Source files / bench/data · raw
notes.json
{ "results": { "2026-11-06T21:5": "first bring-up; pp is wrong (prompt cache hit)", "2026-11-06T22:2": "first bring-up on d90ec36a3", "2026-11-06T22:4": "pin moved to 790a3eaa1", "2026-11-06T23:2": "fork head candidate, not pinned", "2026-11-07T15:5": "baseline after fan control", "2026-11-07T16:0": "power limit 125 W test (not adopted)", "2026-11-07T21:1": "LLAMA_SPEC_SAMPLE_TEMP=0 test (reverted)" }, "concurrency": { "base-np1": "production: 1 slot, MTP, 262k context", "np4-kvu-spec": "4 slots, unified KV, MTP, 262k (OOM before 4 parallel)", "np4-kvu-nospec": "4 slots, unified KV, no MTP, 262k", "np1-nospec-c131k": "1 slot, no MTP, 131k", "np4-kvu-spec-c131k": "4 slots, unified KV, MTP, 131k", "np4-kvu-nospec-c131k": "4 slots, unified KV, no MTP, 131k", "replay-stock": "pinned build, 4 slots", "swarm-stock": "pinned build, 4 slots", "replay-patched": "with server-prompt-cache-keep.patch, first version", "swarm-patched": "with server-prompt-cache-keep.patch, first version", "replay-patched2": "with server-prompt-cache-keep.patch (final)", "swarm-patched2": "with server-prompt-cache-keep.patch (final)", "strata-svc": "Strata (qr0n P100 build), Flash-Next IQ3_S, 131k" }, "lean_bench": { "junk-rig-qwen3.8-27b-ud-q6_k_seed1_standard.jsonl": [ "Qwen3.8 27B Q6_K", "Tersler, llama.cpp, q4_0 KV, 16k cap" ], "junk-rig-qwen3.8-27b-ud-q6_k_seed1_standard_32k.jsonl": [ "Qwen3.8 27B Q6_K", "Tersler, llama.cpp, q4_0 KV, 32k cap" ], "junk-rig-ornith-1.5-35b-a3b-q4_k_m_seed1_standard_32k.jsonl": [ "Ornith 1.5 35B-A3B Q4_K_M", "Tersler, llama.cpp, q8_0 KV, 32k cap" ], "qwen3.8-flash-next-iq3_s_seed1_standard.jsonl": [ "Qwen3.8 Flash-Next IQ3_S", "moppel (RTX 5060 Ti), Strata, 16k cap" ], "qwen3.8-flash-next-iq3_s_seed1_standard_32k.jsonl": [ "Qwen3.8 Flash-Next IQ3_S", "moppel (RTX 5060 Ti), Strata, 32k cap" ], "junk-rig-strata-flash-next-iq3_s_seed1_standard_32k.jsonl": [ "Qwen3.8 Flash-Next IQ3_S", "Tersler, Strata (qr0n P100 build), RAM at 2 channels, 32k cap" ] }, "speed": { "strata-qr0n-iq3s-2ch": "Strata (qr0n P100 build), Flash-Next IQ3_S, RAM at 2 channels" }, "tools": { "strata-svc": "Strata, Flash-Next IQ3_S" } }