DynaDFlash
A DFlash draft model with dynamic context projection from Qwen3-8B (block size 16). Use model.safetensors and router.safetensors together with the included serving adapter.
For more details and results, see the DynaDFlash repository.
A DFlash draft model with dynamic context projection from Qwen3-8B (block size 16). Use model.safetensors and router.safetensors together with the included serving adapter.
For more details and results, see the DynaDFlash repository.
model.safetensors 2.1 GBrouter.safetensors 82 kBz-lab/Qwen3-8B-DFlash-b16