Arko007/Zenyx-V2
0
⚡ Zenyx-V2
A custom language model built entirely from scratch — designed and trained by @Arko007 on TPU v5e-8 using JAX/Flax, now served on ZeroGPU.
🧠 Architecture — Nano-Titan
🗂️ Training
- 200B tokens of Math (45%) · Code (35%) · English (20%)
- Pretrained on FineMath, StarCoderData, FineWeb-Edu
- SFT fine-tuned on 50+ NVIDIA Nemotron / OpenMath / OpenCode datasets
⚠️ Note
This is my first published model, trained from scratch as a learning project. It's a very small model — expect limited reasoning but it's fully custom! 🚀
