strata serve: prompt chunk auto: 8192 tokens, a 384-slot ring
strata prefill timing: 799 tokens, GPU timeline 2985 ms, wall 2994 ms, host staging 58 ms: embed+steps 5 (0.2%) hc read 77 (2.6%) gdn 177 (5.9%) qsa proj 89 (3.0%) qsa indexer 0 (0.0%) qsa select 1 (0.0%) qsa attn 51 (1.7%) router+shared 156 (5.2%) host grouping 12 (0.4%) gather 14 (0.5%) wait copy 1384 (46.4%) dequant 414 (13.9%) gemm gate/up 313 (10.5%) gemm down 154 (5.2%) combine 18 (0.6%) ple 3 (0.1%) kv stage 0 (0.0%) gdn conv+gates 5 (0.2%) gdn recurrence 35 (1.2%) gdn out proj 78 (2.6%)
strata serve: prompt 806 tokens = 0 reused + 806 read in 3296 ms (244.5 tok/s), 2 generated in 140 ms (14.3 tok/s), drafts accepted 3 of 3, 1 checkpoints
strata prefill timing: 495 tokens, GPU timeline 2139 ms, wall 2142 ms, host staging 98 ms: embed+steps 0 (0.0%) hc read 52 (2.5%) gdn 146 (6.8%) qsa proj 66 (3.1%) qsa indexer 0 (0.0%) qsa select 0 (0.0%) qsa attn 24 (1.1%) router+shared 29 (1.3%) host grouping 10 (0.5%) gather 3 (0.2%) wait copy 1075 (50.3%) dequant 318 (14.9%) gemm gate/up 222 (10.4%) gemm down 109 (5.1%) combine 10 (0.5%) ple 2 (0.1%) kv stage 0 (0.0%) gdn conv+gates 3 (0.1%) gdn recurrence 20 (0.9%) gdn out proj 50 (2.3%)
strata serve: prompt 502 tokens = 0 reused + 502 read in 2336 ms (214.9 tok/s), 2 generated in 124 ms (16.1 tok/s), drafts accepted 3 of 3, 1 checkpoints
strata prefill timing: 1108 tokens, GPU timeline 3395 ms, wall 3399 ms, host staging 176 ms: embed+steps 3 (0.1%) hc read 99 (2.9%) gdn 201 (5.9%) qsa proj 109 (3.2%) qsa indexer 0 (0.0%) qsa select 2 (0.0%) qsa attn 88 (2.6%) router+shared 51 (1.5%) host grouping 12 (0.4%) gather 7 (0.2%) wait copy 1637 (48.2%) dequant 448 (13.2%) gemm gate/up 360 (10.6%) gemm down 179 (5.3%) combine 26 (0.8%) ple 3 (0.1%) kv stage 0 (0.0%) gdn conv+gates 11 (0.3%) gdn recurrence 54 (1.6%) gdn out proj 102 (3.0%)
strata serve: prompt 1115 tokens = 0 reused + 1115 read in 3685 ms (302.6 tok/s), 2 generated in 115 ms (17.4 tok/s), drafts accepted 3 of 3, 1 checkpoints
strata prefill timing: 8874 tokens, GPU timeline 11459 ms, wall 11461 ms, host staging 329 ms: embed+steps 39 (0.3%) hc read 898 (7.8%) gdn 1128 (9.8%) qsa proj 595 (5.2%) qsa indexer 1 (0.0%) qsa select 62 (0.5%) qsa attn 1901 (16.6%) router+shared 214 (1.9%) host grouping 41 (0.4%) gather 56 (0.5%) wait copy 1670 (14.6%) dequant 860 (7.5%) gemm gate/up 1434 (12.5%) gemm down 710 (6.2%) combine 783 (6.8%) ple 28 (0.2%) kv stage 8 (0.1%) gdn conv+gates 92 (0.8%) gdn recurrence 492 (4.3%) gdn out proj 447 (3.9%)
strata serve: prompt 8881 tokens = 0 reused + 8881 read in 11952 ms (743.0 tok/s), 3 generated in 155 ms (19.3 tok/s), drafts accepted 3 of 3, 1 checkpoints
