This repository contains the early documentation and development progress for JumpLander Coder 32B, a high-performance code generation model currently hosted privately on Hugging Face.
Hugging Face model page (private):
➡️ https://huggingface.co/jumplander
The model will be publicly available after final tuning and evaluation.
JumpLander Coder 32B is designed as a next-generation coding assistant with:
- High-accuracy code generation
- Deep reasoning across multi-file projects
- Strong debugging and refactoring abilities
- Multilingual support (English + Persian)
- Optimized for developers in Iran
- Local/offline deployment support
This model targets reliability, performance, and real-world coding workflows at scale.
| Component | Status |
|---|---|
| Architecture | ✔️ Complete |
| Tokenizer | 🟡 In progress |
| Code datasets | ✔️ Ready |
| Pretraining | 🔜 Starting soon |
| Instruction tuning | 🔜 Planned |
| HuggingFace release | 🔒 Private (not public yet) |
- Release tokenizer preview
- Publish public architecture notes
- Upload early private checkpoints
- Add Persian coding datasets
- Benchmark vs DeepSeek, Qwen, Llama
- Public release of 32B model
- Website: https://jumplander.com
- Hugging Face page: https://huggingface.co/jumplander
- GitHub releases: coming soon
The project is currently under a proprietary license.
A suitable open-source license will be provided upon public release.
Contributions will open after the first public preview release.
Until then, issues and suggestions are welcome.