home.social

#ssm — Public Fediverse posts

Live and recent posts from across the Fediverse tagged #ssm, aggregated by home.social.

fetched live
  1. Mamba: архитектура, которая шла убивать трансформеры

    В декабре 2023 по ML-тусовке прокатилась волна заголовков в духе «трансформерам конец». Поводом стала статья двух исследователей — Альберта Гу и Три Дао — со скучным названием: «Mamba: моделирование линейно-временных последовательностей с использованием селективных пространств состояний» . Внутри была архитектура, в которой не было механизма внимания, того самого attention, на котором держится весь современный тир-лист нейронок. И при этом она работала на длинных текстах в несколько раз быстрее трансформера, при меньшем расходе памяти. Прошло уже много времени, так что не будет спойлером сказать, что свой трон трансформеры не потеряли. Но история на этом не закончилась, и развязка интереснее, чем «очередной хайп-трейн не взлетел».

    habr.com/ru/companies/selectel

    #Mamba #SSM #трансформеры #attention #языковые_модели #нейросети #машинное_обучение #RNN #гибридные_модели #selectel

  2. New #openaccess publication #SciPost #Physics Core

    Interplay of entanglement structures and stabilizer entropy in spin models

    Michele Viscardi, Marcello Dalmonte, Alioscia Hamma, Emanuele Tirrito
    SciPost Phys. Core 9, 012 (2026)
    scipost.org/SciPostPhysCore.9.

    #UniversityNaplesFederico2 #INFN Napoli #ICTP #SSM
    #MUR

  3. MoE vs. SSM: Два пути из «Тирании Квадрата» Трансформеров

    Архитектура Трансформеров уперлась в стену квадратичной сложности O(n²), или «Тиранию Квадрата». В статье мы разбираем два пути решения этой проблемы: Mixture-of-Experts (MoE), масштабирующий знания, и State Space Models (SSM), масштабирующий контекст. Это сравнительный анализ архитектур, которые определяют будущее AI.

    habr.com/ru/articles/949826/

    #llm #moe #ssm #архитектура_трансформер #квадратичная_сложность #State_Space_Models #Mixture_of_Experts

  4. You know your day is starting when #CloudFront #SSM#GammaGroup proxy servers' in SF, CA try to hack you. Again. 😆

    #infosec 🤝 ☕ 🤝 👨‍💻

  5. So with a local #SSM State Space Model (#Mamba, #RWKV) I can snapshot contexts.

    I've now got a general purpose pipe/filter that can feed and read from any saved context, for stable static or dynamic (by re-saving the checkpoint) sessions.

    Any of which can become an always ready CLI filter I can pipe stuff through.

    *Could* make a local SSM model useful providing instantly ready small named natural language filters that work in the shell just like grep, awk, etc.

    Useful?

    Don't know yet.

  6. Turns out my llama_cpp code needed minimal changes to *just work* swapping Falcon-Mamba 7B in for RWKV7 7B.

    The Mamba is both faster, and based on a brief test with the same primer text - smarter.

    I still haven't figured out if either can actually be useful in any way, but State-Space Models in general are pretty awesome.

    huggingface.co/tiiuae/Falcon3-

    #ssm #mamba #falcon

  7. Thing I've been experimenting with the past few days -- "diegetic role based prompting" for a local State Space Model ( #RWKV7 currently).

    Tiny llama.cpp Python runner for the model and "composer" GUI for stepping and half-stepping through input only or input and generated role specified output, with saving and restoring of KV checkpoints.

    Planning to write runners for #XLSTM 7B & #Falcon #MAMBA 7B to compare.

    Started 'cause no actual #SSM saving, resuming examples.

    github.com/stevenaleach/ssmpro

  8. CW: #uspol #Election2024 #StateSponsoredMalware #DNS

    I.T. would be interesting on the ' what shows up in the 2M+ #GammaGroup installs list.txt from 7 swing states ' targeting 18-25yr olds when their #DNS all update to an open source #infosec provider & finish up their #investigations into #RICO via #SSM & unpaid licenses' on their clients ☣️🔍🧐

    #RTDNA #GammaGroup #SSM #Election2024 #RICO #uspol #ClientDNSupdates #OpenDNS ☣️🔍🧐

  9. Finally! After I don't even know how much head-banging, I've got a State Space Model (RWKV7) running using llama.cpp with GPU offloading and memory states successfully saved and restored. And I *think* the same code can be minimally modified for Mamba (Mistral, Falcon).

    Yay for infinite context length!

    And for massive time and energy savings.

    llama.cpp: github.com/ggml-org/llama.cpp

    Llama CPP Python: pypi.org/project/llama-cpp-pyt

    RWKV7 live demo: huggingface.co/spaces/BlinkDL/

    #SSM #llama #rwkv #mamba

  10. You’re security testing AWS infrastructure. You’ve done the work and need to exfiltrate the evidence files. But there's no internet access and no inbound ports... 🤔

    Here’s how to use AWS Services Systems Manager (SSM) to create a port forwarding session, access what you need, and securely exfiltrate data with a simple Python web server.

    📌 Read the blog post here: pentestpartners.com/security-b

    #AWS #CloudSecurity #PenTesting #CyberSecurity #SSM #Infosec #Python #RedTeam

  11. Часть 4: Mamba — State Space Models vs трансформеры

    Mamba — революция в обработке длинных последовательностей! Mamba — State Space Models vs трансформеры, что лучше?!

    habr.com/ru/articles/925416/

    #mamba #transformer #nlp #ssm

  12. PSA: If you use AWS Incident Manager and trigger incidents with Lambda you may need to allow the `ssm-contacts:StartEngagement` action for your lambda role.

    Our Incident Manager stopped working sometime in the past 2-3 days, and now works again after adding the statement to the policy. Still investigating when and what changed. 🤔

  13. People are often complicit with cruelty and injustice, especially when empathy and love come with a price tag.
    Will I still have the right to marry in the future? That’s unclear.
    buff.ly/3CWl3QM

    #LGBTQ #USPolitics #Transgender #SSM #Michigan

  14. So is 2025 the year we will get #Synology switches? Does this mean that #SSM #SynologySwitchManager will be the new OS? Time will tell...

  15. My first research summary post is about my collaboration with Andrew Anderson and Joseph Mozingo (University of Utah) and Harry Kim (Texas Scottish Rite Hospital for Children) to produce a framework for statistical shape modeling of Perthes' disease!

    lukejohns.online/research/shap

    #SSM #orthopaedics

  16. What iF there was an 18-24yr old group of new voters, let's say 3M of them, & #StateSponsoredMalware was used to ' force a vote a certain way '.

    That would definitely be on the ledger of investigations, for sure, right Brazil? 🙄🤨

    #uspol #SSM#VoterIntimidation #SwingStates #Elwction2024 🙄🤨
    #infosec #delivers_dtignite_com
    #GammaGroup #FinFisher #FinSpy #Finsky