cfg-0116
fork
citable URL: https://halobench.com/records/cfg-0116/ — this address never moves; the anchor /records/#cfg-0116 keeps resolving
| model | DeepSeek-V4-Flash-0731-UD-IQ3_XXS-00001-of-00004.gguf · UD-IQ3_XXS |
| engine | ggml-org/llama.cpp baf6360be · vulkan · host aihydra (igpu) |
| flags | -ngl 999 -fa 1 -dev Vulkan0 --load-mode none -ctk f16 -ctv f16 -p 1024 -n 256 -b 2048 -ub 512 -r 1 |
| template | not recorded at test time |
| tree | fork — carrying con-0002 |
Capability basis
unmeasured — performance numbers on this config stand on a guard alone, not an established capability
Runs on this config (10)
| run | date | kind | suite | metrics |
|---|---|---|---|---|
| run-0441 | 2026-08-18 | performance | llama-bench@1 | prefill_tps=219.360495 |
| run-0442 | 2026-08-18 | performance | llama-bench@1 | decode_tps=19.543966 |
| run-0443 | 2026-08-18 | performance | llama-bench@1 | prefill_tps=181.554246 |
| run-0444 | 2026-08-18 | performance | llama-bench@1 | decode_tps=17.828501 |
| run-0445 | 2026-08-18 | performance | llama-bench@1 | prefill_tps=142.563914 |
| run-0446 | 2026-08-18 | performance | llama-bench@1 | decode_tps=15.412875 |
| run-0447 | 2026-08-18 | performance | llama-bench@1 | prefill_tps=126.504736 |
| run-0448 | 2026-08-18 | performance | llama-bench@1 | decode_tps=14.237001 |
| run-0449 | 2026-08-18 | performance | llama-bench@1 | prefill_tps=115.044597 |
| run-0450 | 2026-08-18 | performance | llama-bench@1 | decode_tps=13.316777 |
Cited by — computed at build time, never stored
model pages deepseek-v4-flash