Skip to content
View vukrosic's full-sized avatar

Block or report vukrosic

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don't include any personal information such as legal names or email addresses. Markdown supported. This note will be visible to only you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
vukrosic/README.md

Hi there 👋

My blog posts:
📄 Early observations: QK Norm seems to hurt Muon optimizer LLM training: English | Chinese

📄 Early LLM (transformer) layers pay attention to many tokens to gain context, later layers focus on fewer "important" tokens: English | Bilingual Chinese

📄 Magnitude Attention: Don't Let Group of Similar Keys Steal Your Probability Mass: English | Chinese

📄 Theoretical Proposal for Hyperbolic Gated Delta Nets: English | Bilingual Chinese

🚀 LLMs, Transformers, Diffusion, JEPA, Reinforcement Learning


🔍 Seeking Open Source AI Lab

Goal: Join a collaborative AI research lab. I aim to do high-quality research with like-minded people and maintain a daily social media presence, posting everything I research to accelerate the adoption and progress of Open Source AI.

Core Requirements:

  • Maintain my current pace of sharing research content on social media without restrictions every day to promote Open Science.
  • Retain research autonomy to ensure that outputs benefit the public to the greatest extent through social media.
  • We may agree on potential publications and directions.

Languages:

  • 🇷🇸 Serbian (Mother Tongue)
  • 🇺🇸 English (Bilingual / Completely Fluent)
  • 🇨🇳 Chinese (Learning/Progressing — Goal: Professional proficiency in 2027)

Let’s build open-source AI together. ☕️


🔍 寻找开源 AI 实验室

目标: 加入一个协作式 AI 研究实验室。我希望与志同道合的伙伴共同开展高质量科研工作,并保持每日社交媒体活跃,发布我研究的所有内容,以此加速开源 AI 的普及与进步。

核心诉求:

  • 保持目前的节奏,每天在社交媒体上无限制地分享研究内容,以促进开放科学 (Open Science)。
  • 保持研究自主权,确保产出能通过社交媒体最大程度地惠及大众。
  • 我们可以就潜在的论文发表和研究方向进行协商。

愿意搬迁至中国。

语言能力:

  • 🇷🇸 塞尔维亚语 (母语)
  • 🇺🇸 英语 (双语/完全精通)
  • 🇨🇳 中文 (学习中 — 目标:2027 年达到专业水平)

让我们一起构建开源 AI。 ☕️


🔗 Connect with me

Pinned Loading

  1. evintunador/gpt-lab evintunador/gpt-lab Public template

    cheap & easy LLM experiments for amateurs (alpha)

    Python 25 11

  2. zero-to-ai-researcher zero-to-ai-researcher Public

    Research on training an LLM with DeepSeek & Kimi architecture

    Python 40 11

  3. Open-Superintelligence-Lab/blueberry-llm Open-Superintelligence-Lab/blueberry-llm Public

    Python 75 59

  4. Open-Superintelligence-Lab/blueberry-llm-t4-gpu Open-Superintelligence-Lab/blueberry-llm-t4-gpu Public

    Train best LLM possible on 1 x NVIDIA T4 (Google Colab, Kaggle,...) GPU

    Python 6 3