gokashi/qwen3.5-4b-lite-jax
This model is a LiteRT (TFLite) export of a Qwen 3.5 Hybrid model.
Architecture
- Hybrid Layout: 3x Gated DeltaNet layers + 1x Gated Attention layer.
- Vocab Size: 248,320
- Target: Mobile/Edge Inference via LiteRT.
Usage
Run via TensorFlow Lite Interpreter with SELECT_TF_OPS enabled.
