strata generate: PCIe probe: 28.7 GB/s host->device -> pcie_frac 0.55 (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: /data/strata-bench/packs/q2_0 experts (largest blob 1.38 MB), token embedding type 11 in mapped host memory (260 MiB)
strata generate: 1416 MiB of weights loaded from /data/strata-bench/packs/q2_0 (303 canonical tensors skipped: served natively)
strata generate: 301 native projection matrices, 1376.20 MiB of weights
strata generate: KV streaming: 32768 of 131072 cells per QSA layer in VRAM, the K/V in 1.55 GiB of pinned RAM
strata generate: PLE on, table 320001536 rows of /data/strata-bench/models/ISTA-ed59f920/Q2_0/Qwen3.8-Flash-Next-GSQ-RCO-Q2_0-00002-of-00002.gguf
strata mtp: draft layer loaded, 835 MiB of VRAM (experts 675, dense 111), files read in 0.43 s (1828 MiB/s)
strata generate: expert arena: cudaHostRegister PORTABLE ok; MAP_HUGETLB unavailable (no hugetlb pool configured?); using 4 KB pages
strata generate: loaded 31.64 GiB at 8.32 GiB/s
strata generate: profile /data/strata-bench/src/strata-8fc40dde/data/expert-profile.bin: 24576 ranked pairs, built for 24576 slots
strata generate: experimental native Q5_K head, 437043200 bytes
strata generate: expert cache auto: 18.19 GiB free, 700 MiB reserved -> 13595 slots
strata generate: expert cache 13595 slots, 17.50 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: pre-filled 13595 of 13595 slots from the profile; slot 0 verified
strata generate: R4 hit path ON - resident experts are computed on the GPU
strata generate: 23 expert-pool workers + the host thread
strata generate: session is up (engine 0.1.25)
strata generate: token graph hit path: 13595 resident experts, decided on the device
strata serve: prompt chunk auto: 8192 tokens
strata serve: the prompt path borrows 3171 cache slots (4.08 GiB)
strata verify: window up to 6 tokens, 55.2 MiB of device buffers
strata mtp: draft head over 40525 tokens (68.0 MiB)
strata serve: 261 MiB of VRAM free with everything loaded
strata verify: captured the 6-token window (upload no error, sync no error)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: captured the 4-token window (upload no error, sync no error)
strata verify: captured the 5-token window (upload no error, sync no error)
strata verify: captured the 2-token window (upload no error, sync no error)
strata verify: captured the 3-token window (upload no error, sync no error)
strata serve: prompt 4038 tokens = 0 reused + 4038 read in 1976 ms (2043.0 tok/s), 256 generated in 2145 ms (119.4 tok/s), drafts accepted 158 of 211, 1 checkpoints
strata serve: decode expert cache hit rate: 95.4% (138160 hits / 144814 lookups)
strata serve: KV streaming: 99.34% of 1941012 block reads hit VRAM, 51.5 MiB read from RAM
strata serve: suffix drafts: 4 windows, 9 of 13 drafts accepted
strata serve: prompt 4038 tokens = 0 reused + 4038 read in 1683 ms (2399.2 tok/s), 256 generated in 1907 ms (134.2 tok/s), drafts accepted 159 of 207, 1 checkpoints
strata serve: decode expert cache hit rate: 98.3% (142408 hits / 144817 lookups)
strata serve: KV streaming: 99.33% of 1910244 block reads hit VRAM, 51.6 MiB read from RAM
strata serve: suffix drafts: 6 windows, 12 of 20 drafts accepted
strata serve: prompt 4038 tokens = 0 reused + 4038 read in 1658 ms (2435.8 tok/s), 256 generated in 1949 ms (131.3 tok/s), drafts accepted 161 of 215, 1 checkpoints
strata serve: decode expert cache hit rate: 98.6% (146829 hits / 148904 lookups)
strata serve: KV streaming: 99.35% of 1959504 block reads hit VRAM, 51.5 MiB read from RAM
strata serve: suffix drafts: 4 windows, 8 of 16 drafts accepted
strata serve: prompt 32038 tokens = 0 reused + 32038 read in 11385 ms (2814.1 tok/s), 256 generated in 2237 ms (114.5 tok/s), drafts accepted 151 of 222, 2 checkpoints
strata serve: decode expert cache hit rate: 97.7% (151762 hits / 155348 lookups)
strata serve: KV streaming: 97.30% of 2051976 block reads hit VRAM, 223.6 MiB read from RAM
strata serve: prompt 32038 tokens = 0 reused + 32038 read in 11309 ms (2832.9 tok/s), 256 generated in 2225 ms (115.1 tok/s), drafts accepted 153 of 230, 2 checkpoints
strata serve: decode expert cache hit rate: 98.8% (157181 hits / 159142 lookups)
strata serve: KV streaming: 97.68% of 2088876 block reads hit VRAM, 195.2 MiB read from RAM
strata serve: suffix drafts: 2 windows, 6 of 10 drafts accepted
strata serve: prompt 32038 tokens = 0 reused + 32038 read in 11438 ms (2801.1 tok/s), 256 generated in 2122 ms (120.6 tok/s), drafts accepted 156 of 218, 2 checkpoints
strata serve: decode expert cache hit rate: 98.8% (150691 hits / 152456 lookups)
strata serve: KV streaming: 97.41% of 2002656 block reads hit VRAM, 208.7 MiB read from RAM
strata serve: suffix drafts: 1 windows, 2 of 3 drafts accepted
strata serve: prompt 128037 tokens = 0 reused + 128037 read in 44870 ms (2853.5 tok/s), 256 generated in 2410 ms (106.2 tok/s), drafts accepted 146 of 223, 6 checkpoints
strata serve: decode expert cache hit rate: 98.9% (158378 hits / 160191 lookups)
strata serve: KV streaming: 94.79% of 2101284 block reads hit VRAM, 441.0 MiB read from RAM
strata serve: suffix drafts: 1 windows, 0 of 5 drafts accepted
strata serve: prompt 128037 tokens = 0 reused + 128037 read in 45032 ms (2843.2 tok/s), 256 generated in 2503 ms (102.3 tok/s), drafts accepted 143 of 242, 6 checkpoints
strata serve: decode expert cache hit rate: 98.9% (167808 hits / 169710 lookups)
strata serve: KV streaming: 94.81% of 2224464 block reads hit VRAM, 465.4 MiB read from RAM
strata serve: suffix drafts: 2 windows, 3 of 10 drafts accepted
strata serve: prompt 128037 tokens = 0 reused + 128037 read in 46819 ms (2734.7 tok/s), 256 generated in 2548 ms (100.5 tok/s), drafts accepted 142 of 218, 6 checkpoints
strata serve: decode expert cache hit rate: 98.9% (157107 hits / 158797 lookups)
strata serve: KV streaming: 94.76% of 2082768 block reads hit VRAM, 439.8 MiB read from RAM
strata serve: suffix drafts: 2 windows, 2 of 7 drafts accepted