Gemma 4 E2B it โ€” "lean" Q4_0 GGUF

Requant of gemma-4-E2B-it with every large tensor forced to Q4_0, including the tied token embedding / output head and the per-layer embedding table (the standard Q4_0 recipe leaves those at Q6_K, costing ~20% generation speed on memory-bandwidth-limited hardware).

  • 2.65 GB vs 3.36 GB standard Q4_0; ~20% faster token generation on iGPU/CPU systems.
  • Made with llama.cpp b9966 llama-quantize --allow-requantize --token-embedding-type q4_0 --output-tensor-type q4_0 from the official lmstudio-community Q4_K_M.
  • Usage subject to the Gemma Terms of Use.
Downloads last month
47
GGUF
Model size
5B params
Architecture
gemma4
Hardware compatibility
Log In to add your hardware

4-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support