cfg-0108
backfilled
citable URL: https://halobench.com/records/cfg-0108/ — this address never moves; the anchor /records/#cfg-0108 keeps resolving
| model | NVIDIA-Nemotron-3-Super-120B-A12B-UD-Q4_K_M-00001-of-00003.gguf · UD-Q4_K_M |
| engine | ggml-org/llama.cpp 3653e6d6d · vulkan · host aihydra (igpu) |
| flags | -ngl 999 -fa 1 -b 2048 -ub 512 -ctk f16 -ctv f16 -t 16 --load-mode none |
| template | not recorded at test time |
| tree | upstream — stock |
aged evidence — reconstructed from the archive. Part 2 deep-cell session (2026-08-17), same fingerprint as cfg-0088 (identical llama-bench invocation: build 3653e6d6d, Vulkan, UD-Q4_K_M, f16 KV, -fa 1, --load-mode none -- only depth differs). d131072 attempt: SURVIVED cleanly on the first of a 2-attempt cap, no vk::DeviceLostError, no GPU wedge -- notable given this box's #25664 device-loss dataset at comparable depth on other candidates (laguna-s-21, deepseek-v4-flash). A fresh config id is minted per house convention. Ingested from llama-bench, which measures throughput and not footprint.
Capability basis
unmeasured — performance numbers on this config stand on a guard alone, not an established capability
Runs on this config (4)
| run | date | kind | suite | metrics |
|---|---|---|---|---|
| run-0389 | 2026-08-17 | performance | llama-bench@1 | prefill_tps=148.617391 |
| run-0390 | 2026-08-17 | performance | llama-bench@1 | decode_tps=16.622963 |
| run-0492 | 2026-08-19 | performance | llama-bench@1 | prefill_tps=130.812678 |
| run-0493 | 2026-08-19 | performance | llama-bench@1 | decode_tps=15.883158 |
Cited by — computed at build time, never stored
model pages nemotron3-super