아래 명령은 실제 benchmark runner가 사용한 llama-server flag 조합입니다. 공개 페이지에서는 로컬 모델 경로·host·port만 placeholder로 치환했습니다.
MXFP4 · non-MTP
Server: non-MTP · reasoning autollama-server -m <MODEL_FILE> --host <HOST> --port <PORT> --ctx-size 8192 --n-gpu-layers 999 --flash-attn on --batch-size 2048 --ubatch-size 512 --threads 20 --metrics --slots --parallel 8 --cache-type-k f16 --cache-type-v f16 --spec-type none
UD-Q4 · non-MTP
Server: non-MTP · reasoning autollama-server -m <MODEL_FILE> --host <HOST> --port <PORT> --ctx-size 8192 --n-gpu-layers 999 --flash-attn on --batch-size 2048 --ubatch-size 512 --threads 20 --metrics --slots --parallel 8 --cache-type-k f16 --cache-type-v f16 --spec-type none
UD-Q5 · non-MTP
Server: non-MTP · reasoning autollama-server -m <MODEL_FILE> --host <HOST> --port <PORT> --ctx-size 8192 --n-gpu-layers 999 --flash-attn on --batch-size 2048 --ubatch-size 512 --threads 20 --metrics --slots --parallel 8 --cache-type-k f16 --cache-type-v f16 --spec-type none
UD-Q6 · non-MTP
Server: non-MTP · reasoning autollama-server -m <MODEL_FILE> --host <HOST> --port <PORT> --ctx-size 8192 --n-gpu-layers 999 --flash-attn on --batch-size 2048 --ubatch-size 512 --threads 20 --metrics --slots --parallel 8 --cache-type-k f16 --cache-type-v f16 --spec-type none