T4
4 records across 1 models (quants: Q4_0).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| T4 | llama 7B | Q4_0 | llama.cpp (CUDA) | 46.38 | 1219.06 | – | – | tg128 | 2025-08-14 | community | – | source |
| T4 | llama 7B | Q4_0 | llama.cpp (CUDA) | 44.03 | 1309.73 | – | – | tg128 | 2025-08-14 | community | – | source |
| T4 | llama 7B | Q4_0 | llama.cpp (Vulkan) | 34.66 | 734.28 | – | – | tg128 | 2025-07-12 | community | – | source |
| T4 | llama 7B | Q4_0 | llama.cpp (Vulkan) | 33.36 | 708.19 | – | – | tg128 | 2025-07-12 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085