mlpc123 @antonio_loison
2
40
Macé Quentin retweeted
Afraid of text2any retrieving? No worries, colqwen-omni has made it simple! Check out Manu's thread for some demos👇
Introducing ColQwen-Omni, a 3B omnimodal retriever that extends the ColPali concept of multimodal retrieval with late interaction to audio chunks and short videos, with no performance degradation on visual document retrieval wrt our best models! (1/N)
1
5
175
Macé Quentin retweeted
all modality RAG 🔥 ColQwen-Omni is a new multimodal retrieval model that can retrieve anything (videos, audios, documents and more!) use with transformers 🤗 here's a smol demo on video retrieval ↙️
7
57
2
407
14,738