clef-flash
17 records across 4 hardware.
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Quadro RTX 4000 | clef-flash | Q5_K_M-dyn | llama.cpp (CUDA) | 1100 | – | – | – | 16384 | – | community | – | source |
| M5 Max | clef-flash | 4bit | oMLX | 100.6 | 2471.0 | – | – | 1024 | 2026-10-07 | community | – | source |
| M5 Max | clef-flash | 4bit | oMLX | 99.4 | 2816.0 | – | – | 4096 | 2026-10-07 | community | – | source |
| M5 Max | clef-flash | 4bit | oMLX | 81.5 | 2586.0 | – | – | 1024 | 2026-10-15 | community | – | source |
| M5 Max | clef-flash | 4bit | oMLX | 81.0 | 2146.0 | – | – | 1024 | 2026-10-15 | community | – | source |
| M5 Max | clef-flash | 4bit | oMLX | 80.4 | 2474.0 | – | – | 4096 | 2026-10-15 | community | – | source |
| M5 Max | clef-flash | 4bit | oMLX | 79.6 | 2837.0 | – | – | 4096 | 2026-10-15 | community | – | source |
| M5 Max | clef-flash | 8bit | oMLX | 57.2 | 3148.0 | – | – | 4096 | 2026-10-06 | community | – | source |
| M5 Max | clef-flash | 8bit | oMLX | 57.2 | 3148.0 | – | – | 4096 | 2026-10-06 | community | – | source |
| M5 Max | clef-flash | 8bit | oMLX | 55.1 | 3140.0 | – | – | 16384 | 2026-10-06 | community | – | source |
| M5 Max | clef-flash | 8bit | oMLX | 55.1 | 3140.0 | – | – | 16384 | 2026-10-06 | community | – | source |
| M6 | clef-flash | 4bit | oMLX | 31.9 | 1039.0 | – | – | 1024 | 2026-10-10 | community | – | source |
| M6 | clef-flash | 4bit | oMLX | 31.2 | 1154.0 | – | – | 4096 | 2026-10-10 | community | – | source |
| M4 | clef-flash | 4bit | oMLX | 21.2 | 201.9 | – | – | 1024 | 2026-10-08 | community | – | source |
| M4 | clef-flash | 4bit | oMLX | 20.9 | 240.4 | – | – | 4096 | 2026-10-06 | community | – | source |
| M4 | clef-flash | 4bit | oMLX | 20.5 | 207.3 | – | – | 4096 | 2026-10-08 | community | – | source |
| M4 | clef-flash | 4bit | oMLX | 20.3 | 228.0 | – | – | 1024 | 2026-10-06 | community | – | source |