
### confirmation/prefill32/warmup
strata verify: capturing the 6-token window (1027 MiB of VRAM free)
strata verify: captured the 6-token window (upload no error, sync no error)
strata verify: capturing the 1-token window (1023 MiB of VRAM free)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: capturing the 2-token window (1013 MiB of VRAM free)
strata verify: captured the 2-token window (upload no error, sync no error)
strata verify: capturing the 3-token window (1009 MiB of VRAM free)
strata verify: captured the 3-token window (upload no error, sync no error)
strata verify: capturing the 4-token window (1001 MiB of VRAM free)
strata verify: captured the 4-token window (upload no error, sync no error)
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 18860 ms (6949.7 tok/s), 64 generated in 485 ms (132.1 tok/s), drafts accepted 31 of 55, 4 checkpoints
strata serve: decode expert cache hit rate: 95.1% (39689 hits / 41721 lookups); 999 more read by the GPU over PCIe or from another GPU (2.3% of all 42720 routed)
strata serve: KV streaming: 91.03% of 585396 block reads hit VRAM, 211.5 MiB read from RAM
strata serve: suffix drafts: 1 windows, 1 of 3 drafts accepted

### confirmation/prefill32/perf-greedy-32768-r0
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 4636 ms (7068.9 tok/s), 512 generated in 2763 ms (185.3 tok/s), drafts accepted 314 of 436, 1 checkpoints
strata serve: decode expert cache hit rate: 95.5% (283651 hits / 296989 lookups); 7331 more read by the GPU over PCIe or from another GPU (2.4% of all 304320 routed)
strata serve: KV streaming: 98.30% of 3943716 block reads hit VRAM, 269.5 MiB read from RAM
strata serve: suffix drafts: 6 windows, 11 of 17 drafts accepted

### confirmation/prefill32/perf-sampled-32768-r0
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 4638 ms (7065.1 tok/s), 512 generated in 2799 ms (182.9 tok/s), drafts accepted 313 of 452, 1 checkpoints
strata serve: decode expert cache hit rate: 96.4% (296284 hits / 307271 lookups); 6169 more read by the GPU over PCIe or from another GPU (2.0% of all 313440 routed)
strata serve: KV streaming: 98.36% of 4048524 block reads hit VRAM, 267.0 MiB read from RAM
strata serve: suffix drafts: 9 windows, 21 of 27 drafts accepted

### confirmation/prefill32/perf-greedy-131072-r0
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 18874 ms (6944.7 tok/s), 512 generated in 2919 ms (175.4 tok/s), drafts accepted 297 of 455, 4 checkpoints
strata serve: decode expert cache hit rate: 96.8% (307680 hits / 317776 lookups); 4784 more read by the GPU over PCIe or from another GPU (1.5% of all 322560 routed)
strata serve: KV streaming: 96.69% of 4177932 block reads hit VRAM, 557.0 MiB read from RAM
strata serve: suffix drafts: 1 windows, 1 of 3 drafts accepted

### confirmation/prefill32/perf-sampled-131072-r0
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 18317 ms (7155.7 tok/s), 512 generated in 3205 ms (159.7 tok/s), drafts accepted 293 of 430, 4 checkpoints
strata serve: decode expert cache hit rate: 95.3% (290034 hits / 304219 lookups); 8261 more read by the GPU over PCIe or from another GPU (2.6% of all 312480 routed)
strata serve: KV streaming: 96.65% of 4036032 block reads hit VRAM, 544.9 MiB read from RAM
strata serve: suffix drafts: 17 windows, 56 of 63 drafts accepted

### confirmation/prefill32/perf-greedy-32768-r1
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 4654 ms (7040.2 tok/s), 512 generated in 2825 ms (181.2 tok/s), drafts accepted 307 of 425, 1 checkpoints
strata serve: decode expert cache hit rate: 96.7% (288710 hits / 298569 lookups); 4791 more read by the GPU over PCIe or from another GPU (1.6% of all 303360 routed)
strata serve: KV streaming: 98.32% of 3931416 block reads hit VRAM, 266.3 MiB read from RAM
strata serve: suffix drafts: 1 windows, 1 of 3 drafts accepted

### confirmation/prefill32/perf-sampled-32768-r1
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 4711 ms (6955.3 tok/s), 512 generated in 2948 ms (173.7 tok/s), drafts accepted 304 of 444, 1 checkpoints
strata serve: decode expert cache hit rate: 95.7% (291950 hits / 305040 lookups); 7920 more read by the GPU over PCIe or from another GPU (2.5% of all 312960 routed)
strata serve: KV streaming: 98.36% of 4042296 block reads hit VRAM, 267.3 MiB read from RAM
strata serve: suffix drafts: 4 windows, 4 of 12 drafts accepted

### confirmation/prefill32/perf-greedy-131072-r1
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 18781 ms (6979.0 tok/s), 512 generated in 3262 ms (157.0 tok/s), drafts accepted 290 of 455, 4 checkpoints
strata serve: decode expert cache hit rate: 96.8% (309731 hits / 320103 lookups); 4857 more read by the GPU over PCIe or from another GPU (1.5% of all 324960 routed)
strata serve: KV streaming: 96.82% of 4208664 block reads hit VRAM, 538.4 MiB read from RAM
strata serve: suffix drafts: 1 windows, 1 of 2 drafts accepted

### confirmation/prefill32/perf-sampled-131072-r1
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 17896 ms (7324.1 tok/s), 512 generated in 2802 ms (182.7 tok/s), drafts accepted 312 of 451, 4 checkpoints
strata serve: decode expert cache hit rate: 95.3% (289946 hits / 304219 lookups); 8741 more read by the GPU over PCIe or from another GPU (2.8% of all 312960 routed)
strata serve: KV streaming: 96.64% of 4042284 block reads hit VRAM, 547.4 MiB read from RAM
strata serve: suffix drafts: 12 windows, 44 of 48 drafts accepted

### confirmation/prefill16/warmup
strata verify: capturing the 6-token window (1028 MiB of VRAM free)
strata verify: captured the 6-token window (upload no error, sync no error)
strata verify: capturing the 1-token window (1024 MiB of VRAM free)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: capturing the 2-token window (1014 MiB of VRAM free)
strata verify: captured the 2-token window (upload no error, sync no error)
strata verify: capturing the 3-token window (1010 MiB of VRAM free)
strata verify: captured the 3-token window (upload no error, sync no error)
strata verify: capturing the 4-token window (1002 MiB of VRAM free)
strata verify: captured the 4-token window (upload no error, sync no error)
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 19198 ms (6827.4 tok/s), 64 generated in 418 ms (153.0 tok/s), drafts accepted 35 of 51, 6 checkpoints
strata serve: decode expert cache hit rate: 94.1% (35438 hits / 37670 lookups); 1210 more read by the GPU over PCIe or from another GPU (3.1% of all 38880 routed)
strata serve: KV streaming: 90.07% of 536124 block reads hit VRAM, 214.4 MiB read from RAM
strata serve: suffix drafts: 1 windows, 1 of 3 drafts accepted

### confirmation/prefill16/perf-greedy-32768-r0
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 4731 ms (6926.7 tok/s), 512 generated in 2741 ms (186.8 tok/s), drafts accepted 308 of 425, 2 checkpoints
strata serve: decode expert cache hit rate: 95.7% (283106 hits / 295683 lookups); 6717 more read by the GPU over PCIe or from another GPU (2.2% of all 302400 routed)
strata serve: KV streaming: 98.27% of 3919008 block reads hit VRAM, 272.6 MiB read from RAM
strata serve: suffix drafts: 5 windows, 12 of 15 drafts accepted

### confirmation/prefill16/perf-sampled-32768-r0
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 4736 ms (6918.8 tok/s), 512 generated in 2650 ms (193.2 tok/s), drafts accepted 328 of 428, 2 checkpoints
strata serve: decode expert cache hit rate: 95.8% (274876 hits / 287035 lookups); 7205 more read by the GPU over PCIe or from another GPU (2.4% of all 294240 routed)
strata serve: KV streaming: 98.28% of 3801996 block reads hit VRAM, 263.8 MiB read from RAM
strata serve: suffix drafts: 12 windows, 26 of 39 drafts accepted

### confirmation/prefill16/perf-greedy-131072-r0
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 19407 ms (6753.9 tok/s), 512 generated in 2873 ms (178.2 tok/s), drafts accepted 306 of 466, 6 checkpoints
strata serve: decode expert cache hit rate: 96.5% (306501 hits / 317551 lookups); 5489 more read by the GPU over PCIe or from another GPU (1.7% of all 323040 routed)
strata serve: KV streaming: 96.89% of 4184016 block reads hit VRAM, 524.5 MiB read from RAM
strata serve: suffix drafts: 1 windows, 1 of 3 drafts accepted

### confirmation/prefill16/perf-sampled-131072-r0
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 19718 ms (6647.4 tok/s), 512 generated in 2832 ms (180.8 tok/s), drafts accepted 334 of 432, 6 checkpoints
strata serve: decode expert cache hit rate: 95.5% (273184 hits / 285997 lookups); 7763 more read by the GPU over PCIe or from another GPU (2.6% of all 293760 routed)
strata serve: KV streaming: 96.34% of 3795828 block reads hit VRAM, 559.6 MiB read from RAM
strata serve: suffix drafts: 12 windows, 47 of 52 drafts accepted

### confirmation/prefill16/perf-greedy-32768-r1
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 4701 ms (6970.3 tok/s), 512 generated in 2778 ms (184.3 tok/s), drafts accepted 298 of 458, 2 checkpoints
strata serve: decode expert cache hit rate: 96.7% (307461 hits / 317914 lookups); 5126 more read by the GPU over PCIe or from another GPU (1.6% of all 323040 routed)
strata serve: KV streaming: 98.46% of 4184052 block reads hit VRAM, 260.0 MiB read from RAM
strata serve: suffix drafts: 1 windows, 2 of 3 drafts accepted

### confirmation/prefill16/perf-sampled-32768-r1
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 4538 ms (7220.9 tok/s), 512 generated in 2836 ms (180.5 tok/s), drafts accepted 295 of 402, 2 checkpoints
strata serve: decode expert cache hit rate: 95.8% (279026 hits / 291400 lookups); 7160 more read by the GPU over PCIe or from another GPU (2.4% of all 298560 routed)
strata serve: KV streaming: 98.17% of 3857400 block reads hit VRAM, 284.5 MiB read from RAM
strata serve: suffix drafts: 10 windows, 18 of 28 drafts accepted

### confirmation/prefill16/perf-greedy-131072-r1
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 18297 ms (7163.6 tok/s), 512 generated in 2672 ms (191.6 tok/s), drafts accepted 314 of 425, 6 checkpoints
strata serve: decode expert cache hit rate: 96.9% (285853 hits / 295099 lookups); 4421 more read by the GPU over PCIe or from another GPU (1.5% of all 299520 routed)
strata serve: KV streaming: 96.65% of 3882060 block reads hit VRAM, 524.2 MiB read from RAM

### confirmation/prefill16/perf-sampled-131072-r1
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 18320 ms (7154.5 tok/s), 512 generated in 2863 ms (178.9 tok/s), drafts accepted 308 of 424, 6 checkpoints
strata serve: decode expert cache hit rate: 95.6% (282066 hits / 295193 lookups); 7687 more read by the GPU over PCIe or from another GPU (2.5% of all 302880 routed)
strata serve: KV streaming: 96.51% of 3912876 block reads hit VRAM, 550.8 MiB read from RAM
strata serve: suffix drafts: 15 windows, 49 of 59 drafts accepted

### confirmation/baseline/warmup
strata verify: capturing the 6-token window (970 MiB of VRAM free)
strata verify: captured the 6-token window (upload no error, sync no error)
strata verify: capturing the 1-token window (966 MiB of VRAM free)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: capturing the 2-token window (957 MiB of VRAM free)
strata verify: captured the 2-token window (upload no error, sync no error)
strata verify: capturing the 4-token window (947 MiB of VRAM free)
strata verify: captured the 4-token window (upload no error, sync no error)
strata verify: capturing the 3-token window (939 MiB of VRAM free)
strata verify: captured the 3-token window (upload no error, sync no error)
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 21097 ms (6212.7 tok/s), 64 generated in 428 ms (149.5 tok/s), drafts accepted 36 of 56, 6 checkpoints
strata serve: decode expert cache hit rate: 94.6% (37117 hits / 39255 lookups); 1065 more read by the GPU over PCIe or from another GPU (2.6% of all 40320 routed)
strata serve: KV streaming: 90.41% of 554568 block reads hit VRAM, 214.3 MiB read from RAM

### confirmation/baseline/perf-greedy-32768-r0
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 4958 ms (6609.7 tok/s), 512 generated in 2540 ms (201.6 tok/s), drafts accepted 322 of 438, 2 checkpoints
strata serve: decode expert cache hit rate: 95.6% (281244 hits / 294251 lookups); 7189 more read by the GPU over PCIe or from another GPU (2.4% of all 301440 routed)
strata serve: KV streaming: 98.28% of 3906696 block reads hit VRAM, 271.2 MiB read from RAM
strata serve: suffix drafts: 6 windows, 10 of 18 drafts accepted

### confirmation/baseline/perf-sampled-32768-r0
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 4821 ms (6796.6 tok/s), 512 generated in 3009 ms (170.2 tok/s), drafts accepted 284 of 406, 2 checkpoints
strata serve: decode expert cache hit rate: 96.4% (287680 hits / 298486 lookups); 5834 more read by the GPU over PCIe or from another GPU (1.9% of all 304320 routed)
strata serve: KV streaming: 98.34% of 3931344 block reads hit VRAM, 262.3 MiB read from RAM
strata serve: suffix drafts: 6 windows, 12 of 15 drafts accepted

### confirmation/baseline/perf-greedy-131072-r0
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 20745 ms (6318.1 tok/s), 512 generated in 2668 ms (191.9 tok/s), drafts accepted 315 of 451, 6 checkpoints
strata serve: decode expert cache hit rate: 97.2% (298826 hits / 307452 lookups); 4068 more read by the GPU over PCIe or from another GPU (1.3% of all 311520 routed)
strata serve: KV streaming: 96.88% of 4036152 block reads hit VRAM, 507.3 MiB read from RAM

### confirmation/baseline/perf-sampled-131072-r0
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 20059 ms (6534.4 tok/s), 512 generated in 2842 ms (180.1 tok/s), drafts accepted 315 of 480, 6 checkpoints
strata serve: decode expert cache hit rate: 96.1% (306006 hits / 318498 lookups); 7422 more read by the GPU over PCIe or from another GPU (2.3% of all 325920 routed)
strata serve: KV streaming: 96.72% of 4208604 block reads hit VRAM, 556.8 MiB read from RAM
strata serve: suffix drafts: 12 windows, 33 of 43 drafts accepted

### confirmation/baseline/perf-greedy-32768-r1
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 4844 ms (6764.1 tok/s), 512 generated in 2610 ms (196.2 tok/s), drafts accepted 312 of 439, 2 checkpoints
strata serve: decode expert cache hit rate: 96.6% (291939 hits / 302103 lookups); 5097 more read by the GPU over PCIe or from another GPU (1.7% of all 307200 routed)
strata serve: KV streaming: 98.36% of 3980664 block reads hit VRAM, 262.2 MiB read from RAM
strata serve: suffix drafts: 1 windows, 1 of 5 drafts accepted

### confirmation/baseline/perf-sampled-32768-r1
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 4847 ms (6761.1 tok/s), 512 generated in 2828 ms (181.0 tok/s), drafts accepted 274 of 401, 2 checkpoints
strata serve: decode expert cache hit rate: 96.1% (288538 hits / 300262 lookups); 6458 more read by the GPU over PCIe or from another GPU (2.1% of all 306720 routed)
strata serve: KV streaming: 98.31% of 3962148 block reads hit VRAM, 270.3 MiB read from RAM
strata serve: suffix drafts: 2 windows, 5 of 8 drafts accepted

### confirmation/baseline/perf-greedy-131072-r1
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 20059 ms (6534.3 tok/s), 512 generated in 2705 ms (189.3 tok/s), drafts accepted 307 of 439, 6 checkpoints
strata serve: decode expert cache hit rate: 97.4% (297240 hits / 305323 lookups); 3797 more read by the GPU over PCIe or from another GPU (1.2% of all 309120 routed)
strata serve: KV streaming: 96.88% of 4005348 block reads hit VRAM, 503.9 MiB read from RAM

### confirmation/baseline/perf-sampled-131072-r1
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 20646 ms (6348.5 tok/s), 512 generated in 2886 ms (177.4 tok/s), drafts accepted 316 of 446, 6 checkpoints
strata serve: decode expert cache hit rate: 96.4% (292068 hits / 302968 lookups); 6152 more read by the GPU over PCIe or from another GPU (2.0% of all 309120 routed)
strata serve: KV streaming: 96.76% of 3992904 block reads hit VRAM, 521.8 MiB read from RAM
strata serve: suffix drafts: 6 windows, 18 of 20 drafts accepted

### screen/baseline/warmup
strata verify: capturing the 6-token window (1016 MiB of VRAM free)
strata verify: captured the 6-token window (upload no error, sync no error)
strata verify: capturing the 1-token window (1012 MiB of VRAM free)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: capturing the 2-token window (1002 MiB of VRAM free)
strata verify: captured the 2-token window (upload no error, sync no error)
strata verify: capturing the 4-token window (998 MiB of VRAM free)
strata verify: captured the 4-token window (upload no error, sync no error)
strata verify: capturing the 3-token window (990 MiB of VRAM free)
strata verify: captured the 3-token window (upload no error, sync no error)
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 20004 ms (6552.4 tok/s), 64 generated in 397 ms (161.2 tok/s), drafts accepted 36 of 57, 6 checkpoints
strata serve: decode expert cache hit rate: 94.5% (37503 hits / 39706 lookups); 1094 more read by the GPU over PCIe or from another GPU (2.7% of all 40800 routed)
strata serve: KV streaming: 90.47% of 560736 block reads hit VRAM, 215.2 MiB read from RAM

### screen/baseline/perf-greedy-32768-r0
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 4818 ms (6800.8 tok/s), 512 generated in 2619 ms (195.5 tok/s), drafts accepted 311 of 429, 2 checkpoints
strata serve: decode expert cache hit rate: 95.4% (281233 hits / 294892 lookups); 7508 more read by the GPU over PCIe or from another GPU (2.5% of all 302400 routed)
strata serve: KV streaming: 98.29% of 3919008 block reads hit VRAM, 269.5 MiB read from RAM
strata serve: suffix drafts: 6 windows, 11 of 18 drafts accepted

### screen/baseline/perf-greedy-131072-r0
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 19947 ms (6571.0 tok/s), 512 generated in 2684 ms (190.8 tok/s), drafts accepted 303 of 440, 6 checkpoints
strata serve: decode expert cache hit rate: 97.9% (302472 hits / 309112 lookups); 2888 more read by the GPU over PCIe or from another GPU (0.9% of all 312000 routed)
strata serve: KV streaming: 96.76% of 4042284 block reads hit VRAM, 527.9 MiB read from RAM
strata serve: suffix drafts: 1 windows, 1 of 3 drafts accepted

### screen/prefill16/warmup
strata verify: capturing the 6-token window (1028 MiB of VRAM free)
strata verify: captured the 6-token window (upload no error, sync no error)
strata verify: capturing the 1-token window (1024 MiB of VRAM free)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: capturing the 2-token window (1014 MiB of VRAM free)
strata verify: captured the 2-token window (upload no error, sync no error)
strata verify: capturing the 3-token window (1010 MiB of VRAM free)
strata verify: captured the 3-token window (upload no error, sync no error)
strata verify: capturing the 4-token window (1002 MiB of VRAM free)
strata verify: captured the 4-token window (upload no error, sync no error)
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 18253 ms (7180.9 tok/s), 64 generated in 431 ms (148.4 tok/s), drafts accepted 34 of 56, 6 checkpoints
strata serve: decode expert cache hit rate: 94.1% (37657 hits / 40031 lookups); 1249 more read by the GPU over PCIe or from another GPU (3.0% of all 41280 routed)
strata serve: KV streaming: 90.56% of 566940 block reads hit VRAM, 215.5 MiB read from RAM
strata serve: suffix drafts: 2 windows, 3 of 5 drafts accepted

### screen/prefill16/perf-greedy-32768-r0
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 4439 ms (7382.3 tok/s), 512 generated in 2520 ms (203.1 tok/s), drafts accepted 325 of 450, 2 checkpoints
strata serve: decode expert cache hit rate: 95.4% (285587 hits / 299278 lookups); 7442 more read by the GPU over PCIe or from another GPU (2.4% of all 306720 routed)
strata serve: KV streaming: 98.29% of 3974472 block reads hit VRAM, 273.3 MiB read from RAM
strata serve: suffix drafts: 6 windows, 12 of 18 drafts accepted

### screen/prefill16/perf-greedy-131072-r0
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 18235 ms (7188.1 tok/s), 512 generated in 2633 ms (194.4 tok/s), drafts accepted 301 of 463, 6 checkpoints
strata serve: decode expert cache hit rate: 97.7% (312696 hits / 320208 lookups); 3312 more read by the GPU over PCIe or from another GPU (1.0% of all 323520 routed)
strata serve: KV streaming: 96.83% of 4190148 block reads hit VRAM, 535.0 MiB read from RAM
strata serve: suffix drafts: 1 windows, 1 of 3 drafts accepted

### screen/prefill32/warmup
strata verify: capturing the 6-token window (1028 MiB of VRAM free)
strata verify: captured the 6-token window (upload no error, sync no error)
strata verify: capturing the 1-token window (1024 MiB of VRAM free)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: capturing the 2-token window (1014 MiB of VRAM free)
strata verify: captured the 2-token window (upload no error, sync no error)
strata verify: capturing the 4-token window (1010 MiB of VRAM free)
strata verify: captured the 4-token window (upload no error, sync no error)
strata verify: capturing the 3-token window (1002 MiB of VRAM free)
strata verify: captured the 3-token window (upload no error, sync no error)
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 17652 ms (7425.4 tok/s), 64 generated in 512 ms (125.0 tok/s), drafts accepted 28 of 56, 4 checkpoints
strata serve: decode expert cache hit rate: 91.2% (38293 hits / 41989 lookups); 2171 more read by the GPU over PCIe or from another GPU (4.9% of all 44160 routed)
strata serve: KV streaming: 91.24% of 603876 block reads hit VRAM, 213.2 MiB read from RAM

### screen/prefill32/perf-greedy-32768-r0
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 4389 ms (7466.5 tok/s), 512 generated in 2530 ms (202.4 tok/s), drafts accepted 316 of 448, 1 checkpoints
strata serve: decode expert cache hit rate: 96.0% (290332 hits / 302468 lookups); 6652 more read by the GPU over PCIe or from another GPU (2.2% of all 309120 routed)
strata serve: KV streaming: 98.33% of 4005288 block reads hit VRAM, 268.8 MiB read from RAM
strata serve: suffix drafts: 8 windows, 17 of 24 drafts accepted

### screen/prefill32/perf-greedy-131072-r0
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 17641 ms (7430.1 tok/s), 512 generated in 2542 ms (201.4 tok/s), drafts accepted 312 of 448, 4 checkpoints
strata serve: decode expert cache hit rate: 97.5% (301563 hits / 309157 lookups); 3323 more read by the GPU over PCIe or from another GPU (1.1% of all 312480 routed)
strata serve: KV streaming: 96.74% of 4048440 block reads hit VRAM, 531.1 MiB read from RAM
strata serve: suffix drafts: 1 windows, 1 of 3 drafts accepted

### screen/gdn/warmup
strata verify: capturing the 6-token window (992 MiB of VRAM free)
strata verify: captured the 6-token window (upload no error, sync no error)
strata verify: capturing the 1-token window (988 MiB of VRAM free)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: capturing the 2-token window (978 MiB of VRAM free)
strata verify: captured the 2-token window (upload no error, sync no error)
strata verify: capturing the 3-token window (974 MiB of VRAM free)
strata verify: captured the 3-token window (upload no error, sync no error)
strata verify: capturing the 4-token window (966 MiB of VRAM free)
strata verify: captured the 4-token window (upload no error, sync no error)
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 19547 ms (6705.4 tok/s), 64 generated in 432 ms (148.2 tok/s), drafts accepted 34 of 54, 6 checkpoints
strata serve: decode expert cache hit rate: 95.7% (38653 hits / 40409 lookups); 871 more read by the GPU over PCIe or from another GPU (2.1% of all 41280 routed)
strata serve: KV streaming: 91.64% of 566940 block reads hit VRAM, 190.9 MiB read from RAM

### screen/gdn/perf-greedy-32768-r0
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 4712 ms (6954.6 tok/s), 512 generated in 2556 ms (200.3 tok/s), drafts accepted 315 of 451, 2 checkpoints
strata serve: decode expert cache hit rate: 95.7% (290851 hits / 303911 lookups); 7129 more read by the GPU over PCIe or from another GPU (2.3% of all 311040 routed)
strata serve: KV streaming: 98.34% of 4029972 block reads hit VRAM, 269.3 MiB read from RAM
strata serve: suffix drafts: 4 windows, 9 of 12 drafts accepted

### screen/gdn/perf-greedy-131072-r0
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 19578 ms (6694.7 tok/s), 512 generated in 2770 ms (184.9 tok/s), drafts accepted 295 of 427, 6 checkpoints
strata serve: decode expert cache hit rate: 97.7% (299534 hits / 306705 lookups); 2895 more read by the GPU over PCIe or from another GPU (0.9% of all 309600 routed)
strata serve: KV streaming: 96.67% of 4011456 block reads hit VRAM, 538.4 MiB read from RAM
strata serve: suffix drafts: 1 windows, 1 of 3 drafts accepted

### screen/gdn16/warmup
strata verify: capturing the 6-token window (1038 MiB of VRAM free)
strata verify: captured the 6-token window (upload no error, sync no error)
strata verify: capturing the 1-token window (1034 MiB of VRAM free)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: capturing the 2-token window (1024 MiB of VRAM free)
strata verify: captured the 2-token window (upload no error, sync no error)
strata verify: capturing the 4-token window (1020 MiB of VRAM free)
strata verify: captured the 4-token window (upload no error, sync no error)
strata verify: capturing the 3-token window (1012 MiB of VRAM free)
strata verify: captured the 3-token window (upload no error, sync no error)
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 19105 ms (6860.4 tok/s), 64 generated in 498 ms (128.4 tok/s), drafts accepted 29 of 55, 6 checkpoints
strata serve: decode expert cache hit rate: 95.1% (40626 hits / 42701 lookups); 979 more read by the GPU over PCIe or from another GPU (2.2% of all 43680 routed)
strata serve: KV streaming: 91.46% of 597708 block reads hit VRAM, 205.5 MiB read from RAM

### screen/gdn16/perf-greedy-32768-r0
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 4743 ms (6908.5 tok/s), 512 generated in 3135 ms (163.3 tok/s), drafts accepted 311 of 466, 2 checkpoints
strata serve: decode expert cache hit rate: 95.5% (298175 hits / 312263 lookups); 7897 more read by the GPU over PCIe or from another GPU (2.5% of all 320160 routed)
strata serve: KV streaming: 98.36% of 4147032 block reads hit VRAM, 273.4 MiB read from RAM
strata serve: suffix drafts: 7 windows, 15 of 21 drafts accepted

### screen/gdn16/perf-greedy-131072-r0
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 18365 ms (7136.9 tok/s), 512 generated in 2535 ms (202.0 tok/s), drafts accepted 311 of 426, 6 checkpoints
strata serve: decode expert cache hit rate: 97.9% (292869 hits / 299232 lookups); 2688 more read by the GPU over PCIe or from another GPU (0.9% of all 301920 routed)
strata serve: KV streaming: 96.69% of 3912876 block reads hit VRAM, 522.0 MiB read from RAM

### screen/gdn32/warmup
strata verify: capturing the 6-token window (1042 MiB of VRAM free)
strata verify: captured the 6-token window (upload no error, sync no error)
strata verify: capturing the 1-token window (1038 MiB of VRAM free)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: capturing the 2-token window (1028 MiB of VRAM free)
strata verify: captured the 2-token window (upload no error, sync no error)
strata verify: capturing the 4-token window (1024 MiB of VRAM free)
strata verify: captured the 4-token window (upload no error, sync no error)
strata verify: capturing the 3-token window (1016 MiB of VRAM free)
strata verify: captured the 3-token window (upload no error, sync no error)
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 17801 ms (7363.1 tok/s), 64 generated in 400 ms (160.1 tok/s), drafts accepted 36 of 49, 4 checkpoints
strata serve: decode expert cache hit rate: 94.3% (33755 hits / 35812 lookups); 1148 more read by the GPU over PCIe or from another GPU (3.1% of all 36960 routed)
strata serve: KV streaming: 89.92% of 511464 block reads hit VRAM, 207.7 MiB read from RAM

### screen/gdn32/perf-greedy-32768-r0
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 4328 ms (7571.3 tok/s), 512 generated in 2702 ms (189.5 tok/s), drafts accepted 322 of 435, 1 checkpoints
strata serve: decode expert cache hit rate: 95.4% (278873 hits / 292326 lookups); 7674 more read by the GPU over PCIe or from another GPU (2.6% of all 300000 routed)
strata serve: KV streaming: 98.30% of 3888276 block reads hit VRAM, 266.8 MiB read from RAM
strata serve: suffix drafts: 6 windows, 13 of 20 drafts accepted

### screen/gdn32/perf-greedy-131072-r0
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 17987 ms (7287.1 tok/s), 512 generated in 2696 ms (189.9 tok/s), drafts accepted 305 of 450, 4 checkpoints
strata serve: decode expert cache hit rate: 97.7% (305185 hits / 312229 lookups); 3131 more read by the GPU over PCIe or from another GPU (1.0% of all 315360 routed)
strata serve: KV streaming: 96.76% of 4085412 block reads hit VRAM, 533.7 MiB read from RAM
strata serve: suffix drafts: 2 windows, 4 of 6 drafts accepted

### decode/baseline/warmup-greedy-r0
strata verify: capturing the 6-token window (1015 MiB of VRAM free)
strata verify: captured the 6-token window (upload no error, sync no error)
strata verify: capturing the 1-token window (1011 MiB of VRAM free)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: capturing the 4-token window (1001 MiB of VRAM free)
strata verify: captured the 4-token window (upload no error, sync no error)
strata verify: capturing the 2-token window (997 MiB of VRAM free)
strata verify: captured the 2-token window (upload no error, sync no error)
strata verify: capturing the 3-token window (989 MiB of VRAM free)
strata verify: captured the 3-token window (upload no error, sync no error)
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 5611 ms (5839.8 tok/s), 512 generated in 3150 ms (162.6 tok/s), drafts accepted 315 of 438, 2 checkpoints
strata serve: decode expert cache hit rate: 94.5% (279389 hits / 295584 lookups); 9216 more read by the GPU over PCIe or from another GPU (3.0% of all 304800 routed)
strata serve: KV streaming: 98.31% of 3949752 block reads hit VRAM, 268.9 MiB read from RAM
strata serve: suffix drafts: 8 windows, 23 of 26 drafts accepted

### decode/baseline/point-greedy-r0-pcie0.35-spec0.3
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 31 ms (223.9 tok/s), 512 generated in 2810 ms (182.2 tok/s), drafts accepted 337 of 498, 2 checkpoints
strata serve: decode expert cache hit rate: 96.0% (309002 hits / 321881 lookups); 2599 more read by the GPU over PCIe or from another GPU (0.8% of all 324480 routed)
strata serve: KV streaming: 99.09% of 8152236 block reads hit VRAM, 298.9 MiB read from RAM
strata serve: suffix drafts: 8 windows, 25 of 32 drafts accepted

### decode/baseline/point-greedy-r0-pcie0.35-spec0.5
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 33 ms (211.0 tok/s), 512 generated in 2777 ms (184.4 tok/s), drafts accepted 311 of 415, 2 checkpoints
strata serve: decode expert cache hit rate: 96.1% (283223 hits / 294648 lookups); 1992 more read by the GPU over PCIe or from another GPU (0.7% of all 296640 routed)
strata serve: KV streaming: 99.37% of 11997324 block reads hit VRAM, 305.1 MiB read from RAM
strata serve: suffix drafts: 5 windows, 15 of 18 drafts accepted

### decode/baseline/point-greedy-r0-pcie0.35-spec0.7
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 32 ms (219.6 tok/s), 512 generated in 2625 ms (195.0 tok/s), drafts accepted 299 of 349, 2 checkpoints
strata serve: decode expert cache hit rate: 97.0% (260878 hits / 268954 lookups); 1286 more read by the GPU over PCIe or from another GPU (0.5% of all 270240 routed)
strata serve: KV streaming: 99.51% of 15503544 block reads hit VRAM, 307.7 MiB read from RAM
strata serve: suffix drafts: 7 windows, 18 of 23 drafts accepted

### decode/baseline/point-greedy-r0-pcie0.55-spec0.3
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 24 ms (286.7 tok/s), 512 generated in 2603 ms (196.7 tok/s), drafts accepted 326 of 514, 2 checkpoints
strata serve: decode expert cache hit rate: 97.4% (323704 hits / 332220 lookups); 4260 more read by the GPU over PCIe or from another GPU (1.3% of all 336480 routed)
strata serve: KV streaming: 99.61% of 19860096 block reads hit VRAM, 311.8 MiB read from RAM
strata serve: suffix drafts: 8 windows, 17 of 23 drafts accepted

### decode/baseline/point-greedy-r0-pcie0.55-spec0.5
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 31 ms (227.5 tok/s), 512 generated in 2800 ms (182.9 tok/s), drafts accepted 312 of 439, 2 checkpoints
strata serve: decode expert cache hit rate: 97.5% (297010 hits / 304648 lookups); 3512 more read by the GPU over PCIe or from another GPU (1.1% of all 308160 routed)
strata serve: KV streaming: 99.67% of 23853000 block reads hit VRAM, 315.1 MiB read from RAM
strata serve: suffix drafts: 6 windows, 18 of 23 drafts accepted

### decode/baseline/point-greedy-r0-pcie0.55-spec0.7
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 31 ms (228.4 tok/s), 512 generated in 2683 ms (190.9 tok/s), drafts accepted 300 of 347, 2 checkpoints
strata serve: decode expert cache hit rate: 97.4% (258168 hits / 265020 lookups); 3300 more read by the GPU over PCIe or from another GPU (1.2% of all 268320 routed)
strata serve: KV streaming: 99.71% of 27334488 block reads hit VRAM, 316.2 MiB read from RAM
strata serve: suffix drafts: 6 windows, 16 of 18 drafts accepted

### decode/baseline/point-greedy-r0-pcie0.75-spec0.3
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 28 ms (254.0 tok/s), 512 generated in 2643 ms (193.7 tok/s), drafts accepted 323 of 518, 2 checkpoints
strata serve: decode expert cache hit rate: 98.0% (326035 hits / 332824 lookups); 6536 more read by the GPU over PCIe or from another GPU (1.9% of all 339360 routed)
strata serve: KV streaming: 99.75% of 31728000 block reads hit VRAM, 317.3 MiB read from RAM
strata serve: suffix drafts: 5 windows, 14 of 19 drafts accepted

### decode/baseline/point-greedy-r0-pcie0.75-spec0.5
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 30 ms (234.1 tok/s), 512 generated in 2565 ms (199.6 tok/s), drafts accepted 319 of 437, 2 checkpoints
strata serve: decode expert cache hit rate: 97.7% (289542 hits / 296210 lookups); 6670 more read by the GPU over PCIe or from another GPU (2.2% of all 302880 routed)
strata serve: KV streaming: 99.78% of 35653092 block reads hit VRAM, 319.3 MiB read from RAM
strata serve: suffix drafts: 6 windows, 16 of 18 drafts accepted

### decode/baseline/point-greedy-r0-pcie0.75-spec0.7
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 30 ms (233.5 tok/s), 512 generated in 2493 ms (205.4 tok/s), drafts accepted 300 of 343, 2 checkpoints
strata serve: decode expert cache hit rate: 97.8% (256906 hits / 262609 lookups); 4751 more read by the GPU over PCIe or from another GPU (1.8% of all 267360 routed)
strata serve: KV streaming: 99.80% of 39122364 block reads hit VRAM, 319.5 MiB read from RAM
strata serve: suffix drafts: 5 windows, 17 of 19 drafts accepted

### decode/baseline/point-greedy-r0-pcie0.9-spec0.3
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 29 ms (239.8 tok/s), 512 generated in 2622 ms (195.3 tok/s), drafts accepted 323 of 528, 2 checkpoints
strata serve: decode expert cache hit rate: 98.3% (333613 hits / 339510 lookups); 6090 more read by the GPU over PCIe or from another GPU (1.8% of all 345600 routed)
strata serve: KV streaming: 99.82% of 43595904 block reads hit VRAM, 320.5 MiB read from RAM
strata serve: suffix drafts: 5 windows, 15 of 18 drafts accepted

### decode/baseline/point-greedy-r0-pcie0.9-spec0.5
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 32 ms (219.1 tok/s), 512 generated in 2480 ms (206.4 tok/s), drafts accepted 321 of 434, 2 checkpoints
strata serve: decode expert cache hit rate: 98.1% (288626 hits / 294075 lookups); 5925 more read by the GPU over PCIe or from another GPU (2.0% of all 300000 routed)
strata serve: KV streaming: 99.83% of 47484132 block reads hit VRAM, 320.6 MiB read from RAM
strata serve: suffix drafts: 6 windows, 19 of 22 drafts accepted

### decode/baseline/point-greedy-r0-pcie0.9-spec0.7
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 31 ms (226.9 tok/s), 512 generated in 2628 ms (194.8 tok/s), drafts accepted 302 of 373, 2 checkpoints
strata serve: decode expert cache hit rate: 97.8% (267078 hits / 273083 lookups); 7237 more read by the GPU over PCIe or from another GPU (2.6% of all 280320 routed)
strata serve: KV streaming: 99.84% of 51119676 block reads hit VRAM, 322.4 MiB read from RAM
strata serve: suffix drafts: 7 windows, 17 of 23 drafts accepted

### decode/baseline/warmup-sampled-r0
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 5240 ms (6253.4 tok/s), 512 generated in 2773 ms (184.6 tok/s), drafts accepted 330 of 408, 2 checkpoints
strata serve: decode expert cache hit rate: 95.7% (264170 hits / 276091 lookups); 7109 more read by the GPU over PCIe or from another GPU (2.5% of all 283200 routed)
strata serve: KV streaming: 98.14% of 3660180 block reads hit VRAM, 274.0 MiB read from RAM
strata serve: suffix drafts: 1 windows, 1 of 2 drafts accepted

### decode/baseline/point-sampled-r0-pcie0.35-spec0.3
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 22 ms (224.0 tok/s), 512 generated in 3077 ms (166.4 tok/s), drafts accepted 306 of 499, 2 checkpoints
strata serve: decode expert cache hit rate: 95.9% (322224 hits / 336156 lookups); 2724 more read by the GPU over PCIe or from another GPU (0.8% of all 338880 routed)
strata serve: KV streaming: 99.07% of 8035212 block reads hit VRAM, 302.3 MiB read from RAM
strata serve: suffix drafts: 5 windows, 7 of 16 drafts accepted

### decode/baseline/point-sampled-r0-pcie0.35-spec0.5
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 30 ms (164.2 tok/s), 512 generated in 2918 ms (175.5 tok/s), drafts accepted 322 of 441, 2 checkpoints
strata serve: decode expert cache hit rate: 95.3% (286329 hits / 300322 lookups); 3038 more read by the GPU over PCIe or from another GPU (1.0% of all 303360 routed)
strata serve: KV streaming: 99.35% of 11954292 block reads hit VRAM, 311.0 MiB read from RAM
strata serve: suffix drafts: 8 windows, 25 of 36 drafts accepted

### decode/baseline/point-sampled-r0-pcie0.35-spec0.7
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 27 ms (186.6 tok/s), 512 generated in 2956 ms (173.2 tok/s), drafts accepted 291 of 355, 2 checkpoints
strata serve: decode expert cache hit rate: 96.3% (264919 hits / 275119 lookups); 1841 more read by the GPU over PCIe or from another GPU (0.7% of all 276960 routed)
strata serve: KV streaming: 99.49% of 15534348 block reads hit VRAM, 316.8 MiB read from RAM
strata serve: suffix drafts: 7 windows, 4 of 19 drafts accepted

### decode/baseline/point-sampled-r0-pcie0.55-spec0.3
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 25 ms (200.1 tok/s), 512 generated in 2950 ms (173.6 tok/s), drafts accepted 328 of 509, 2 checkpoints
strata serve: decode expert cache hit rate: 97.1% (318057 hits / 327417 lookups); 5223 more read by the GPU over PCIe or from another GPU (1.6% of all 332640 routed)
strata serve: KV streaming: 99.59% of 19829256 block reads hit VRAM, 323.6 MiB read from RAM
strata serve: suffix drafts: 1 windows, 3 of 3 drafts accepted

### decode/baseline/point-sampled-r0-pcie0.55-spec0.5
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 25 ms (200.4 tok/s), 512 generated in 3237 ms (158.2 tok/s), drafts accepted 277 of 403, 2 checkpoints
strata serve: decode expert cache hit rate: 97.0% (292886 hits / 302066 lookups); 4654 more read by the GPU over PCIe or from another GPU (1.5% of all 306720 routed)
strata serve: KV streaming: 99.66% of 23791344 block reads hit VRAM, 326.8 MiB read from RAM
strata serve: suffix drafts: 1 windows, 1 of 3 drafts accepted

### decode/baseline/point-sampled-r0-pcie0.55-spec0.7
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 29 ms (173.2 tok/s), 512 generated in 3011 ms (170.1 tok/s), drafts accepted 300 of 363, 2 checkpoints
strata serve: decode expert cache hit rate: 97.0% (264584 hits / 272649 lookups); 4311 more read by the GPU over PCIe or from another GPU (1.6% of all 276960 routed)
strata serve: KV streaming: 99.70% of 27371424 block reads hit VRAM, 329.5 MiB read from RAM
strata serve: suffix drafts: 9 windows, 15 of 25 drafts accepted

### decode/baseline/point-sampled-r0-pcie0.75-spec0.3
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 26 ms (191.9 tok/s), 512 generated in 3178 ms (161.1 tok/s), drafts accepted 344 of 475, 2 checkpoints
strata serve: decode expert cache hit rate: 97.8% (294746 hits / 301240 lookups); 7400 more read by the GPU over PCIe or from another GPU (2.4% of all 308640 routed)
strata serve: KV streaming: 99.74% of 31358268 block reads hit VRAM, 331.6 MiB read from RAM
strata serve: suffix drafts: 9 windows, 23 of 33 drafts accepted

### decode/baseline/point-sampled-r0-pcie0.75-spec0.5
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 31 ms (163.3 tok/s), 512 generated in 3163 ms (161.9 tok/s), drafts accepted 328 of 436, 2 checkpoints
strata serve: decode expert cache hit rate: 98.2% (287992 hits / 293307 lookups); 4773 more read by the GPU over PCIe or from another GPU (1.6% of all 298080 routed)
strata serve: KV streaming: 99.77% of 35209512 block reads hit VRAM, 333.0 MiB read from RAM
strata serve: suffix drafts: 7 windows, 4 of 20 drafts accepted

### decode/baseline/point-sampled-r0-pcie0.75-spec0.7
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 31 ms (160.8 tok/s), 512 generated in 3390 ms (151.0 tok/s), drafts accepted 299 of 363, 2 checkpoints
strata serve: decode expert cache hit rate: 98.3% (267972 hits / 272639 lookups); 3841 more read by the GPU over PCIe or from another GPU (1.4% of all 276480 routed)
strata serve: KV streaming: 99.79% of 38783508 block reads hit VRAM, 333.9 MiB read from RAM
strata serve: suffix drafts: 4 windows, 3 of 12 drafts accepted

### decode/baseline/point-sampled-r0-pcie0.9-spec0.3
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 23 ms (215.4 tok/s), 512 generated in 3208 ms (159.6 tok/s), drafts accepted 323 of 518, 2 checkpoints
strata serve: decode expert cache hit rate: 98.2% (326619 hits / 332554 lookups); 7766 more read by the GPU over PCIe or from another GPU (2.3% of all 340320 routed)
strata serve: KV streaming: 99.81% of 43176996 block reads hit VRAM, 334.8 MiB read from RAM
strata serve: suffix drafts: 3 windows, 12 of 12 drafts accepted

### decode/baseline/point-sampled-r0-pcie0.9-spec0.5
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 29 ms (172.5 tok/s), 512 generated in 3214 ms (159.3 tok/s), drafts accepted 304 of 463, 2 checkpoints
strata serve: decode expert cache hit rate: 98.3% (310104 hits / 315476 lookups); 6604 more read by the GPU over PCIe or from another GPU (2.1% of all 322080 routed)
strata serve: KV streaming: 99.82% of 47336364 block reads hit VRAM, 335.6 MiB read from RAM
strata serve: suffix drafts: 8 windows, 21 of 27 drafts accepted

### decode/baseline/point-sampled-r0-pcie0.9-spec0.7
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 25 ms (201.2 tok/s), 512 generated in 3127 ms (163.7 tok/s), drafts accepted 277 of 338, 2 checkpoints
strata serve: decode expert cache hit rate: 98.2% (265111 hits / 269892 lookups); 5148 more read by the GPU over PCIe or from another GPU (1.9% of all 275040 routed)
strata serve: KV streaming: 99.84% of 50891820 block reads hit VRAM, 336.9 MiB read from RAM
strata serve: suffix drafts: 4 windows, 1 of 11 drafts accepted

### decode/baseline/warmup-greedy-r1
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 5910 ms (5544.3 tok/s), 512 generated in 2809 ms (182.3 tok/s), drafts accepted 322 of 432, 2 checkpoints
strata serve: decode expert cache hit rate: 95.7% (280571 hits / 293136 lookups); 6864 more read by the GPU over PCIe or from another GPU (2.3% of all 300000 routed)
strata serve: KV streaming: 98.29% of 3888252 block reads hit VRAM, 267.3 MiB read from RAM
strata serve: suffix drafts: 3 windows, 13 of 13 drafts accepted

### decode/baseline/point-greedy-r1-pcie0.9-spec0.7
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 29 ms (243.9 tok/s), 512 generated in 2876 ms (178.0 tok/s), drafts accepted 285 of 340, 2 checkpoints
strata serve: decode expert cache hit rate: 97.8% (261086 hits / 266993 lookups); 5647 more read by the GPU over PCIe or from another GPU (2.1% of all 272640 routed)
strata serve: KV streaming: 99.04% of 7425240 block reads hit VRAM, 287.8 MiB read from RAM
strata serve: suffix drafts: 4 windows, 12 of 12 drafts accepted

### decode/baseline/point-greedy-r1-pcie0.9-spec0.5
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 30 ms (231.5 tok/s), 512 generated in 2669 ms (191.9 tok/s), drafts accepted 319 of 422, 2 checkpoints
strata serve: decode expert cache hit rate: 97.9% (282051 hits / 288006 lookups); 7194 more read by the GPU over PCIe or from another GPU (2.4% of all 295200 routed)
strata serve: KV streaming: 99.35% of 11251848 block reads hit VRAM, 296.7 MiB read from RAM
strata serve: suffix drafts: 4 windows, 14 of 14 drafts accepted

### decode/baseline/point-greedy-r1-pcie0.9-spec0.3
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 32 ms (219.8 tok/s), 512 generated in 2751 ms (186.1 tok/s), drafts accepted 334 of 509, 2 checkpoints
strata serve: decode expert cache hit rate: 97.9% (313904 hits / 320744 lookups); 9016 more read by the GPU over PCIe or from another GPU (2.7% of all 329760 routed)
strata serve: KV streaming: 99.51% of 15522096 block reads hit VRAM, 305.0 MiB read from RAM
strata serve: suffix drafts: 3 windows, 12 of 13 drafts accepted

### decode/baseline/point-greedy-r1-pcie0.75-spec0.7
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 32 ms (220.6 tok/s), 512 generated in 2699 ms (189.7 tok/s), drafts accepted 294 of 342, 2 checkpoints
strata serve: decode expert cache hit rate: 97.7% (257346 hits / 263456 lookups); 5344 more read by the GPU over PCIe or from another GPU (2.0% of all 268800 routed)
strata serve: KV streaming: 99.60% of 19009740 block reads hit VRAM, 307.5 MiB read from RAM
strata serve: suffix drafts: 4 windows, 12 of 12 drafts accepted

### decode/baseline/point-greedy-r1-pcie0.75-spec0.5
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 30 ms (229.8 tok/s), 512 generated in 2593 ms (197.4 tok/s), drafts accepted 317 of 420, 2 checkpoints
strata serve: decode expert cache hit rate: 98.1% (286008 hits / 291557 lookups); 4603 more read by the GPU over PCIe or from another GPU (1.6% of all 296160 routed)
strata serve: KV streaming: 99.66% of 22848684 block reads hit VRAM, 310.4 MiB read from RAM
strata serve: suffix drafts: 3 windows, 11 of 11 drafts accepted

### decode/baseline/point-greedy-r1-pcie0.75-spec0.3
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 30 ms (231.6 tok/s), 512 generated in 2754 ms (185.9 tok/s), drafts accepted 318 of 521, 2 checkpoints
strata serve: decode expert cache hit rate: 98.1% (331341 hits / 337925 lookups); 5755 more read by the GPU over PCIe or from another GPU (1.7% of all 343680 routed)
strata serve: KV streaming: 99.72% of 27297564 block reads hit VRAM, 312.0 MiB read from RAM
strata serve: suffix drafts: 4 windows, 10 of 12 drafts accepted

### decode/baseline/point-greedy-r1-pcie0.55-spec0.7
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 30 ms (235.2 tok/s), 512 generated in 2664 ms (192.2 tok/s), drafts accepted 297 of 340, 2 checkpoints
strata serve: decode expert cache hit rate: 97.6% (257801 hits / 264069 lookups); 2811 more read by the GPU over PCIe or from another GPU (1.1% of all 266880 routed)
strata serve: KV streaming: 99.75% of 30760620 block reads hit VRAM, 312.5 MiB read from RAM
strata serve: suffix drafts: 4 windows, 14 of 14 drafts accepted

### decode/baseline/point-greedy-r1-pcie0.55-spec0.5
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 29 ms (243.2 tok/s), 512 generated in 2618 ms (195.6 tok/s), drafts accepted 311 of 412, 2 checkpoints
strata serve: decode expert cache hit rate: 97.6% (283726 hits / 290829 lookups); 3411 more read by the GPU over PCIe or from another GPU (1.2% of all 294240 routed)
strata serve: KV streaming: 99.77% of 34574928 block reads hit VRAM, 314.3 MiB read from RAM
strata serve: suffix drafts: 5 windows, 14 of 17 drafts accepted

### decode/baseline/point-greedy-r1-pcie0.55-spec0.3
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 31 ms (229.3 tok/s), 512 generated in 2674 ms (191.5 tok/s), drafts accepted 325 of 516, 2 checkpoints
strata serve: decode expert cache hit rate: 97.4% (324219 hits / 333033 lookups); 4407 more read by the GPU over PCIe or from another GPU (1.3% of all 337440 routed)
strata serve: KV streaming: 99.80% of 38943804 block reads hit VRAM, 315.4 MiB read from RAM
strata serve: suffix drafts: 4 windows, 13 of 14 drafts accepted

### decode/baseline/point-greedy-r1-pcie0.35-spec0.7
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 32 ms (218.5 tok/s), 512 generated in 2750 ms (186.2 tok/s), drafts accepted 296 of 336, 2 checkpoints
strata serve: decode expert cache hit rate: 96.7% (255914 hits / 264528 lookups); 1392 more read by the GPU over PCIe or from another GPU (0.5% of all 265920 routed)
strata serve: KV streaming: 99.82% of 42394488 block reads hit VRAM, 315.6 MiB read from RAM
strata serve: suffix drafts: 4 windows, 12 of 12 drafts accepted

### decode/baseline/point-greedy-r1-pcie0.35-spec0.5
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 34 ms (208.0 tok/s), 512 generated in 2925 ms (175.0 tok/s), drafts accepted 311 of 443, 2 checkpoints
strata serve: decode expert cache hit rate: 96.5% (296946 hits / 307830 lookups); 1770 more read by the GPU over PCIe or from another GPU (0.6% of all 309600 routed)
strata serve: KV streaming: 99.83% of 46406040 block reads hit VRAM, 320.2 MiB read from RAM
strata serve: suffix drafts: 2 windows, 10 of 10 drafts accepted

### decode/baseline/point-greedy-r1-pcie0.35-spec0.3
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 34 ms (203.1 tok/s), 512 generated in 2918 ms (175.4 tok/s), drafts accepted 319 of 529, 2 checkpoints
strata serve: decode expert cache hit rate: 96.5% (332403 hits / 344414 lookups); 2146 more read by the GPU over PCIe or from another GPU (0.6% of all 346560 routed)
strata serve: KV streaming: 99.84% of 50892000 block reads hit VRAM, 320.8 MiB read from RAM
strata serve: suffix drafts: 5 windows, 15 of 18 drafts accepted

### decode/baseline/warmup-sampled-r1
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 6212 ms (5274.8 tok/s), 512 generated in 2833 ms (180.7 tok/s), drafts accepted 327 of 402, 2 checkpoints
strata serve: decode expert cache hit rate: 95.5% (262510 hits / 274844 lookups); 7396 more read by the GPU over PCIe or from another GPU (2.6% of all 282240 routed)
strata serve: KV streaming: 98.14% of 3647940 block reads hit VRAM, 273.9 MiB read from RAM
strata serve: suffix drafts: 6 windows, 20 of 20 drafts accepted

### decode/baseline/point-sampled-r1-pcie0.9-spec0.7
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 38 ms (130.7 tok/s), 512 generated in 3477 ms (147.3 tok/s), drafts accepted 285 of 324, 2 checkpoints
strata serve: decode expert cache hit rate: 98.0% (254668 hits / 259744 lookups); 5216 more read by the GPU over PCIe or from another GPU (2.0% of all 264960 routed)
strata serve: KV streaming: 98.98% of 7074000 block reads hit VRAM, 292.1 MiB read from RAM
strata serve: suffix drafts: 8 windows, 11 of 21 drafts accepted

### decode/baseline/point-sampled-r1-pcie0.9-spec0.5
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 28 ms (175.8 tok/s), 512 generated in 3019 ms (169.6 tok/s), drafts accepted 321 of 430, 2 checkpoints
strata serve: decode expert cache hit rate: 98.0% (285082 hits / 290842 lookups); 7718 more read by the GPU over PCIe or from another GPU (2.6% of all 298560 routed)
strata serve: KV streaming: 99.32% of 10931412 block reads hit VRAM, 299.2 MiB read from RAM
strata serve: suffix drafts: 4 windows, 12 of 15 drafts accepted

### decode/baseline/point-sampled-r1-pcie0.9-spec0.3
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 26 ms (191.5 tok/s), 512 generated in 2926 ms (175.0 tok/s), drafts accepted 335 of 490, 2 checkpoints
strata serve: decode expert cache hit rate: 98.2% (306880 hits / 312559 lookups); 8081 more read by the GPU over PCIe or from another GPU (2.5% of all 320640 routed)
strata serve: KV streaming: 99.50% of 15072276 block reads hit VRAM, 306.2 MiB read from RAM
strata serve: suffix drafts: 6 windows, 18 of 20 drafts accepted

### decode/baseline/point-sampled-r1-pcie0.75-spec0.7
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 28 ms (181.1 tok/s), 512 generated in 3028 ms (169.1 tok/s), drafts accepted 300 of 349, 2 checkpoints
strata serve: decode expert cache hit rate: 97.7% (257661 hits / 263842 lookups); 6398 more read by the GPU over PCIe or from another GPU (2.4% of all 270240 routed)
strata serve: KV streaming: 99.59% of 18566172 block reads hit VRAM, 308.7 MiB read from RAM
strata serve: suffix drafts: 6 windows, 15 of 22 drafts accepted

### decode/baseline/point-sampled-r1-pcie0.75-spec0.5
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 24 ms (208.5 tok/s), 512 generated in 2753 ms (186.0 tok/s), drafts accepted 331 of 424, 2 checkpoints
strata serve: decode expert cache hit rate: 98.1% (279587 hits / 285051 lookups); 5349 more read by the GPU over PCIe or from another GPU (1.8% of all 290400 routed)
strata serve: KV streaming: 99.65% of 22318848 block reads hit VRAM, 313.0 MiB read from RAM

### decode/baseline/point-sampled-r1-pcie0.75-spec0.3
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 30 ms (164.9 tok/s), 512 generated in 2784 ms (183.9 tok/s), drafts accepted 338 of 485, 2 checkpoints
strata serve: decode expert cache hit rate: 97.9% (303736 hits / 310340 lookups); 7420 more read by the GPU over PCIe or from another GPU (2.3% of all 317760 routed)
strata serve: KV streaming: 99.70% of 26422716 block reads hit VRAM, 316.3 MiB read from RAM
strata serve: suffix drafts: 8 windows, 20 of 24 drafts accepted

### decode/baseline/point-sampled-r1-pcie0.55-spec0.7
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 21 ms (235.5 tok/s), 512 generated in 2644 ms (193.7 tok/s), drafts accepted 300 of 363, 2 checkpoints
strata serve: decode expert cache hit rate: 97.0% (264083 hits / 272205 lookups); 4275 more read by the GPU over PCIe or from another GPU (1.5% of all 276480 routed)
strata serve: KV streaming: 99.73% of 29996748 block reads hit VRAM, 322.0 MiB read from RAM
strata serve: suffix drafts: 8 windows, 10 of 23 drafts accepted

### decode/baseline/point-sampled-r1-pcie0.55-spec0.5
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 24 ms (210.2 tok/s), 512 generated in 3324 ms (154.0 tok/s), drafts accepted 290 of 454, 2 checkpoints
strata serve: decode expert cache hit rate: 96.6% (307872 hits / 318577 lookups); 5903 more read by the GPU over PCIe or from another GPU (1.8% of all 324480 routed)
strata serve: KV streaming: 99.76% of 34186884 block reads hit VRAM, 327.7 MiB read from RAM
strata serve: suffix drafts: 3 windows, 4 of 9 drafts accepted

### decode/baseline/point-sampled-r1-pcie0.55-spec0.3
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 25 ms (197.2 tok/s), 512 generated in 2880 ms (177.8 tok/s), drafts accepted 315 of 514, 2 checkpoints
strata serve: decode expert cache hit rate: 97.2% (326550 hits / 336069 lookups); 5211 more read by the GPU over PCIe or from another GPU (1.5% of all 341280 routed)
strata serve: KV streaming: 99.79% of 38592780 block reads hit VRAM, 332.0 MiB read from RAM
strata serve: suffix drafts: 3 windows, 7 of 9 drafts accepted

### decode/baseline/point-sampled-r1-pcie0.35-spec0.7
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 33 ms (150.5 tok/s), 512 generated in 2905 ms (176.2 tok/s), drafts accepted 301 of 348, 2 checkpoints
strata serve: decode expert cache hit rate: 96.1% (256383 hits / 266837 lookups); 1963 more read by the GPU over PCIe or from another GPU (0.7% of all 268800 routed)
strata serve: KV streaming: 99.80% of 42068136 block reads hit VRAM, 333.7 MiB read from RAM
strata serve: suffix drafts: 7 windows, 12 of 21 drafts accepted

### decode/baseline/point-sampled-r1-pcie0.35-spec0.5
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 27 ms (184.6 tok/s), 512 generated in 2646 ms (193.5 tok/s), drafts accepted 332 of 413, 2 checkpoints
strata serve: decode expert cache hit rate: 96.7% (273364 hits / 282836 lookups); 1804 more read by the GPU over PCIe or from another GPU (0.6% of all 284640 routed)
strata serve: KV streaming: 99.82% of 45746892 block reads hit VRAM, 335.3 MiB read from RAM
strata serve: suffix drafts: 7 windows, 23 of 23 drafts accepted

### decode/baseline/point-sampled-r1-pcie0.35-spec0.3
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 24 ms (206.4 tok/s), 512 generated in 2780 ms (184.1 tok/s), drafts accepted 333 of 495, 2 checkpoints
strata serve: decode expert cache hit rate: 97.0% (312724 hits / 322235 lookups); 1765 more read by the GPU over PCIe or from another GPU (0.5% of all 324000 routed)
strata serve: KV streaming: 99.83% of 49930872 block reads hit VRAM, 336.5 MiB read from RAM

### decode/baseline/warmup-greedy-r2
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 5557 ms (5896.3 tok/s), 512 generated in 3103 ms (165.0 tok/s), drafts accepted 310 of 435, 2 checkpoints
strata serve: decode expert cache hit rate: 95.6% (285763 hits / 299024 lookups); 7216 more read by the GPU over PCIe or from another GPU (2.4% of all 306240 routed)
strata serve: KV streaming: 98.29% of 3968280 block reads hit VRAM, 274.0 MiB read from RAM
strata serve: suffix drafts: 3 windows, 12 of 13 drafts accepted

### decode/baseline/point-greedy-r2-pcie0.35-spec0.3
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 37 ms (186.9 tok/s), 512 generated in 2910 ms (176.0 tok/s), drafts accepted 322 of 523, 2 checkpoints
strata serve: decode expert cache hit rate: 96.5% (329100 hits / 341149 lookups); 2051 more read by the GPU over PCIe or from another GPU (0.6% of all 343200 routed)
strata serve: KV streaming: 99.15% of 8411136 block reads hit VRAM, 288.7 MiB read from RAM
strata serve: suffix drafts: 4 windows, 13 of 14 drafts accepted

### decode/baseline/point-greedy-r2-pcie0.35-spec0.5
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 33 ms (212.6 tok/s), 512 generated in 2783 ms (184.0 tok/s), drafts accepted 322 of 431, 2 checkpoints
strata serve: decode expert cache hit rate: 96.4% (286915 hits / 297601 lookups); 1919 more read by the GPU over PCIe or from another GPU (0.6% of all 299520 routed)
strata serve: KV streaming: 99.41% of 12293196 block reads hit VRAM, 294.3 MiB read from RAM
strata serve: suffix drafts: 5 windows, 13 of 17 drafts accepted

### decode/baseline/point-greedy-r2-pcie0.35-spec0.7
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 33 ms (214.8 tok/s), 512 generated in 2794 ms (183.2 tok/s), drafts accepted 295 of 339, 2 checkpoints
strata serve: decode expert cache hit rate: 96.5% (256122 hits / 265408 lookups); 1472 more read by the GPU over PCIe or from another GPU (0.6% of all 266880 routed)
strata serve: KV streaming: 99.54% of 15756312 block reads hit VRAM, 295.0 MiB read from RAM
strata serve: suffix drafts: 4 windows, 12 of 12 drafts accepted

### decode/baseline/point-greedy-r2-pcie0.55-spec0.3
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 28 ms (245.8 tok/s), 512 generated in 2571 ms (199.2 tok/s), drafts accepted 332 of 509, 2 checkpoints
strata serve: decode expert cache hit rate: 97.3% (317570 hits / 326282 lookups); 4438 more read by the GPU over PCIe or from another GPU (1.3% of all 330720 routed)
strata serve: KV streaming: 99.62% of 20038872 block reads hit VRAM, 302.7 MiB read from RAM
strata serve: suffix drafts: 5 windows, 14 of 17 drafts accepted

### decode/baseline/point-greedy-r2-pcie0.55-spec0.5
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 28 ms (248.7 tok/s), 512 generated in 2420 ms (211.6 tok/s), drafts accepted 317 of 417, 2 checkpoints
strata serve: decode expert cache hit rate: 97.3% (282334 hits / 290087 lookups); 3673 more read by the GPU over PCIe or from another GPU (1.3% of all 293760 routed)
strata serve: KV streaming: 99.68% of 23847012 block reads hit VRAM, 303.4 MiB read from RAM
strata serve: suffix drafts: 4 windows, 12 of 12 drafts accepted

### decode/baseline/point-greedy-r2-pcie0.55-spec0.7
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 25 ms (275.3 tok/s), 512 generated in 2461 ms (208.0 tok/s), drafts accepted 297 of 350, 2 checkpoints
strata serve: decode expert cache hit rate: 97.4% (260877 hits / 267893 lookups); 3307 more read by the GPU over PCIe or from another GPU (1.2% of all 271200 routed)
strata serve: KV streaming: 99.72% of 27365508 block reads hit VRAM, 305.1 MiB read from RAM
strata serve: suffix drafts: 6 windows, 14 of 20 drafts accepted

### decode/baseline/point-greedy-r2-pcie0.75-spec0.3
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 28 ms (251.3 tok/s), 512 generated in 2667 ms (192.0 tok/s), drafts accepted 322 of 524, 2 checkpoints
strata serve: decode expert cache hit rate: 97.7% (327627 hits / 335299 lookups); 7901 more read by the GPU over PCIe or from another GPU (2.3% of all 343200 routed)
strata serve: KV streaming: 99.76% of 31808364 block reads hit VRAM, 308.3 MiB read from RAM
strata serve: suffix drafts: 7 windows, 15 of 21 drafts accepted

### decode/baseline/point-greedy-r2-pcie0.75-spec0.5
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 30 ms (229.9 tok/s), 512 generated in 2691 ms (190.3 tok/s), drafts accepted 313 of 420, 2 checkpoints
strata serve: decode expert cache hit rate: 97.8% (284885 hits / 291324 lookups); 5796 more read by the GPU over PCIe or from another GPU (2.0% of all 297120 routed)
strata serve: KV streaming: 99.79% of 35659656 block reads hit VRAM, 308.6 MiB read from RAM
strata serve: suffix drafts: 3 windows, 9 of 9 drafts accepted

### decode/baseline/point-greedy-r2-pcie0.75-spec0.7
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 32 ms (221.9 tok/s), 512 generated in 2995 ms (171.0 tok/s), drafts accepted 288 of 340, 2 checkpoints
strata serve: decode expert cache hit rate: 97.9% (260593 hits / 266113 lookups); 4607 more read by the GPU over PCIe or from another GPU (1.7% of all 270720 routed)
strata serve: KV streaming: 99.80% of 39172008 block reads hit VRAM, 311.2 MiB read from RAM
strata serve: suffix drafts: 6 windows, 18 of 23 drafts accepted

### decode/baseline/point-greedy-r2-pcie0.9-spec0.3
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 33 ms (211.6 tok/s), 512 generated in 2989 ms (171.3 tok/s), drafts accepted 318 of 539, 2 checkpoints
strata serve: decode expert cache hit rate: 98.1% (337920 hits / 344393 lookups); 7447 more read by the GPU over PCIe or from another GPU (2.1% of all 351840 routed)
strata serve: KV streaming: 99.82% of 43725756 block reads hit VRAM, 312.0 MiB read from RAM
strata serve: suffix drafts: 4 windows, 14 of 17 drafts accepted

### decode/baseline/point-greedy-r2-pcie0.9-spec0.5
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 32 ms (216.1 tok/s), 512 generated in 2515 ms (203.6 tok/s), drafts accepted 318 of 433, 2 checkpoints
strata serve: decode expert cache hit rate: 98.1% (290542 hits / 296190 lookups); 6210 more read by the GPU over PCIe or from another GPU (2.1% of all 302400 routed)
strata serve: KV streaming: 99.84% of 47644764 block reads hit VRAM, 312.5 MiB read from RAM
strata serve: suffix drafts: 5 windows, 14 of 17 drafts accepted

### decode/baseline/point-greedy-r2-pcie0.9-spec0.7
strata serve: prompt 32768 tokens = 32761 reused + 7 read in 31 ms (223.3 tok/s), 512 generated in 2576 ms (198.7 tok/s), drafts accepted 310 of 361, 2 checkpoints
strata serve: decode expert cache hit rate: 98.2% (261794 hits / 266562 lookups); 4638 more read by the GPU over PCIe or from another GPU (1.7% of all 271200 routed)
strata serve: KV streaming: 99.85% of 51163284 block reads hit VRAM, 314.5 MiB read from RAM
strata serve: suffix drafts: 4 windows, 12 of 12 drafts accepted

### decode/baseline/warmup-sampled-r2
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 5575 ms (5877.5 tok/s), 512 generated in 2747 ms (186.4 tok/s), drafts accepted 326 of 430, 2 checkpoints
strata serve: decode expert cache hit rate: 94.7% (271635 hits / 286742 lookups); 9418 more read by the GPU over PCIe or from another GPU (3.2% of all 296160 routed)
strata serve: KV streaming: 98.18% of 3826572 block reads hit VRAM, 279.9 MiB read from RAM

### decode/baseline/point-sampled-r2-pcie0.35-spec0.3
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 24 ms (204.7 tok/s), 512 generated in 2992 ms (171.1 tok/s), drafts accepted 302 of 531, 2 checkpoints
strata serve: decode expert cache hit rate: 95.0% (334181 hits / 351884 lookups); 3796 more read by the GPU over PCIe or from another GPU (1.1% of all 355680 routed)
strata serve: KV streaming: 99.11% of 8417292 block reads hit VRAM, 302.5 MiB read from RAM
strata serve: suffix drafts: 6 windows, 24 of 26 drafts accepted

### decode/baseline/point-sampled-r2-pcie0.35-spec0.5
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 26 ms (190.6 tok/s), 512 generated in 2773 ms (184.6 tok/s), drafts accepted 325 of 432, 2 checkpoints
strata serve: decode expert cache hit rate: 95.8% (283065 hits / 295440 lookups); 2640 more read by the GPU over PCIe or from another GPU (0.9% of all 298080 routed)
strata serve: KV streaming: 99.36% of 12268536 block reads hit VRAM, 316.4 MiB read from RAM
strata serve: suffix drafts: 6 windows, 20 of 21 drafts accepted

### decode/baseline/point-sampled-r2-pcie0.35-spec0.7
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 28 ms (180.1 tok/s), 512 generated in 2737 ms (187.1 tok/s), drafts accepted 294 of 349, 2 checkpoints
strata serve: decode expert cache hit rate: 96.4% (261051 hits / 270785 lookups); 1855 more read by the GPU over PCIe or from another GPU (0.7% of all 272640 routed)
strata serve: KV streaming: 99.50% of 15793224 block reads hit VRAM, 320.4 MiB read from RAM
strata serve: suffix drafts: 7 windows, 17 of 25 drafts accepted

### decode/baseline/point-sampled-r2-pcie0.55-spec0.3
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 23 ms (216.6 tok/s), 512 generated in 3069 ms (166.8 tok/s), drafts accepted 300 of 507, 2 checkpoints
strata serve: decode expert cache hit rate: 97.2% (331625 hits / 341177 lookups); 4903 more read by the GPU over PCIe or from another GPU (1.4% of all 346080 routed)
strata serve: KV streaming: 99.60% of 20260668 block reads hit VRAM, 327.1 MiB read from RAM
strata serve: suffix drafts: 3 windows, 5 of 7 drafts accepted

### decode/baseline/point-sampled-r2-pcie0.55-spec0.5
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 24 ms (211.4 tok/s), 512 generated in 2577 ms (198.7 tok/s), drafts accepted 318 of 435, 2 checkpoints
strata serve: decode expert cache hit rate: 96.7% (286763 hits / 296493 lookups); 5427 more read by the GPU over PCIe or from another GPU (1.8% of all 301920 routed)
strata serve: KV streaming: 99.66% of 24161244 block reads hit VRAM, 332.1 MiB read from RAM
strata serve: suffix drafts: 3 windows, 6 of 8 drafts accepted

### decode/baseline/point-sampled-r2-pcie0.55-spec0.7
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 24 ms (209.9 tok/s), 512 generated in 2546 ms (201.1 tok/s), drafts accepted 295 of 350, 2 checkpoints
strata serve: decode expert cache hit rate: 97.3% (261575 hits / 268970 lookups); 3670 more read by the GPU over PCIe or from another GPU (1.3% of all 272640 routed)
strata serve: KV streaming: 99.70% of 27685872 block reads hit VRAM, 333.4 MiB read from RAM
strata serve: suffix drafts: 6 windows, 13 of 17 drafts accepted

### decode/baseline/point-sampled-r2-pcie0.75-spec0.3
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 23 ms (220.8 tok/s), 512 generated in 2851 ms (179.6 tok/s), drafts accepted 318 of 541, 2 checkpoints
strata serve: decode expert cache hit rate: 97.8% (337597 hits / 345285 lookups); 8475 more read by the GPU over PCIe or from another GPU (2.4% of all 353760 routed)
strata serve: KV streaming: 99.74% of 32251920 block reads hit VRAM, 334.4 MiB read from RAM
strata serve: suffix drafts: 6 windows, 18 of 18 drafts accepted

### decode/baseline/point-sampled-r2-pcie0.75-spec0.5
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 23 ms (219.2 tok/s), 512 generated in 2586 ms (198.0 tok/s), drafts accepted 349 of 421, 2 checkpoints
strata serve: decode expert cache hit rate: 96.6% (258312 hits / 267318 lookups); 13002 more read by the GPU over PCIe or from another GPU (4.6% of all 280320 routed)
strata serve: KV streaming: 99.76% of 35875188 block reads hit VRAM, 340.4 MiB read from RAM
strata serve: suffix drafts: 6 windows, 21 of 22 drafts accepted

### decode/baseline/point-sampled-r2-pcie0.75-spec0.7
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 26 ms (189.7 tok/s), 512 generated in 3067 ms (166.9 tok/s), drafts accepted 262 of 322, 2 checkpoints
strata serve: decode expert cache hit rate: 96.7% (256729 hits / 265542 lookups); 9018 more read by the GPU over PCIe or from another GPU (3.3% of all 274560 routed)
strata serve: KV streaming: 99.78% of 39424488 block reads hit VRAM, 342.3 MiB read from RAM
strata serve: suffix drafts: 5 windows, 6 of 11 drafts accepted

### decode/baseline/point-sampled-r2-pcie0.9-spec0.3
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 27 ms (184.2 tok/s), 512 generated in 2633 ms (194.4 tok/s), drafts accepted 332 of 497, 2 checkpoints
strata serve: decode expert cache hit rate: 98.0% (311678 hits / 318044 lookups); 8356 more read by the GPU over PCIe or from another GPU (2.6% of all 326400 routed)
strata serve: KV streaming: 99.80% of 43639296 block reads hit VRAM, 344.5 MiB read from RAM
strata serve: suffix drafts: 4 windows, 6 of 10 drafts accepted

### decode/baseline/point-sampled-r2-pcie0.9-spec0.5
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 24 ms (205.6 tok/s), 512 generated in 2713 ms (188.7 tok/s), drafts accepted 305 of 430, 2 checkpoints
strata serve: decode expert cache hit rate: 98.1% (293767 hits / 299604 lookups); 7116 more read by the GPU over PCIe or from another GPU (2.3% of all 306720 routed)
strata serve: KV streaming: 99.82% of 47601504 block reads hit VRAM, 346.2 MiB read from RAM
strata serve: suffix drafts: 5 windows, 13 of 14 drafts accepted

### decode/baseline/point-sampled-r2-pcie0.9-spec0.7
strata serve: prompt 32768 tokens = 32763 reused + 5 read in 23 ms (213.1 tok/s), 512 generated in 2583 ms (198.2 tok/s), drafts accepted 293 of 342, 2 checkpoints
strata serve: decode expert cache hit rate: 98.2% (260664 hits / 265328 lookups); 4912 more read by the GPU over PCIe or from another GPU (1.8% of all 270240 routed)
strata serve: KV streaming: 99.83% of 51095352 block reads hit VRAM, 346.6 MiB read from RAM
strata serve: suffix drafts: 3 windows, 2 of 7 drafts accepted

### disk/initial/initial-32768
strata verify: capturing the 6-token window (1019 MiB of VRAM free)
strata verify: captured the 6-token window (upload no error, sync no error)
strata verify: capturing the 1-token window (1015 MiB of VRAM free)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: capturing the 4-token window (1005 MiB of VRAM free)
strata verify: captured the 4-token window (upload no error, sync no error)
strata serve: prompt 32768 tokens = 0 reused + 32768 read in 5204 ms (6296.6 tok/s), 169 generated in 1038 ms (162.8 tok/s), drafts accepted 126 of 128, 2 checkpoints
strata serve: decode expert cache hit rate: 80.2% (56930 hits / 70968 lookups); 11112 more read by the GPU over PCIe or from another GPU (13.5% of all 82080 routed)
strata serve: KV streaming: 95.65% of 1090680 block reads hit VRAM, 191.3 MiB read from RAM
strata serve: suffix drafts: 1 windows, 3 of 5 drafts accepted

### disk/control/depth-32768-r0-continuation
strata verify: capturing the 2-token window (1001 MiB of VRAM free)
strata verify: captured the 2-token window (upload no error, sync no error)
strata serve: prompt 32987 tokens = 32936 reused + 51 read in 209 ms (243.9 tok/s), 169 generated in 825 ms (204.9 tok/s), drafts accepted 127 of 127, 3 checkpoints
strata serve: decode expert cache hit rate: 90.3% (68083 hits / 75370 lookups); 5750 more read by the GPU over PCIe or from another GPU (7.1% of all 81120 routed)
strata serve: KV streaming: 97.57% of 2440164 block reads hit VRAM, 238.5 MiB read from RAM
strata serve: suffix drafts: 2 windows, 10 of 10 drafts accepted

### disk/restored/depth-32768-r0-continuation
strata verify: capturing the 6-token window (1077 MiB of VRAM free)
strata verify: captured the 6-token window (upload no error, sync no error)
strata verify: capturing the 2-token window (1073 MiB of VRAM free)
strata verify: captured the 2-token window (upload no error, sync no error)
strata verify: capturing the 1-token window (1065 MiB of VRAM free)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: capturing the 5-token window (1059 MiB of VRAM free)
strata verify: captured the 5-token window (upload no error, sync no error)
strata verify: capturing the 4-token window (1051 MiB of VRAM free)
strata verify: captured the 4-token window (upload no error, sync no error)
strata serve: prompt 32987 tokens = 32936 reused + 51 read in 219 ms (232.9 tok/s), 169 generated in 1084 ms (155.9 tok/s), drafts accepted 140 of 140, 2 checkpoints
strata serve: decode expert cache hit rate: 76.6% (53060 hits / 69305 lookups); 14215 more read by the GPU over PCIe or from another GPU (17.0% of all 83520 routed)
strata serve: KV streaming: 96.28% of 1380288 block reads hit VRAM, 207.1 MiB read from RAM
strata serve: suffix drafts: 23 windows, 110 of 110 drafts accepted

### disk/32768/save
strata serve: session saved 32936 tokens, 1 of 2 checkpoints, 739131980 bytes to <MEASUREMENT_ROOT>/checkpoints/disk-disk-base-1m/orchestrator.bin in 401.9 ms (state 65.9 ms)

### disk/32768/restore
strata serve: session restored 32936 tokens, 1 checkpoints, 739131980 bytes from <MEASUREMENT_ROOT>/checkpoints/disk-disk-base-1m/orchestrator.bin in 269.7 ms (read+check 240.5 ms)

### disk/initial/initial-131072
strata verify: capturing the 6-token window (920 MiB of VRAM free)
strata verify: captured the 6-token window (upload no error, sync no error)
strata verify: capturing the 1-token window (916 MiB of VRAM free)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: capturing the 4-token window (906 MiB of VRAM free)
strata verify: captured the 4-token window (upload no error, sync no error)
strata verify: capturing the 2-token window (902 MiB of VRAM free)
strata verify: captured the 2-token window (upload no error, sync no error)
strata verify: capturing the 3-token window (894 MiB of VRAM free)
strata verify: captured the 3-token window (upload no error, sync no error)
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 22060 ms (5941.7 tok/s), 163 generated in 1121 ms (145.4 tok/s), drafts accepted 119 of 137, 6 checkpoints
strata serve: decode expert cache hit rate: 83.1% (64546 hits / 77702 lookups); 10618 more read by the GPU over PCIe or from another GPU (12.0% of all 88320 routed)
strata serve: KV streaming: 93.62% of 1170780 block reads hit VRAM, 300.8 MiB read from RAM
strata serve: suffix drafts: 1 windows, 2 of 5 drafts accepted

### disk/control/depth-131072-r0-continuation
strata serve: prompt 131285 tokens = 131065 reused + 220 read in 314 ms (700.3 tok/s), 163 generated in 807 ms (201.9 tok/s), drafts accepted 124 of 124, 6 checkpoints
strata serve: decode expert cache hit rate: 91.2% (67520 hits / 74061 lookups); 5139 more read by the GPU over PCIe or from another GPU (6.5% of all 79200 routed)
strata serve: KV streaming: 95.32% of 2224476 block reads hit VRAM, 419.6 MiB read from RAM
strata serve: suffix drafts: 29 windows, 91 of 91 drafts accepted

### disk/restored/depth-131072-r0-continuation
strata verify: capturing the 6-token window (1016 MiB of VRAM free)
strata verify: captured the 6-token window (upload no error, sync no error)
strata verify: capturing the 1-token window (1012 MiB of VRAM free)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: capturing the 5-token window (1002 MiB of VRAM free)
strata verify: captured the 5-token window (upload no error, sync no error)
strata verify: capturing the 4-token window (998 MiB of VRAM free)
strata verify: captured the 4-token window (upload no error, sync no error)
strata serve: prompt 131285 tokens = 131065 reused + 220 read in 402 ms (547.9 tok/s), 163 generated in 1035 ms (157.5 tok/s), drafts accepted 135 of 135, 2 checkpoints
strata serve: decode expert cache hit rate: 79.0% (53828 hits / 68165 lookups); 12475 more read by the GPU over PCIe or from another GPU (15.5% of all 80640 routed)
strata serve: KV streaming: 92.91% of 1072176 block reads hit VRAM, 306.4 MiB read from RAM
strata serve: suffix drafts: 20 windows, 99 of 99 drafts accepted

### disk/131072/save
strata serve: session saved 131237 tokens, 1 of 6 checkpoints, 2240424512 bytes to <MEASUREMENT_ROOT>/checkpoints/disk-disk-base-1m/orchestrator.bin in 993.7 ms (state 31.8 ms)

### disk/131072/restore
strata serve: session restored 131237 tokens, 1 checkpoints, 2240424512 bytes from <MEASUREMENT_ROOT>/checkpoints/disk-disk-base-1m/orchestrator.bin in 819.0 ms (read+check 706.6 ms)

### disk/initial/initial-1000000
strata verify: capturing the 6-token window (913 MiB of VRAM free)
strata verify: captured the 6-token window (upload no error, sync no error)
strata verify: capturing the 1-token window (909 MiB of VRAM free)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: capturing the 4-token window (899 MiB of VRAM free)
strata verify: captured the 4-token window (upload no error, sync no error)
strata verify: capturing the 3-token window (895 MiB of VRAM free)
strata verify: captured the 3-token window (upload no error, sync no error)
strata serve: prompt 1000000 tokens = 0 reused + 1000000 read in 243642 ms (4104.4 tok/s), 167 generated in 1295 ms (129.0 tok/s), drafts accepted 117 of 137, 6 checkpoints
strata serve: decode expert cache hit rate: 85.3% (68738 hits / 80623 lookups); 9137 more read by the GPU over PCIe or from another GPU (10.2% of all 89760 routed)
strata serve: KV streaming: 91.25% of 1189260 block reads hit VRAM, 419.0 MiB read from RAM
strata serve: suffix drafts: 1 windows, 1 of 5 drafts accepted

### disk/control/depth-1000000-r0-continuation
strata verify: capturing the 2-token window (887 MiB of VRAM free)
strata verify: captured the 2-token window (upload no error, sync no error)
strata serve: prompt 1000217 tokens = 1000166 reused + 51 read in 259 ms (196.8 tok/s), 167 generated in 885 ms (188.8 tok/s), drafts accepted 136 of 136, 6 checkpoints
strata serve: decode expert cache hit rate: 91.7% (69536 hits / 75838 lookups); 5282 more read by the GPU over PCIe or from another GPU (6.5% of all 81120 routed)
strata serve: KV streaming: 92.88% of 2538732 block reads hit VRAM, 728.2 MiB read from RAM
strata serve: suffix drafts: 20 windows, 100 of 100 drafts accepted

### disk/restored/depth-1000000-r0-continuation
strata verify: capturing the 6-token window (1076 MiB of VRAM free)
strata verify: captured the 6-token window (upload no error, sync no error)
strata verify: capturing the 2-token window (1072 MiB of VRAM free)
strata verify: captured the 2-token window (upload no error, sync no error)
strata verify: capturing the 1-token window (1064 MiB of VRAM free)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: capturing the 5-token window (1058 MiB of VRAM free)
strata verify: captured the 5-token window (upload no error, sync no error)
strata verify: capturing the 4-token window (1050 MiB of VRAM free)
strata verify: captured the 4-token window (upload no error, sync no error)
strata serve: prompt 1000217 tokens = 1000166 reused + 51 read in 255 ms (200.0 tok/s), 167 generated in 1110 ms (150.5 tok/s), drafts accepted 136 of 136, 2 checkpoints
strata serve: decode expert cache hit rate: 81.7% (57573 hits / 70431 lookups); 11169 more read by the GPU over PCIe or from another GPU (13.7% of all 81600 routed)
strata serve: KV streaming: 91.58% of 1355640 block reads hit VRAM, 459.6 MiB read from RAM
strata serve: suffix drafts: 19 windows, 94 of 94 drafts accepted

### disk/1000000/save
strata serve: session saved 1000166 tokens, 1 of 6 checkpoints, 15510692932 bytes to <MEASUREMENT_ROOT>/checkpoints/disk-disk-base-1m/orchestrator.bin in 7101.7 ms (state 53.2 ms)

### disk/1000000/restore
strata serve: session restored 1000166 tokens, 1 checkpoints, 15510692932 bytes from <MEASUREMENT_ROOT>/checkpoints/disk-disk-base-1m/orchestrator.bin in 6106.4 ms (read+check 5165.5 ms)

### learned-profile/initial/initial-131072
strata verify: capturing the 6-token window (1018 MiB of VRAM free)
strata verify: captured the 6-token window (upload no error, sync no error)
strata verify: capturing the 1-token window (1014 MiB of VRAM free)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: capturing the 4-token window (1004 MiB of VRAM free)
strata verify: captured the 4-token window (upload no error, sync no error)
strata verify: capturing the 2-token window (1000 MiB of VRAM free)
strata verify: captured the 2-token window (upload no error, sync no error)
strata verify: capturing the 3-token window (992 MiB of VRAM free)
strata verify: captured the 3-token window (upload no error, sync no error)
strata serve: prompt 131072 tokens = 0 reused + 131072 read in 23944 ms (5474.2 tok/s), 163 generated in 1278 ms (127.6 tok/s), drafts accepted 119 of 137, 6 checkpoints
strata serve: decode expert cache hit rate: 82.8% (64264 hits / 77573 lookups); 10747 more read by the GPU over PCIe or from another GPU (12.2% of all 88320 routed)
strata serve: KV streaming: 93.57% of 1170768 block reads hit VRAM, 303.3 MiB read from RAM
strata serve: suffix drafts: 1 windows, 1 of 5 drafts accepted

### learned-profile/control/depth-131072-r0-continuation
strata serve: prompt 131285 tokens = 131065 reused + 220 read in 352 ms (625.0 tok/s), 163 generated in 856 ms (190.3 tok/s), drafts accepted 131 of 131, 6 checkpoints
strata serve: decode expert cache hit rate: 90.8% (65911 hits / 72586 lookups); 5654 more read by the GPU over PCIe or from another GPU (7.2% of all 78240 routed)
strata serve: KV streaming: 95.32% of 2212140 block reads hit VRAM, 417.2 MiB read from RAM
strata serve: suffix drafts: 19 windows, 95 of 95 drafts accepted

### learned-profile/restored/depth-131072-r0-continuation
strata verify: capturing the 6-token window (995 MiB of VRAM free)
strata verify: captured the 6-token window (upload no error, sync no error)
strata verify: capturing the 1-token window (991 MiB of VRAM free)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: capturing the 5-token window (981 MiB of VRAM free)
strata verify: captured the 5-token window (upload no error, sync no error)
strata verify: capturing the 4-token window (977 MiB of VRAM free)
strata verify: captured the 4-token window (upload no error, sync no error)
strata serve: prompt 131285 tokens = 131065 reused + 220 read in 388 ms (567.5 tok/s), 163 generated in 801 ms (203.5 tok/s), drafts accepted 135 of 135, 2 checkpoints
strata serve: decode expert cache hit rate: 94.3% (72767 hits / 77192 lookups); 3448 more read by the GPU over PCIe or from another GPU (4.3% of all 80640 routed)
strata serve: KV streaming: 92.89% of 1072176 block reads hit VRAM, 307.2 MiB read from RAM
strata serve: suffix drafts: 20 windows, 99 of 99 drafts accepted
