bitmamba-zen-distillix-300m
This is a 1.58-bit BitMamba (Zen) model trained via Superposition Distillation.
Model Details
- Teacher: Qwen/Qwen2.5-Coder-1.5B-Instruct
- Architecture: Mamba SSM (Zen Variant) + BitLinear (1.58-bit weights)
- Params: ~278M
- Training Data: CodeAlpaca (Distilled)
Usage
This model requires the custom BitMambaStudent class definition to run.
It was trained as a Proof-of-Concept for Superposition Distillation.