Filename	Quant Type	Description
`gemma-4-E4B-it-obliterated-Q3_K_M.gguf`	Q3_K_M	Very small, high quality loss. Good for extreme low-VRAM scenarios.
`gemma-4-E4B-it-obliterated-Q4_0.gguf`	Q4_0	Legacy format. Fast, but generally superseded by K-quants.
`gemma-4-E4B-it-obliterated-Q4_K_M.gguf`	Q4_K_M	Recommended. Excellent balance of size, speed, and minimal quality loss.
`gemma-4-E4B-it-obliterated-Q5_0.gguf`	Q5_0	Legacy format. Slightly higher quality and larger than Q4_0.
`gemma-4-E4B-it-obliterated-Q5_K_M.gguf`	Q5_K_M	High quality. Recommended if you have enough RAM/VRAM to spare over Q4_K_M.
`gemma-4-E4B-it-obliterated-Q6_K.gguf`	Q6_K	Very high quality. Near-perfect recreation of the original unquantized model.
`gemma-4-E4B-it-obliterated-Q8_0.gguf`	Q8_0	Extremely high quality. Almost indistinguishable from fp16, but requires significant memory.

How to Use with llama.cpp

Once you have downloaded llama.cpp and compiled it, you can run this model via the command line.

Basic CLI usage:

./llama-cli -m gemma-4-E4B-it-obliterated-Q4_K_M.gguf -p "Your prompt goes here" -n 512

Abiray/gemma-4-E4B-it-OBLITERATED-GGUF