Rohan Jha

PhD Student @jhuclsp | Prev: @mixedbreadai, @JinaAI_

Interested in IR & NLP

[Rohan Jha @Robro612]

New 📄: we replicate XTR, a multi-vector retrieval method that makes ColBERT faster by avoiding its expensive step of gathering full document embeddings

XTR is not a free lunch over ColBERT, but its training objective is useful for modern efficient engines like PLAID and WARP 👇🏼

[Rohan Jha @Robro612]

@antoine_chaffin I was just re-searching for it again this morning (was curious if e-means intuitions could be applied to scale up ward)!

I wish we could search kernels from the main HF searchbar too rather than segmented to https://t.co/EEAm2vzxwz

[Swann Bessa @swann_bessa]

Introducing e-means, an algorithm beating k-means in 10x fewer epochs.

With order of magnitude wall-clock speedups, e-means is released in FAISS and needs no hyperparameter tuning beyond k. It is robust to various scales and datasets. Still skeptical ? Just try e-means [1/10]🧵

[Alexander Martin @alexdmartin314]

No one told me recreating the @cartesia blog post would be this fun (and this exciting!)

Thinking of blog posting some of the recreation / improvement process, but for now my favorite findings are below. (1/4)

[Rohan Jha @Robro612]

@swann_bessa I first heard about this technique in VecFlow-Chamfer (quite similar intuitions to TACHIOM but for unified-memory GPUs). It has the benefit of not requiring a tf-idf inductive bias (visual docs) but is definitely worse than TAC when you have token ids.

https://t.co/X0lxTH37Hs

[Rohan Jha @Robro612]

Very excited by the new e-means paper from @swann_bessa et al.! Better convergence at a fraction of the training time than conventional Lloyd kmeans.

The blogpost has lots of beautiful visualizations/animations of the studies in the paper!

[Rohan Jha @Robro612]

@swann_bessa One example of the cool methods/experiments discussed in the paper: index-accelerated approximate assignment!

Incidentally, I just submitted a PR to TACHIOM supporting this same technique (cpu, HNSW, not-batched) as an alternative to TAC clustering.

[Rohan Jha @Robro612]

@ManuelFaysse I came across his profile on here and saw that this work was upcoming. This was literally in my drafts 😂

[Rohan Jha @Robro612]

@AmelieTabatta @helloiamleonie +1, and any specific guidance on how to elicit these 🙂

[Diane @dianetc_]

Is there a way to scale a reasoning model’s judgment for searching a massive corpus?

We introduce the Amortized Reasoning Index (ARI), a new kind of retriever that updates its weights during search, using gradients derived at inference time from the ranking preferences of a reasoning model.

Every session, current agents spend lots of reasoning deciding which search results matter, and then throw all of that judgment away upon compaction. ARI is our attempt to “amortize” that judgment, so that reasoning spent on past queries keeps improving search on future ones.

With no relevance labels, ARI takes a 400M cross-encoder that starts out weak and grows its Recall@100 on held-out queries by up to 7×, ending above Qwen3-Embedding-8B on various oblique/reasoning-intensive search tasks.

Read our blogpost: https://t.co/UEPpEpw22d

Preliminary results and discussions are below!

[Neha Verma @n_verma1]

Need different model weights? Just change them 🙃

Our new paper VFold does exactly that to enable low-overhead cross-layer value cache compression: half the value cache, >98% of full-cache performance 🧵
📄 https://t.co/8oR3sRB31D

[tusKANNy @tuskannysearch]

Our paper on the core architecture behind tusKANNy is out.

Two libraries, one idea: separate the data from the index. 🧵

#VectorSearch #NeuralIR #RustLang

[Benjamin Van Durme @ben_vandurme]

An awesome final project by @zhengping_jiang as he wraps up the PhD. Since 2018 when I first met him there have been so many good papers. This one is especially topical to a lot of labs and businesses (he's on the market!).

[Zhengping Jiang @zhengping_jiang]

Long-horizon agents can take hundreds of actions before receiving a single outcome reward. Training them raises two questions:
➡️ Which actions deserve credit?
➡️ How does training fit with existing compaction and memory strategies?
Introducing FOCUS: active critic learning over managed context states, with modular, customizable context protocols and integrations with Tinker and slime.
Docs: https://t.co/luN4jQzfmG (1/4)

[Daniel Israel @danielmisrael]

I think LLM inference might be in the middle of a paradigm shift. Read my blog if you want my full thoughts. TLDR is that I think we are moving from batch parallelism to sequence parallelism

[Antoine Chaffin @antoine_chaffin]

@din0s_ @_reachsumit 🫡
this one is a bit bigger than the previous one, we promised a real paper

[Yuda Song @yus167]

Foundation models are trained in stages: pretraining, midtraining, and post-training. But is this separation fundamental?
We introduce unified training: a single algorithm for the full pipeline. This gives
⚙️ Simpler infrastructure
🔬 A framework for better science
(1/n)

[Rohan Jha @Robro612]

Highest compliment one can give to research-applicable software. I feel the same about PyLate, Sentence Transformers, and TuskANNy's libraries.

[Ivan Rubachev @puhsuuu]

btw, tabpack doesn't just give you good out-of-the-box performance on benchmarks, it also speeds up research. We had a lot of MLP models to run for our paper's scaling study:
https://t.co/hmiFCEQ00H

We wouldn't even have considered running a grid this big before tabpack

[samsja @samsja19]

you can just sparsify and compress thing

[Saron Samuel @itssaronsamuel]

Applying to PhD programs is hard! Take advantage of this opportunity, especially if you may not otherwise have access to application support:
https://t.co/kM3cevFbNU