﻿[strata] 2026-10-06 17:11:38 engine started: F:\Strata\engine\strata.exe --serve --pack F:\Strata-data\packs\sc117-q2_0 --native F:\Strata-data\models\sc117-q2_0\Qwen3.8-Flash-Next-GSQ-RCO-abliterated-Q2_0-00001-of-00002.gguf --ple-gguf F:\Strata-data\models\sc117-q2_0\Qwen3.8-Flash-Next-GSQ-RCO-abliterated-Q2_0-00002-of-00002.gguf --expert-profile F:\Strata\data\expert-profile.bin --expert-cache auto --prefill auto --spec 4 --spec-min-p 0.5 --mtp F:\Strata-data\mtp\rt --max-context 65536 --kv int8 --kv-resident 32768 --resident-experts --vision
strata generate: PCIe probe: 7.0 GB/s host->device (best of 7.0 6.9 6.9 7.0) -> pcie_frac 0.19 (default 0.55)
strata generate: this CPU has no AVX-512: the expert kernels run on AVX-2 (multi-token for the i-quant gate/up rows)
strata generate: native pack: F:\Strata-data\packs\sc117-q2_0 experts (largest blob 1.38 MB), token embedding Q3_K in mapped host memory (260 MiB, 0.3 s)
strata generate: 1416 MiB of weights loaded from F:\Strata-data\packs\sc117-q2_0 in 1.4 s (303 canonical tensors skipped: served natively)
strata generate: 301 native projection matrices, 1755.42 MiB of weights, in 3.4 s
strata generate: the SSD is kept awake while rows are read: one page of the table after 100 ms without a read, until 60 s after the last request (STRATA_SSD_KEEPALIVE=0 turns it off)
strata generate: profile F:\Strata\data\expert-profile.bin: 24576 ranked pairs, built for 24576 slots
strata generate: KV streaming: 32768 of 65536 cells per QSA layer in VRAM, the K/V in 0.77 GiB of pinned RAM
strata generate: PLE on, table 320001536 rows of F:\Strata-data\models\sc117-q2_0\Qwen3.8-Flash-Next-GSQ-RCO-abliterated-Q2_0-00002-of-00002.gguf
strata mtp: draft layer loaded, 835 MiB of VRAM (experts 675, dense 111), files read in 0.83 s (942 MiB/s)
strata generate: experimental native Q5_K head, 437043200 bytes, in 0.6 s
strata generate: GPU 0: AMD Radeon RX 6800M (gfx1031)
strata generate: HIP runtime F:\Strata\engine\amdhip64_7.dll
strata generate: experts via mmap (--mmap-experts; the GGUF shards in place, no experts.bin)
strata generate: 7 pool workers on logical processors 2,4,6,8,10,12,14, the host thread on 0 (draining too) (--host-core first)
strata: Windows budgets 11474 of this card's 12272 MiB for this process; free VRAM is counted within that (STRATA_WDDM_BUDGET=0: off)
strata generate: expert cache auto: 6.05 GiB free, 700 MiB reserved (+184 MiB for the draft head) -> 4026 slots
strata prefill: HIP device 0 gfx1031 - tip: STRATA_HIP_PROMPT_F16=1 reads long prompts about 2x faster here (FP16 prompt GEMMs, rocBLAS is tuned only for FP16 on gfx103x; the numbers differ slightly, see docs/AMD_HIP.md)
strata generate: expert cache 4026 slots, 5.18 GiB of VRAM; policy is
strata generate: the GPU computes the experts in the cache; it rounds differently from the CPU,
                 so a reply can differ slightly from a run without the cache (same quality:
                 bench/results/2026-09-27-cache-parity).
                 PROFILE, ranked by routing frequency, no eviction.
strata generate: the profile fill asks for 0 pairs ahead (STRATA_FILL_AHEAD)
strata generate: pre-filled 4026 of 4026 slots from the profile in 18.9 s (294 MB/s); slot 0 verified
strata generate: routing-aware prefetch of the file tier on (the next layer's router)
