3090
2 records across 2 models (quants: EXL2 2.5bpw, EXL2 4bpw).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 3090 | 70B 70B | EXL2 2.5bpw | ExLlamaV2 | 20 | – | – | – | – | 2023-10 | community | – | source |
| 3090 | Llama 2 7B 7B | EXL2 4bpw | ExLlamaV2 | 80 | – | – | – | – | 2023-10 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085