strata generate: layer split across 2 GPUs: CUDA0, then CUDA1 (split 25)
strata generate: PCIe probe: 7.0 GB/s host->device (best of 7.0 7.0 7.0 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: <data root>/packs/qwen-iq2_xs experts (largest blob 1.51 MB), token embedding IQ4_XS in mapped host memory (322 MiB)
strata generate: layer split: CUDA0 loads the dense weights of layers 0-24 only
strata generate: 799 MiB of weights loaded from <data root>/packs/qwen-iq2_xs (670 canonical tensors skipped: served natively)
strata generate: 156 native projection matrices, 766.85 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: layer split: CUDA1 loads the dense weights of layers 25-47 only
strata generate: layer split: CUDA1 PCIe probe 7.0 GB/s (best of 7.0 7.0 7.0 7.0) -> pcie_frac 0.19
strata generate: layer split: CUDA1 holds its weights; 22.36 GiB free (its session follows the split search)
strata generate: profile data/expert-profile.bin: 24576 ranked pairs, built for 24576 slots
strata generate: PLE on, table 320001536 rows of <data root>/models/qwen-iq2_xs/Qwen3.8-Flash-Next-GSQ-RCO-IQ2_XS-00002-of-00002.gguf
strata generate: layer split: CUDA1 holds its weights, session [25, 48) and the head; 21.93 GiB free
strata mtp: draft layer loaded, 802 MiB of VRAM (experts 675, dense 111), files read in 0.75 s (1053 MiB/s)
strata generate: GPU 0: Intel(R) Arc(TM) Pro B60 Graphics, compute capability 20.1
strata generate: WARNING: this engine was compiled with CUDA 12.8 headers but loaded a CUDA 0.2 runtime (libcudart): GPU properties can read wrong and some kernels go unused. Rebuild it against one toolkit (cmake -DCUDAToolkit_ROOT=<the toolkit>, with its libcudart.so present) (#542)
strata generate: experts streamed from the GGUF on demand (--stream-experts): no host arena; every expert must fit the VRAM cache
strata generate: expert cache auto: 22.11 GiB free, 1536 MiB reserved (+0 MiB for the draft head) -> 12800 slots
strata generate: expert cache 12800 slots, 16.91 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: profile fill: 16.91 GiB in 3.8 s (4.81 GB/s, pipelined)
strata generate: pre-filled 12800 of 12800 slots from the profile; slot 0 verified
strata generate: 4473 of 11776 experts missing from VRAM mirrored in pinned host memory (6.00 GiB, 1.1 s); the GPU reads them over PCIe
strata generate: WARNING: 7303 experts are neither in VRAM nor mirrored; with STRATA_VERIFY_NO_HOST the device plan cannot run them and their layers' windows fall back slowly - raise STRATA_MIRROR_MIB or the free RAM, or lower --max-context
strata generate: layer split, CUDA1: 21.14 GiB free of 23.91, room for experts 17.93 GiB
strata generate: layer split: CUDA1 runs layers 25-47, expert cache 11776 slots (16.11 GiB), 11776 of its 11776 profiled pairs; slot 0 verified
strata generate: layer split: CUDA0 runs layers 0-24
strata generate: R4 hit path ON - resident experts are computed on the GPU
strata generate: 5 expert-pool workers + the host thread
strata generate: session is up (engine 0.1.39-sycl)
strata generate: token graph hit path: 24576 resident experts, decided on the device
strata serve: layer split: 100% of the experts resident, the prompt path's streamed ring 96 slots
strata serve: the prompt path allocates its own buffers (too few cache slots to borrow)
strata verify: window up to 6 tokens, 56.4 MiB of device buffers
strata serve: layer split: layers 0-24 (CUDA0), 25-47 (CUDA1), one hand-off per window
strata verify: window up to 6 tokens, 56.4 MiB of device buffers
strata mtp: draft head over 106299 tokens (137.9 MiB)
strata serve: 4235 MiB of VRAM free with everything loaded
strata verify: captured the 6-token window (upload <FIXME: Placeholder>, sync <FIXME: Placeholder>)
strata verify: captured the 6-token window (upload <FIXME: Placeholder>, sync <FIXME: Placeholder>)
strata verify: captured the 4-token window (upload <FIXME: Placeholder>, sync <FIXME: Placeholder>)
strata verify: captured the 4-token window (upload <FIXME: Placeholder>, sync <FIXME: Placeholder>)
strata verify: captured the 1-token window (upload <FIXME: Placeholder>, sync <FIXME: Placeholder>)
strata verify: captured the 1-token window (upload <FIXME: Placeholder>, sync <FIXME: Placeholder>)
strata verify: captured the 3-token window (upload <FIXME: Placeholder>, sync <FIXME: Placeholder>)
strata verify: captured the 3-token window (upload <FIXME: Placeholder>, sync <FIXME: Placeholder>)
strata serve: prompt 23 tokens = 0 reused + 23 read in 456 ms (50.4 tok/s), 112 generated in 1910 ms (58.6 tok/s), drafts accepted 84 of 88, 1 checkpoints
strata serve: prompt 23 tokens = 16 reused + 7 read in 86 ms (81.7 tok/s), 112 generated in 1827 ms (61.3 tok/s), drafts accepted 84 of 89, 1 checkpoints
strata serve: suffix drafts: 1 windows, 1 of 3 drafts accepted
strata serve: prompt 37 tokens = 0 reused + 37 read in 505 ms (73.3 tok/s), 32 generated in 518 ms (61.8 tok/s), drafts accepted 23 of 24, 1 checkpoints
strata verify: captured the 5-token window (upload <FIXME: Placeholder>, sync <FIXME: Placeholder>)
strata verify: captured the 5-token window (upload <FIXME: Placeholder>, sync <FIXME: Placeholder>)
strata serve: prompt 42 tokens = 0 reused + 42 read in 591 ms (71.1 tok/s), 4 generated in 157 ms (25.4 tok/s), drafts accepted 4 of 6, 1 checkpoints
strata verify: captured the 2-token window (upload <FIXME: Placeholder>, sync <FIXME: Placeholder>)
strata verify: captured the 2-token window (upload <FIXME: Placeholder>, sync <FIXME: Placeholder>)
strata serve: prompt 2129 tokens = 0 reused + 2129 read in 4878 ms (436.4 tok/s), 104 generated in 2680 ms (38.8 tok/s), drafts accepted 56 of 87, 1 checkpoints
strata serve: prompt 2129 tokens = 2122 reused + 7 read in 88 ms (79.2 tok/s), 104 generated in 2602 ms (40.0 tok/s), drafts accepted 56 of 87, 1 checkpoints
strata serve: prompt 1922 tokens = 0 reused + 1922 read in 4318 ms (445.1 tok/s), 256 generated in 4616 ms (55.5 tok/s), drafts accepted 183 of 213, 1 checkpoints
strata serve: suffix drafts: 1 windows, 1 of 5 drafts accepted
strata serve: prompt 2199 tokens = 1915 reused + 284 read in 1604 ms (177.0 tok/s), 256 generated in 3532 ms (72.5 tok/s), drafts accepted 211 of 217, 2 checkpoints
strata serve: suffix drafts: 41 windows, 205 of 205 drafts accepted
strata serve: prompt 30 tokens = 0 reused + 30 read in 454 ms (66.1 tok/s), 11 generated in 255 ms (43.2 tok/s), drafts accepted 7 of 9, 1 checkpoints
strata serve: prompt 70 tokens = 41 reused + 29 read in 412 ms (70.3 tok/s), 3 generated in 97 ms (30.9 tok/s), drafts accepted 3 of 3, 2 checkpoints
strata serve: prompt 108 tokens = 74 reused + 34 read in 479 ms (70.9 tok/s), 14 generated in 307 ms (45.6 tok/s), drafts accepted 8 of 11, 3 checkpoints
strata serve: prompt 33 tokens = 0 reused + 33 read in 487 ms (67.8 tok/s), 39 generated in 813 ms (47.9 tok/s), drafts accepted 28 of 38, 1 checkpoints
