shape           K      M  unfused ms  fused ms  ratio  sign  unf GB/s  fus GB/s  SM MHz
out_proj     6144     48       0.009     0.005  1.911  9/9      308.3     196.4    2520 [+-1 on 0.61%, scale 4.2e-04]
             6144     48  per-128-group     0.004 ms  paired group/per-token 0.9634x [0.8557-1.3371]      208.4 GB/s
out_proj     6144     64       0.009     0.005  1.841  9/9      409.0     251.0    2520 [+-1 on 0.62%, scale 4.6e-04]
             6144     64  per-128-group     0.004 ms  paired group/per-token 0.9534x [0.8649-1.0116]      270.0 GB/s
out_proj     6144   2048       0.089     0.040  2.202  9/9     1274.8     935.6    2520 [+-1 on 0.60%, scale 4.8e-04]
             6144   2048  per-128-group     0.038 ms  paired group/per-token 0.9320x [0.9284-0.9355]     1003.1 GB/s
out_proj     6144   8192       0.474     0.167  2.847  9/9      955.7     906.9    2520 [+-1 on 0.60%, scale 4.8e-04]
             6144   8192  per-128-group     0.165 ms  paired group/per-token 0.9911x [0.9891-0.9927]      914.3 GB/s
gate_proj    5120     48       0.008     0.004  2.177  9/9      260.2     188.9    2520 [+-1 on 0.57%, scale 4.4e-04]
             5120     48  per-128-group     0.004 ms  paired group/per-token 0.9697x [0.8912-1.0262]      194.9 GB/s
gate_proj    5120     64       0.008     0.004  2.019  9/9      350.8     236.0    2520 [+-1 on 0.60%, scale 4.6e-04]
             5120     64  per-128-group     0.004 ms  paired group/per-token 0.9241x [0.8639-1.0130]      259.5 GB/s
gate_proj    5120   2048       0.070     0.035  1.984  9/9     1355.3     896.4    2520 [+-1 on 0.60%, scale 4.8e-04]
             5120   2048  per-128-group     0.032 ms  paired group/per-token 0.9163x [0.9154-0.9368]      977.2 GB/s
gate_proj    5120   8192       0.380     0.143  2.655  9/9      993.6     879.2    2520 [+-1 on 0.60%, scale 4.8e-04]
             5120   8192  per-128-group     0.138 ms  paired group/per-token 0.9613x [0.9594-0.9627]      914.4 GB/s
