Marco-Mini-Instruct
9 records across 1 hardware.
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| RTX 3090 | Marco-Mini-Instruct | Q4_K_M | llama.cpp (CUDA) | 316 | 6966 | – | – | tg128 | 2026-10-06 | community | – | source |
| RTX 3090 | Marco-Mini-Instruct | Q4_K_M | llama.cpp (CUDA) | 287 | 9335 | – | – | tg128 | 2026-10-06 | community | – | source |
| RTX 3090 | Marco-Mini-Instruct | Q4_K_M | llama.cpp (CUDA) | 281 | 6482 | – | – | tg128 | 2026-10-06 | community | – | source |
| RTX 3090 | Marco-Mini-Instruct | Q4_K_M | llama.cpp (CUDA) | 278 | 7661 | – | – | tg128 | 2026-10-06 | community | – | source |
| RTX 3090 | Marco-Mini-Instruct | Q4_K_M | llama.cpp (CUDA) | 278 | 6970 | – | – | tg128 | 2026-10-06 | community | – | source |
| RTX 3090 | Marco-Mini-Instruct | Q4_K_M | llama.cpp (CUDA) | 278 | 9713 | – | – | tg128 | 2026-10-06 | community | – | source |
| RTX 3090 | Marco-Mini-Instruct | Q4_K_M | llama.cpp (CUDA) | 277 | 6005 | – | – | tg128 | 2026-10-06 | community | – | source |
| RTX 3090 | Marco-Mini-Instruct | Q4_K_M | llama.cpp (CUDA) | 256 | 6806 | – | – | tg128 | 2026-10-06 | community | – | source |
| RTX 3090 | Marco-Mini-Instruct | Q4_K_M | llama.cpp (CUDA) | 155 | 5160 | – | – | tg128 | 2026-10-06 | community | – | source |