strata generate: PCIe probe: 26.3 GB/s host->device (best of 26.3 26.3 26.3 26.3) -> 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: /workshop/bench-flashnext/packs-6f32ec07/iq3_s experts (largest blob 2.66 MB), token embedding IQ4_XS in mapped host memory (322 MiB)
strata generate: 1466 MiB of weights loaded from /workshop/bench-flashnext/packs-6f32ec07/iq3_s (302 canonical tensors skipped: served natively)
strata generate: 300 native projection matrices, 2018.88 MiB of weights
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 /workshop/bench-flashnext/strata-6f32ec07/data/expert-profile.bin: 24576 ranked pairs, built for 24576 slots
strata generate: PLE on, table 320001536 rows of /workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf
strata mtp: draft layer loaded, 950 MiB of VRAM (experts 675, dense 111), files read in 0.33 s (2360 MiB/s)
strata generate: experimental native Q5_K head, 521472000 bytes
strata generate: GPU 0: NVIDIA GeForce RTX 3090 Ti, compute capability 8.6
strata generate: expert arena read through the file cache (buffered (not Windows))
strata generate: expert arena: cudaHostRegister PORTABLE ok; large pages skipped (STRATA_NO_LARGEPAGES); using 4 KB pages
strata generate: loaded 46.84 GiB at 4.87 GiB/s
strata generate: expert cache auto: 16.22 GiB free, 700 MiB reserved (+86 MiB for the draft head) -> 6229 slots
strata generate: expert cache 8144 slots, 15.44 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 8144 of 8144 slots from the profile; slot 0 verified
strata generate: R4 hit path ON - resident experts are computed on the GPU
strata generate: 31 expert-pool workers + the host thread
strata generate: session is up (engine 0.1.39)
strata generate: token graph hit path: 8144 resident experts, decided on the device
strata serve: prompt chunk auto: 8192 tokens, a 384-slot ring
strata serve: the prompt path borrows 2180 CUDA0 cache slots (4.12 GiB)
strata hc: CUDA0: the hyper-connection read runs as staged (the norm per token and stream, the down projection's activations staged ahead by cp.async); checked bit for bit against the plain read on this card (STRATA_HC_SPLIT=1 or 0 for the earlier ones)
strata verify: window up to 6 tokens, 74.7 MiB of device buffers
strata mtp: draft head over 40525 tokens (81.2 MiB)
strata serve: 431 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 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 2156 ms (1872.9 tok/s), 256 generated in 2927 ms (87.4 tok/s), drafts accepted 163 of 220, 1 checkpoints
strata serve: decode expert cache hit rate: 87.4% (120532 hits / 137987 lookups); 12253 more read by the GPU over PCIe or from another GPU (8.2% of all 150240 routed)
strata serve: suffix drafts: 2 windows, 7 of 10 drafts accepted
strata serve: prompt 4038 tokens = 0 reused + 4038 read in 1866 ms (2163.8 tok/s), 256 generated in 2597 ms (98.6 tok/s), drafts accepted 161 of 216, 1 checkpoints
strata serve: decode expert cache hit rate: 92.1% (131689 hits / 142967 lookups); 7753 more read by the GPU over PCIe or from another GPU (5.1% of all 150720 routed)
strata serve: prompt 4038 tokens = 0 reused + 4038 read in 1884 ms (2143.1 tok/s), 256 generated in 2610 ms (98.1 tok/s), drafts accepted 157 of 220, 1 checkpoints
strata serve: decode expert cache hit rate: 92.7% (136258 hits / 146976 lookups); 7104 more read by the GPU over PCIe or from another GPU (4.6% of all 154080 routed)
strata serve: suffix drafts: 1 windows, 2 of 3 drafts accepted
strata serve: prompt 32038 tokens = 0 reused + 32038 read in 11992 ms (2671.6 tok/s), 256 generated in 2771 ms (92.4 tok/s), drafts accepted 151 of 220, 2 checkpoints
strata serve: decode expert cache hit rate: 92.2% (136712 hits / 148292 lookups); 7708 more read by the GPU over PCIe or from another GPU (4.9% of all 156000 routed)
strata serve: suffix drafts: 2 windows, 1 of 5 drafts accepted
strata serve: prompt 32038 tokens = 0 reused + 32038 read in 11963 ms (2678.0 tok/s), 256 generated in 2720 ms (94.1 tok/s), drafts accepted 150 of 217, 2 checkpoints
strata serve: decode expert cache hit rate: 92.5% (137004 hits / 148044 lookups); 6996 more read by the GPU over PCIe or from another GPU (4.5% of all 155040 routed)
strata serve: prompt 32038 tokens = 0 reused + 32038 read in 12044 ms (2660.0 tok/s), 256 generated in 2844 ms (90.0 tok/s), drafts accepted 165 of 225, 2 checkpoints
strata serve: decode expert cache hit rate: 89.6% (127730 hits / 142531 lookups); 10589 more read by the GPU over PCIe or from another GPU (6.9% of all 153120 routed)
strata serve: suffix drafts: 1 windows, 1 of 2 drafts accepted
strata serve: prompt 128037 tokens = 0 reused + 128037 read in 50197 ms (2550.7 tok/s), 256 generated in 3212 ms (79.7 tok/s), drafts accepted 138 of 236, 6 checkpoints
strata serve: decode expert cache hit rate: 92.2% (148729 hits / 161294 lookups); 8626 more read by the GPU over PCIe or from another GPU (5.1% of all 169920 routed)
strata serve: suffix drafts: 3 windows, 2 of 6 drafts accepted
strata serve: prompt 128037 tokens = 0 reused + 128037 read in 50559 ms (2532.4 tok/s), 256 generated in 3147 ms (81.4 tok/s), drafts accepted 144 of 239, 6 checkpoints
strata serve: decode expert cache hit rate: 92.5% (148721 hits / 160740 lookups); 8220 more read by the GPU over PCIe or from another GPU (4.9% of all 168960 routed)
strata serve: suffix drafts: 2 windows, 4 of 4 drafts accepted
strata serve: prompt 128037 tokens = 0 reused + 128037 read in 50508 ms (2535.0 tok/s), 256 generated in 2875 ms (89.0 tok/s), drafts accepted 152 of 228, 6 checkpoints
strata serve: decode expert cache hit rate: 92.8% (141286 hits / 152211 lookups); 7149 more read by the GPU over PCIe or from another GPU (4.5% of all 159360 routed)
strata serve: suffix drafts: 3 windows, 4 of 7 drafts accepted