31 Photos and videos
Jun 6
1/ποΈ Meet Argus-Retriever: the first late-interaction visual doc retriever where the document representation adapts to the query: D(q).
- 86.0 NDCG@5 on ViDoRe V1 V2 (SOTA open model)
- π¦ 1024-dim head, 4.5Γ smaller index
Jun 4
Argus-Retriever: Vision-LLM Late-Interaction Retrieval with Region-Aware Query-Conditioned MoE for Visual Document Retrieval
Introduces a query-conditioned late-interaction visual document retriever.
π arxiv.org/abs/2606.04300
π¨π½βπ» github.com/DataScienceUIBK/Aβ¦
1
7
52
8,392
Jun 6
2/ The idea: a region-aware Mixture-of-Experts inside the document encoder.
A router reads each region's content, its 2D position, and a pooled query context z_q, then mixes 4 latent experts ( 1 shared).
The page is now encoded differently per query β D(q). Still MaxSim.
2
250
Abdelrahman retweeted
Jun 4
Argus-Retriever: Vision-LLM Late-Interaction Retrieval with Region-Aware Query-Conditioned MoE for Visual Document Retrieval
Introduces a query-conditioned late-interaction visual document retriever.
π arxiv.org/abs/2606.04300
π¨π½βπ» github.com/DataScienceUIBK/Aβ¦
6
19
9,192
May 17
Thrilled to share two paper acceptances! π
π MM-BRIGHT β Multimodal Benchmark for Reasoning-Intensive Retrieval β #KDD2026
π οΈ Rankify β Python toolkit for Retrieval, Re-Ranking & RAG (670 β) β #ACL2026 Demo
Thanks to @ supervisor & collaborators π
#NLP #RAG #IR
3
7
51
2,695
another late interaction w
Apr 23
[1/3] tiny 32M reasoning Colber retriever, just dropped. π€
Reason-mxbai-colbert-v0-32m a late-interaction ColBERT fine-tuned for BRIGHT-style retrieval.
19.0 avg nDCG@10 on BRIGHT, ~5x smaller than Reason-ModernColBERT.
huggingface.co/DataScience-Uβ¦
1
2
14
1,465
Apr 23
[1/3] tiny 32M reasoning Colber retriever, just dropped. π€
Reason-mxbai-colbert-v0-32m a late-interaction ColBERT fine-tuned for BRIGHT-style retrieval.
19.0 avg nDCG@10 on BRIGHT, ~5x smaller than Reason-ModernColBERT.
huggingface.co/DataScience-Uβ¦
4
14
92
7,342
Apr 23
[1/3] Built on @mixedbreadai's mxbai-edge-colbert-v0-32m with one key change: we widened the projection head from 64β128 dim using small-random init (zero-pad = dead neurons under L2 MaxSim, classic gotcha).
1
6
451
Apr 23
[1/3] Still limited on splits (leetcode/theorem) β base model's sans_pos lowercased tokeniser caps code retrieval.
But Imporved Pony split from 8.0 NDCG@10 by Reason-ModernColBERT β 20 NDCG@10
4
362
Abdelrahman retweeted
Apr 19
For the people not on Reddit, yesterday ICLR desk-rejected an accepted oral paper from a US-sanctioned institution.
The conference is in 4 days; this could probably have been done better.
openreview.net/forum?id=w2tnβ¦
7
22
181
50,465
Abdelrahman retweeted
Apr 5
The paper acceptance notifications will be out by the 6th of April, AoE. The PCs are working hard throughout the holiday season to finalize the decisions.
Apologies for the delay!
35
19
136
43,211
Apr 2
We just released Claw Code Agent, a full Python reimplementation of the Claude Code agent architecture, based on the reverse-engineering work shared in this tweet:
x.com/Fried_rice/status/2038β¦
Repo: github.com/HarnessLab/claw-cβ¦
We're actively working on this and happy to add features.
Mar 31
Claude code source code has been leaked via a map file in their npm registry!
Code: pub-aea8527898604c1bbb12468bβ¦
1
2
393
27 Feb 2025
1
78
10 Feb 2025
π Introducing Rankify: A comprehensive Python toolkit for #retrieval, #reranking, and #RAG
π Supercharge your AI & NLP models with cutting-edge retrieval techniques.
β If you find it useful, give it a star!
π GitHub: github.com/DataScienceUIBK/rβ¦
#AI #NLP #Retrieval #RAG
1
70
Abdelrahman retweeted
22 Dec 2024
20 Dec 2024
$ethbtc
reached the retracement zone
lets see if this will play out
ΩΨ΅Ω Ω
ΩΨ§Ψ·Ω Ψ§ΩΨ§Ψ±ΨͺΨ―Ψ§Ψ―
ΩΩΨ±Ω Ψ§Ψ°Ψ§ Ψ§ΩΨ³ΩΩΨ§Ψ±ΩΩ Ψ³ΩΨͺΨΩΩ
6
5
104
38,949
Abdelrahman retweeted
22 Nov 2023
The top 15 most-liked organizations on @huggingface
1. @StabilityAI 20k likes
2. @AIatMeta 20k
3. @runwayml 11k
4. CompVis 10k
5. @thukeg 7k
6. @BigscienceW 7k
7. @TIIuae 7k
8. @Microsoft 6.5k
9. @GoogleAI 6k
10. @OpenAI 4k
11. @BigCodeProject 4k
12. @MosaicML 4k
13. @UKPLab 3k
14. @AiEleuther 3k
15. @salesforce 3k
huggingface.co/spaces/Pulsarβ¦
9
85
435
155,918