Top Tweets for #Sglang
Open models need open infrastructure, and this crowd shows how many people are building it with @sgl_project @radixark Packed halls, big energy! #SGLang #Summit
Honored to present our work, “Boreal: Towards Self-Improving Video Generation,” at the SGLang Summit! @sgl_project #SGLang
An AI agent can spit out tokens and still leave you staring at a spinner. SGLang’s main stage starts at 11:30 pm IST tonight.
I want to see two numbers on the same slide. Tokens per second, and time until the actual job is done.
#SGLang
SGLang Flaw Let an Attacker Breach NEAR AI's "Confidential" Inference Servers
#LLMSecurity #NearAI #OpenSourceSoftware #Sglang #Vulnerability
https://t.co/kw5l1ZwZaq
🚀 Crypto expert: NEAR AI patched a SGLang flaw exposing ~100 users, 18 legacy agents, and 2 API keys; Cloud service is back online. Security-first: AI-clouds need constant audits. #NEAR #AI #SGLang #CloudSecurity #Web3 #crypto #cryptonews 🛡️
NEAR AI, SGLang açığıyla bir sunucusunun ele geçirildiğini duyurdu: ~100 kullanıcının hesap bilgisi ve 18 ajanın verisi alındı. Gizli moddaki istekler de görülmüş olabilir.
https://t.co/Joevr06LUM
#NEARAI #SGLang #yapayZeka #veriİhlali
A critical #SGLang RCE vulnerability exposed sensitive data from NEAR AI’s inference servers, impacting agents and users. SOC teams, prioritize monitoring and testing open-source AI frameworks to prevent similar breaches. #NEARAI #RCE #AIsecurity #SOCMinute
NEAR AI Breach Hit 100 Users Via SGLang Flaw
Read the full story: https://t.co/TmLIYiLRA3
An SGLang remote-code-execution flaw let an attacker access records for roughly 100 NEAR AI users and data from 18 legacy agents.
#NEARAI #SGLang #databreach #remotecodeexecution #AIsecurity
AI Attribution:
Using AI to create intelligent content ◈ = AI creator
It's been a lot of fun and learning to work it out, with @connorliml, @tianyi___zhang, @JueWANG26088228, @ZhongzhuZhou, and @Chenfeng_X!
#AgenticAI #MLsys #vLLM #LLMinference #DeltaNet #mamba #attention #gpu #inference #infra #nvidia #KVCache #AIinfra #SGLang #kimi #glm #qwen #LLM
I’ll be speaking at SGLang Summit 2026 on November 12 in San Francisco. Looking forward to connecting with the SGLang community.
https://t.co/P6fOEWQZyD
#SGLang #AIInference #Nebius
📋 How to run eight #NVIDIA #DGXSpark boxes as a headless #LLM inference cluster: two groups of four, each serving #DeepSeek V4.1 Flash with tensor parallelism over #RoCE, #SGLang on one group and a #vLLM fork on the other, linked by two #MikroTik CRS804 switches. Stock NVIDIA kernel and driver; all changes are settings, not patches.
⚡ Power: a GB10 has no power-limit setting, the GPU clock is the only lever. At stock clocks the rack averaged 1,717 W and tripped its 15 A strip after about eight minutes. A 2,200 MHz clock cap via a systemd unit brought it to 1,345 W (−22%) with no drop in decode speed, because token generation is memory-bound. Measure at the wall: nvidia-smi shows only the GPU rail, the wall reads 3–4.5× higher.
🧠 Memory: CPU and GPU share 128 GB, so the desktop, the display reservation and the page cache all come out of the model's budget. Headless mode plus 0.5 GB display memory (needs the October 2026 firmware) lifts claimable GPU memory from 117.0 to 119.0 GiB per box, +16 GiB across eight. The page cache is flushed before every model start.
🛡️ Security: two open kernel bugs. A public local root exploit is blocked with user.max_user_namespaces=0. The remote CVE-2026-98432 (CVSS 9.8, use-after-free in the IPv4 TCP handshake, no login needed) is mitigated by an #nftables rule that drops TCP packets carrying IP options, plus MPTCP off. The rule lives in its own table, not the stock nftables.service, whose default config would flush Docker's rules.
🌐 Network: two subnets per 200G cable, MTU 9000, seven #NCCL variables (one fix took a 1 MiB all_reduce from 1.36 to 11.5 GB/s), RDMA in its own switch queue with ECN and no PFC. Test with NCCL, not just perftest: on one kernel perftest passed 72 of 72 runs while NCCL failed every one.
🔧 Updates: auto-updates off, a monthly manual routine, four boxes first so the other four keep serving. One package silently flashed the ConnectX-7 firmware during apt install. NVIDIA's display-reset fix also brought RDMA latency from 3.35 back to 2.50 µs. Allow 20 minutes per reboot and never pull power.
📊 Every setting has an ID, a check command, apply, undo and reboot flag, plus a machine-readable JSON copy and a read-only check script for AI agents. #devops #homelab
https://t.co/4vFNgvYtmg
@MicrosoftAI recruiting and infra/ research teams will be at #sglang and #neurips over the next month! Like and follow for updates on our presence, say 👋 if you want to meet our team directly - can’t wait! #microsoftai
Last Seen Hashtags on Sotwe
realorgasm
Seen from Spain
แตกในตูด
Seen from Thailand
rojaselvamanihot
Seen from Oman
MGAI(((************************************************))) filter:videos since:2026-11-01
Seen from Germany
pemandianairhangat
Seen from United States
ibuibustw
Seen from Indonesia
brosis
Seen from Ireland
doggy
Seen from Latvia
Konya
Seen from Turkey
fried desire
Seen from Korea
Trends for you
Most Popular Users
Elon Musk 
@elonmusk
241M followers
Barack Obama
@barackobama
118.8M followers
Cristiano Ronaldo
@cristiano
114.5M followers
Donald J. Trump
@realdonaldtrump
111.9M followers
Narendra Modi
@narendramodi
107M followers
Rihanna
@rihanna
98.7M followers
NASA
@nasa
92.2M followers
Justin Bieber
@justinbieber
91.8M followers
KATY PERRY
@katyperry
90M followers
Taylor Swift
@taylorswift13
84M followers
Lady Gaga
@ladygaga
75.5M followers
Virat Kohli
@imvkohli
73.4M followers
Kim Kardashian
@kimkardashian
70.8M followers
YouTube
@youtube
68.7M followers
Neymar Jr
@neymarjr
66.4M followers
Bill Gates
@billgates
65.2M followers
Selena Gomez
@selenagomez
63.1M followers
The Ellen Show
@theellenshow
62.2M followers
CNN
@cnn
61.7M followers
X
@x
60.6M followers