Web This site Trusted
30 result(s) for “okapi bm25 site www eyespy cy site www is gov site does nvidia com site can site:arxiv.org” of 54 matched · 500 ms
Scope: Nishi’s own index (our crawl + Common Crawl) — not the whole live web. Some sites block all automated crawlers (anti-bot walls), so their pages may be absent here even though we never filter lawful content.
Freshness: newest result fetched 24 d ago · median age of these results 24 d · 1 of 30 carry a fetch time
For each evaluation sample, we construct an Okapi BM25 index over its decoded text chunks; the index is therefore sample-specific rather than a shared external knowledge base
from the top result below
https://arxiv.org/html/2608.03048v1
com/JetAstra/PI-Mem
Introduction
The ability to reason over long contexts is increasingly important for large language models (LLMs), supporting applications such as long-document question answering (Gemini Team 2025; Yang et al
https://arxiv.org/html/2608.02613v1
A million-token cloud context window does not help here, as it requires off-device upload, which privacy-sensitive personal memory management is meant to avoid
https://arxiv.org/list/cs.AI/new
, the meaning of a customized field) originates from users' domain knowledge and is rarely documented publicly
https://arxiv.org/list/cs.CL/new
LG)
Half a million cuneiform clay tablets survive in museums worldwide, yet modern users can neither read nor write in the world's oldest writing system, leaving a 4,000-year cultural barrier that existing NLP tools have only partially addressed
https://arxiv.org/html/2501.13956v1
com
Abstract
We introduce Zep, a novel memory layer service for AI agents that outperforms the current state-of-the-art system, MemGPT, in the Deep Memory Retrieval (DMR) benchmark
https://arxiv.org/list/cs.IR/new
IR)
Industrial recommenders increasingly adopt the pretrain-then-transfer paradigm, yet behavioral distribution drift raises two questions: what to learn from behavior sequences, and how to transfer the learned knowledge while the pretrained model is continually refreshed
https://arxiv.org/html/2608.01085v1
Its defining property is threshold-conditioned collective activation: cue realizations can vary, while activation is governed by the aggregate peer evidence they convey
https://arxiv.org/html/2608.02668v1
com/hazdzz/deep_connection \ncudata[Correspondence]Jie Zhang (),
Correspondence: Cheng-Fang Su ()
“Geometry is not true, it is advantageous
https://arxiv.org/html/2608.03703v1
This project was begun on July 20, 2026; the present version is dated August 4, 2026
https://arxiv.org/list/cs.CV/new
CV)
Fine-grained wildlife classification in aerial imagery is limited not only by model performance, but also by unreliable labels: animals occupy few pixels, key visual cues vary seasonally, and modality-specific evidence can be ambiguous
https://arxiv.org/list/math/new
It is natural to expect that if many pairs $(x_i,x_j)$ lie at distance close to $1$ from each other, then some clustering phenomenon must occur, implying that a significant number of these pairs are also very close to each other
https://arxiv.org/abs/2608.03883
Our systems combine sparse BM25 retrieval, evidence-aware rescoring for self-referential symptom reports, embedding-based reranking, query-prototype expansion, and LLM-based reranking
https://arxiv.org/abs/2104.08663
Our results show BM25 is a robust baseline and re-ranking and late-interaction-based models on average achieve the best zero-shot performances, however, at high computational costs
https://arxiv.org/abs/2608.03860
The pipeline combines sentence-window chunking, BM25, BGE-M3 dense retrieval, reciprocal rank fusion, optional cross-encoder reranking, and grounded answer generation
https://arxiv.org/list/cs.IR/recent
02048 [pdf, html, other]
Title: SmartGR: Hierarchy and Beam-Aware Knowledge Distillation for Generative Recommendation
Ziheng Zhang, Yu Cui, Bohao Wang, Yong He, Chao Yu, Chuan Yuan, Wujie Sun, Can Wa
https://arxiv.org/tb/2005.14165
Trackback guide
By sending a trackback, you can notify arXiv
https://arxiv.org/html/2407.16741v3
How can we ensure that development is safe and avoids negative side effects on the users’ systems
https://arxiv.org/html/2608.03494v1
Abstract
Vocabulary extension is an efficient way to adapt pretrained large language models (LLMs) to new languages, but the initialization of newly added token embeddings can strongly affect continued pre-training (CPT) efficiency
https://arxiv.org/html/2608.03035v1
Moreover, the truth of factual statements usually does not need to be updated throughout the course of a conversation, whereas an LLM needs to constantly update its representation of a contextual statement: in multi-agent collaborative settings, “truth” is often whatever has been established in the conversation
https://arxiv.org/html/2607.28127v1
It is therefore natural to ask:
•
Can we develop a financial sentiment analysis framework that extends beyond static labelled training data; can this be achieved by incorporating real-time market dynamics, thereby aligning textual sentiment with subsequent market actions
https://arxiv.org/html/2608.01024v1
While the evolutionary methods can in principle be run under such non-uniform demand, their search is not informed by it: random mutation does not know which shelves are requested frequently, so placing the high-frequency shelves well is left to chance
https://arxiv.org/list/cs.DC/new
The main question of interest is how close one can remain to the desired palette size of $\Delta+1$ while computing, in few distributed rounds, a coloring whose frequencies are close to $\sigma$
https://arxiv.org/html/2608.00648v1
The framework is suitable for a broad range of distributed task-planning applications and can be flexibly transferred to related scenarios with different practical constraints and settings
https://arxiv.org/abs/2404.05321
265 (HEVC), AV1 (SVT-AV1) along with direct comparisons to the latest AV1 encoder inside NVIDIA GPUs (40 series), and AWS Mediaconvert's AV1 implementation
https://arxiv.org/abs/2609.14845
Finally, as a proof of concept for what experimental research can be done with the models, we have utilised them in two demonstrative numerical applications, quantifying application-level accuracy differences between AMD matrix cores and the NVIDIA tensor cores
https://arxiv.org/abs/2608.01891
We implement AFlex in SGLang and evaluate it on NVIDIA A800 GPUs using Qwen3-32B and Mixtral-8$\times$7B under production Conversation and Coding traces
https://arxiv.org/abs/2608.02122
8% on average on NVIDIA L40S and L4, and reduces EDP by 9
https://arxiv.org/html/2510.03215v2
Report GitHub Issue
Description:
arXiv is now an independent nonprofit
https://arxiv.org/list/cs.MM/new
MM)
The next generation of AI agents is increasingly moving beyond systems that answer isolated questions toward persistent personal assistants that can understand, remember, and continuously learn from users' experiences
https://arxiv.org/list/cs.AR/new
Our analysis shows that, despite its limited write performance, HBF can significantly improve the batch size, throughput, and flexibility of LLM-serving systems while reducing the minimum GPU requirements, but realizing these benefits critically depends on sustaining HBM-comparable read bandwidth and requires