Standing under Longmen Grottoes? Wow. No words really. It's wild to think generations of people carved all this into the cliff over 1,000 years ago. I truly hope this human miracle stays safe and is well protected for many more generations to see. Peace and love. #LongmenGrottoes# #UNESCO# #Luoyang# #Henan#

When I dress this set of Hanfu, I visited the world famous Longmen Grottoes, where Buddha and Bodhisattva statues are preserved from Wu Zetian Emperor (China's only female emperor) time. I immediately dive into the peace of the Buddhism spiritual world. I feel I'm part of its universe, merging into the elements of ancient Chinese civilization. Absolutely a life-changing experience. #buddha# #tourists# #hanfu#

Chinese and Greek scholars explore the art of stone carvings, noting that it makes moments timeless — this is the power of art. #Travelinchina# #ChinaTourismDay# #AllTheWayToLuoyang# #peony# #ChinaTravel# #fypシ゚viral# #Trending# #LUOYANG# #BlossomsAllTheWayToLuoyang# #ZoomerChina# #longmen# #GREECE#

#Luoyang#, an ancient capital with a history of over 5,000 years of civilization, stands as a major cradle of Chinese culture. Nestled against Mount Mang to the north and overlooking the Funiu Mountains to the south, with four rivers flowing through the city, it boasts a topography unrivaled under heaven. As the ancient capital of thirteen dynasties—from Zhenxun of the Xia era to the Divine Capital of the Wu Zhou period—historical imprints quietly endure in cultural heritage sites such as the Longmen Grottoes and the White Horse Temple, bearing witness to the glorious past and enduring vitality of Chinese civilization. (Video by Huayun Bilingual Cultural Communication Society of North Minzu University) #ChinaCulture#

This is China — Longdeng Grassland. Home to the self-growing Command Platform of King Gesar, the jade-green Zhumucuo Lake ，and a mystical meadow that changes its colors 12 times a year. An ultimate destination for explorers and culture enthusiasts. #OffTheBeatenPath# #KingGesar#

Every AI product will need memory. Here's EverOS. Next release (internal): LoCoMo 94.74% · AMA 80.89% · BEAM-1M 69.42% · SWE-ContextBench 34.34% Current release: LongMemEval 94.00% Built around two ideas: Self-evolving. Every task becomes experience, experience becomes skills, and skills keep improving with use. Your agent gets better without retraining a model or rewriting a prompt. Universal. One memory layer for any agent, any model, any app. Each agent gets exactly the context its task needs, in open source or in the cloud, on the same API. Full write-up 👇

From 8,594 tokens to 1,946 per answer. EverOS selects the memories a question actually needs instead of injecting a fixed 20 every time. In our LongMemEval-S test, that meant 77% fewer answer-stage input tokens, with 91% accuracy versus 93.4%. Less context to process. Relevant evidence to work with.

Dear Doctor, Experience the convenience forged by Yan's craftsmanship. With ample storage for your daily needs and a unique design that showcases your style, this Lungmen Backpack is not to be missed. Sling it over shoulders, Doctor. Now you're one of Lungmen. Order Now: #Yostar# #Arknights# #Arknights6thAnniv#

🧠 Is agent memory really a better deal than plain RAG? Title: Memory Tax: Dissecting the Net Value of Agent Memory URL: ❓ What is the "net value" of memory? 💡 Compared with naive RAG, queries memory newly solves are its "benefit," and queries RAG solved but memory dropped are its "tax." The paper measures the balance between the two. ❓ What did they find? 💡 Across 10 memory systems and 3 benchmarks, the tax often outweighed the benefit. On LongMemEval-s, with RAG at 90.0, MemoryOS scored 80.0, MAGMA 76.0, and OpenViking 63.0. ❓ Why does the tax arise? 💡 Three reasons: ・Facts get lost during construction (38.0% of facts had no memory unit in OpenViking) ・Query-evidence similarity drifts after construction, making evidence harder to retrieve ・Swapping the backbone between construction and answering swings scores more than for RAG ❓ So how should memory be designed? 💡 The paper suggests also retrieving from raw history and reranking jointly with memory, and storing past queries and retrieval paths as recursive memory. #AIAgents# #rag#

For long-context LLM inference, should the KV cache be offloaded to disk or just recomputed on the GPU? There's no universal right answer, and this paper builds a system that decides quantitatively. Title: Building py-kvcache: A Performance Characterization of External KV Caching for vLLM with NVMe SSDs URL: 📝 Overview The paper proposes py-kvcache, an external KV cache system for vLLM. Using io_uring for async I/O, even a Python implementation pulls near-full SSD read bandwidth of 13.5GB/s. ❗ Problem it solves Existing external KV caches like LMCache have no criteria for when to actually use them — they load from disk unconditionally even when a short prefix or a fast GPU makes recomputation cheaper. ⚙️ Methodology It introduces "scheduler-aware preloading," which starts disk reads while a request is still waiting to be scheduled, plus a "break-even gate" that rejects a load whenever it wouldn't improve time-to-first-token. 📊 Results On LongBench multi-document QA it's 6.02-7.43x faster than GPU recomputation and 2.77-3.64x faster than LMCache. On multi-turn SCBench, native vLLM read 3.4TB from disk with completion times over 1200s, while py-kvcache kept disk reads to 85GB and completion time to 480s. 🖥️ Use cases On a high-end H100, requests often don't even clear the break-even point, so skipping external caching is fine — but on a lower-end RTX 4000 Ada, external caching clearly wins, giving concrete hardware-specific guidance. #LLMInference# #vLLM#