AI on request

Benchmarks:

lama-server -hf unsloth/gemma-4-E4B-it-GGUF:UD-Q3_K_XL --reasoning off -fa on -ngl 99 -b 2048 -ub 2048 -c 4096 --temp 1.0 --top-p 0.95 --top-k 64 -> 130 t/s

llama-server -hf unsloth/gemma-4-E2B-it-GGUF:Q4_K_S --reasoning off -fa on -ngl 99 -b 2048 -ub 2048 -c 4096 --temp 1.0 --top-p 0.95 --top-k 64 -> 187 t/s

llama-server -hf unsloth/gemma-4-E2B-it-GGUF:Q3_K_S --reasoning off -fa on -ngl 99 -b 2048 -ub 2048 -c 4096 --temp 1.0 --top-p 0.95 --top-k 64 -> 186 t/s

llama-server -hf unsloth/gemma-4-E2B-it-GGUF:Q6_K --reasoning off -fa on -ngl 99 -b 2048 -ub 2048 -c 4096 --temp 1.0 --top-p 0.95 --top-k 64 -> 160 t/s

llama-server -hf unsloth/gemma-4-E2B-it-GGUF:Q5_K_S --reasoning off -fa on -ngl 99 -b 2048 -ub 2048 -c 4096 --temp 1.0 --top-p 0.95 --top-k 64 -> 177 t/s

llama-server -hf unsloth/gemma-4-E2B-it-GGUF:Q5_K_S --reasoning off -fa on -ngl 99 -b 2048 -ub 2048 -c 4096 --temp 1.0 --top-p 0.95 --top-k 64 --no-mmap -t 4 -> 181 t/s

llama-server -hf unsloth/gemma-4-E2B-it-GGUF:Q4_K_S --reasoning off -fa on -ngl 99 -b 2048 -ub 2048 -c 4096 --temp 1.0 --top-p 0.95 --top-k 64 --no-mmap -t -> 194 t/s

llama-server -hf unsloth/gemma-4-E4B-it-GGUF:Q3_K_XL --reasoning off -fa on -ngl 99 -b 2048 -ub 2048 -c 4096 --temp 1.0 --top-p 0.95 --top-k 64 --no-mmap -t 4 -> 126 t/s