radelbamor

Iterate fast at nano-scale and scale what works. Architecting the future of efficient AI 17y+ Engineering. Systems × Generative AI.

[radelbamor @ramobledar]

https://t.co/jOfcHOe9H1

[radelbamor @ramobledar]

Screen-recorded mini eval of the Noeum-1-Nano (0.6B) AI/LLM on 28 random questions from a 10,000 https://t.co/qk1iX9h8Ip via @YouTube

[radelbamor @ramobledar]

Inspired by @karpathy, we built the LLMification engine. Static Math books → Dynamic LLM Data

Worked problems >SFT
Practice problems >RL environments
Infinite, verified variations
SymPy verification
Algebra to Olympic-level math
Micro-step process supervision
and much more ..

[radelbamor @ramobledar]

What efficient AI looks like when you start from first principles. https://t.co/E0woeUSWwo model & weights on Hugging Face.

[radelbamor @ramobledar]

@karpathy Inspired by this post! I have this running now: PDF → Markdown → Infinite Generator → Code Execution Env. It's evolved into a full engine at this point. Sharing the code in the next few days.for math it's working fine, but physics is another story. Complexity increases 100x

[radelbamor @ramobledar]

Noeum-1-Nano is live

Competitive AI model trained on just 18B tokens (vs 2-12T standard)

- 20-667× more data-efficient
- Matches major lab nano models
- Built entirely from scratch
- Solo research from Vienna

Model: https://t.co/zhK5dHe9li
#AI #EfficientAI #SovereignAI