π BananaAI Developer 1
A 1.3B parameter coding model trained completely from scratch. ?
- No pretrained weights - initialized from random
- No fine-tuning - trained from scratch on raw data
- No APIs - pure open-source training
- Full transparency - all code, data, and weights public
Model Details
- Parameters: 1.3B or less (scaling to 3B)
- Architecture: GPT-NeoX style decoder
- Context Length: 4096 tokens
- Training Data: 30B+ tokens
- The Stack (code)
- OpenWebMath
- Wikipedia
- Reasoning chains
Training Progress
Status: Preparing for training
Start Date: TBD (awaiting compute allocation)
Training Logs: Coming soon
(I really hope this gets approved and to build my own AI...)
Inference Providers NEW
This model isn't deployed by any Inference Provider. π Ask for provider support