{"type": "meta", "benchmark_run_id": "91751afd-068a-477b-8f40-6e1963f803f1", "harness_version": "vps50-cpu-matrix-1", "started_at_utc": "2026-05-05T10:23:13Z", "host": "Slobodans-MacBook-Air.local", "node": "vps50", "engine": "llamacpp", "predator_target_url": "http://81.17.99.14:11436", "node_specs": {"cpu": "AMD EPYC 16 vCPUs", "ram_gb": 62, "free_ram_gb": 57, "gpu": "none"}, "purpose": "Sloba 2026-05-05 ask: bench VPS50 CPU box with the established llama.cpp config; primarily 30B-A3B + 70B-tier Qwen + bigger Gemma-4 + a 'Chinchilla smaller' (Phi-4 14B as proxy)", "models": ["phi-4", "gemma-4-26b-a4b", "qwen3-30b-a3b", "qwen2.5-72b"]}
{"type": "ram_snapshot", "model_key": "gemma-4-26b-a4b", "cell_id": "vps50:llamacpp:gemma-4-26b-a4b-it-q4km-cpu-ctx32k", "raw": "total        used        free      shared  buff/cache   available\nMem:           64301       20415       19503          37       25287       43885\nSwap:              0           0           0"}
{"type": "llama_bench", "model_key": "gemma-4-26b-a4b", "cell_id": "vps50:llamacpp:gemma-4-26b-a4b-it-q4km-cpu-ctx32k", "raw": "| model                          |       size |     params | backend    | threads |            test |                  t/s |\n| ------------------------------ | ---------: | ---------: | ---------- | ------: | --------------: | -------------------: |\n| gemma4 26B.A4B Q4_K - Medium   |  15.77 GiB |    25.23 B | CPU        |      16 |           pp256 |          0.86 \u00b1 0.13 |"}