TP4
6 records across 1 models (quants: BF16).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| TP4 | GLM-5.3 | BF16 | vLLM | 870.4 | – | – | – | tg1000 | 2026-09-26 | community | – | source |
| TP4 | GLM-5.3 | BF16 | vLLM | 826.2 | – | – | – | tg1000 | 2026-09-26 | community | – | source |
| TP4 | GLM-5.3 | BF16 | vLLM | 412.4 | – | – | – | tg1000 | 2026-09-26 | community | – | source |
| TP4 | GLM-5.3 | BF16 | vLLM | 398.5 | – | – | – | tg1000 | 2026-09-26 | community | – | source |
| TP4 | GLM-5.3 | BF16 | vLLM | 176.8 | – | – | – | tg1000 | 2026-09-26 | community | – | source |
| TP4 | GLM-5.3 | BF16 | vLLM | 175.4 | – | – | – | tg1000 | 2026-09-26 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085