Rohan Jha

385 posts

Rohan Jha banner
Rohan Jha

Rohan Jha

@Robro612

CS PhD Student @jhuclsp Previously: Intern @JinaAI_, MS CS @UTAustin, BS AI @carnegiemellon Interested in Information Retrieval and NLP

Baltimore, MD Katılım Haziran 2015
429 Takip Edilen281 Takipçiler
Rohan Jha
Rohan Jha@Robro612·
Claude code skill that oneshot adds a dataset to ir_datasets
Rohan Jha tweet media
English
1
0
7
333
Rohan Jha retweetledi
Sumit
Sumit@_reachsumit·
AMES: Approximate Multi-modal Enterprise Search via Late Interaction Retrieval Apple presents a multimodal late interaction retrieval method deployed in Apache Solr, combining parallel token-level ANN candidate generation with Exact MaxSim re-ranking. 📝 arxiv.org/abs/2603.13537
English
2
6
36
6.4K
Haocheng Xi
Haocheng Xi@HaochengXiUCB·
𝗞-𝗺𝗲𝗮𝗻𝘀 𝗶𝘀 𝘀𝗶𝗺𝗽𝗹𝗲. 𝗠𝗮𝗸𝗶𝗻𝗴 𝗶𝘁 𝗳𝗮𝘀𝘁 𝗼𝗻 𝗚𝗣𝗨𝘀 𝗶𝘀𝗻’𝘁. That’s why we built Flash-KMeans — an IO-aware implementation of exact k-means that rethinks the algorithm around modern GPU bottlenecks. By attacking the memory bottlenecks directly, Flash-KMeans achieves 30x speedup over cuML and 200x speedup over FAISS — with the same exact algorithm, just engineered for today’s hardware. At the million-scale, Flash-KMeans can complete a k-means iteration in milliseconds. A classic algorithm — redesigned for modern GPUs. Paper: arxiv.org/abs/2603.09229 Code: github.com/svg-project/fl…
English
35
196
1.7K
278.1K
Rohan Jha retweetledi
Pau
Pau@hugemensa·
WARP just got faster and more memory efficient ⚡️ New 0.2 version focuses on CPU upgrades, increasing QPS by almost twofold and slashing memory usage during search by 60-80%, all without a change in the metrics ⚙️ Technical details below
English
1
2
5
146
Rohan Jha retweetledi
Ben Clavié
Ben Clavié@bclavie·
I'm so excited to introduce this! We've worked on a million different moving parts to produce this. I'm fairly confident it's the best multimodal model that exists, period -- and it's not too shabby at pushing back the LIMITs of retrieval either...
Mixedbread@mixedbreadai

Introducing Mixedbread Wholembed v3, our new SOTA retrieval model across all modalities and 100+ languages. Wholembed v3 brings best-in-class search to text, audio, images, PDFs, videos... You can now get the best retrieval performance on your data, no matter its format.

English
37
40
410
138.3K
Rohan Jha retweetledi
Sumit
Sumit@_reachsumit·
Does Reasoning Make Search More Fair? Comparing Fairness in Reasoning and Non-Reasoning Rerankers @itssaronsamuel et al. present a systematic comparison of fairness between reasoning and non-reasoning rerankers. 📝 arxiv.org/abs/2603.10332
English
0
3
10
647
Antoine Chaffin
Antoine Chaffin@antoine_chaffin·
let's call it a vocabulary
English
1
0
3
54
Aamir
Aamir@aaxsh18·
seems like bm25 works beyond text. single dense vector retrieval is dead.
English
13
1
67
16.9K
Rohan Jha retweetledi
Sumit
Sumit@_reachsumit·
Beyond Relevance: On the Relationship Between Retrieval and RAG Information Coverage @itssaronsamuel et al. investigate whether upstream retrieval metrics can predict downstream RAG information coverage. 📝 arxiv.org/abs/2603.08819
English
1
7
19
1.1K
Rohan Jha
Rohan Jha@Robro612·
@antoine_chaffin @raphaelsrty IMO you need to hammer harder on the point in this tweet of yours. People saying grep is all you need don’t realize that you can be more targeted and that it has tangible cost benefits at the agent task level, rather than just search problems x.com/antoine_chaffi…
Antoine Chaffin@antoine_chaffin

Tokens aren’t free On our 135 questions bench, we saved around 32$ As a rule of thumb, this means 243$/1k question It starts to add up pretty quickly, especially given large team usages

English
2
1
9
1.6K
Antoine Chaffin
Antoine Chaffin@antoine_chaffin·
We have receive a lot of very positive feedback on ColGrep and LateOn-Code from people trying it out Also read a lot about how it should be more popular given its power Any idea how we could spread the word? Should we make a collab with CC/Codex somehow?
Alex@santangelx

I shared this with friends and colleagues and now all the smart ones are using it Someone even built it into its product with postgres pgvector Ironic that a search product is hard to find 😅

English
3
3
21
1.8K
Sumit
Sumit@_reachsumit·
Visual Words Meet BM25: Sparse Auto-Encoder Visual Word Scoring for Image Retrieval Introduces BM25-V, a sparse image retrieval method that applies Okapi BM25 scoring to Sparse Auto-Encoder visual words from ViT patch features. 📝 arxiv.org/abs/2603.05781
English
1
7
44
2.6K
Rohan Jha retweetledi
Leo Boytsov
Leo Boytsov@srchvrs·
🧵For the last seven years, I kept re-implementing the same pattern: A parallel map loop that divides the work among several processes or threads. My very first attempts were built on Python’s standard tools, e.g., multiprocessing.map... ↩️
English
4
12
70
15.8K
Rohan Jha retweetledi
Jeremy Howard
Jeremy Howard@jeremyphoward·
According to OpenAI, their contract with the US DoW locks in current law, "even if those laws or policies change in the future". Our legal analysis, with Virgil Law CEO @LukeVerswey, shows that this is almost certainly incorrect. answer.ai/posts/2026-03-…
English
17
51
456
31.5K
Rohan Jha retweetledi
Neha Verma
Neha Verma@n_verma1·
Introducing "DOTResize: Reducing LLM Width via Discrete Optimal Transport-based Neuron Merging" ! We introduce an optimal transport framework for Transformer width compression that redistributes signal across neurons rather than eliminating them 🚚⚖️ 🧵 1/6
Neha Verma tweet media
English
1
7
16
1.1K

Keşfet

@HaochengXiUCB @antoine_chaffin @itssaronsamuel @aaxsh18 @ManuelFaysse @rasdani_ @willccbb @PrimeIntellect