cfg-0114
backfilled
citable URL: https://halobench.com/records/cfg-0114/ — this address never moves; the anchor /records/#cfg-0114 keeps resolving
| model | Ling-3.0-flash-Q4_K_M.gguf · Q4_K_M |
| engine | ggml-org/llama.cpp 7077abb · rocm · host aihydra (igpu) |
| flags | -ngl 999 -fa 1 -b 2048 -ub 512 -ctk f16 -ctv f16 -t 16 -lm none |
| template | not recorded at test time |
| tree | upstream — stock |
aged evidence — reconstructed from the archive. Ingested from llama-bench JSON, which measures throughput and not footprint: the memory block, chat-template hash and model revision are absent by construction. Peak memory comes from a Tier 0 capability probe, not from this record. The rows behind this config cover d0 and d32768, each as a single llama-bench invocation containing prefill and decode rows. Capability guard is recorded separately as run-0434 against the matching serving config cfg-0115.
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-0435 | 2026-08-18 | performance | llama-bench@50 | prefill_tps=409.548592 |
| run-0436 | 2026-08-18 | performance | llama-bench@50 | decode_tps=36.080858 |
| run-0437 | 2026-08-18 | performance | llama-bench@50 | prefill_tps=229.055626 |
| run-0438 | 2026-08-18 | performance | llama-bench@50 | decode_tps=32.886856 |
Cited by — computed at build time, never stored
model pages ling-30-flash