Skip to content

GitLab

Commit 85762c1a authored by Xiaowei.zhang's avatar Xiaowei.zhang
Browse files

Init the main branch for aiter

parent ae0b3521
Pipeline #3505 canceled with stages
Expand all Show whitespace changes
Inline Side-by-side
Showing with 5937 additions and 0 deletions
aiter/configs/ck_tune/tuned_fmoe_ck.csv 0 → 100644
View file @ 85762c1a
arch,quant_type,indtype,token,inter_dim,model_dim,expert,topk,q_size_n,q_size_k,sol_type,sol_id,time_us
gfx936,no_quant,torch.float16,1,256,8192,256,4,0,0,ck,576,309.8021428571405
gfx936,no_quant,torch.float16,8,256,8192,256,4,0,0,ck,576,364.40785714285965
gfx936,no_quant,torch.float16,16,256,8192,256,4,0,0,ck,576,705.001571428565
gfx936,no_quant,torch.float16,24,256,8192,256,4,0,0,ck,272,1028.2664285714334
gfx936,no_quant,torch.float16,32,256,8192,256,4,0,0,ck,576,1293.2290000000055
gfx936,no_quant,torch.float16,48,256,8192,256,4,0,0,ck,272,1877.2707142857207
gfx936,no_quant,torch.float16,64,256,8192,256,4,0,0,ck,576,2193.247142857142
gfx936,no_quant,torch.float16,96,256,8192,256,4,0,0,ck,272,2471.966714285715
gfx936,no_quant,torch.float16,128,256,8192,256,4,0,0,ck,576,2724.7889999999948
gfx936,no_quant,torch.float16,256,256,8192,256,4,0,0,ck,576,3391.9192857142875
gfx936,no_quant,torch.float16,512,256,8192,256,4,0,0,ck,576,3537.358999999999
gfx936,no_quant,torch.float16,1024,256,8192,256,4,0,0,ck,272,5022.04200000001
gfx936,no_quant,torch.float16,2048,256,8192,256,4,0,0,ck,272,6217.148285714277
gfx936,no_quant,torch.float16,4096,256,8192,256,4,0,0,ck,272,8764.503571428571
gfx936,no_quant,torch.float16,8192,256,8192,256,4,0,0,ck,272,16520.489142857128
gfx936,no_quant,torch.float16,1,256,7168,256,8,0,0,ck,576,274.4422857142844
gfx936,no_quant,torch.float16,8,256,7168,256,8,0,0,ck,576,584.9560000000014
gfx936,no_quant,torch.float16,16,256,7168,256,8,0,0,ck,576,1056.360857142859
gfx936,no_quant,torch.float16,24,256,7168,256,8,0,0,ck,576,1480.196999999999
gfx936,no_quant,torch.float16,32,256,7168,256,8,0,0,ck,576,1867.970714285712
gfx936,no_quant,torch.float16,48,256,7168,256,8,0,0,ck,576,2312.378428571425
gfx936,no_quant,torch.float16,64,256,7168,256,8,0,0,ck,576,2382.001142857145
gfx936,no_quant,torch.float16,96,256,7168,256,8,0,0,ck,576,2795.691714285705
gfx936,no_quant,torch.float16,128,256,7168,256,8,0,0,ck,576,2820.2631428571426
gfx936,no_quant,torch.float16,256,256,7168,256,8,0,0,ck,576,3055.348428571423
gfx936,no_quant,torch.float16,512,256,7168,256,8,0,0,ck,272,4095.392285714287
gfx936,no_quant,torch.float16,1024,256,7168,256,8,0,0,ck,272,5514.589714285707
gfx936,no_quant,torch.float16,2048,256,7168,256,8,0,0,ck,272,7718.288428571438
gfx936,no_quant,torch.float16,4096,256,7168,256,8,0,0,ck,272,14435.327142857148
gfx936,no_quant,torch.float16,8192,256,7168,256,8,0,0,ck,272,27862.7415
aiter/configs/ck_tune/tuned_fmoe_ck_int8_w8a8_group.csv 0 → 100644
View file @ 85762c1a
arch,quant_type,indtype,token,inter_dim,model_dim,expert,topk,q_size_n,q_size_k,sol_type,sol_id,time_us
gfx936,int8_w8a8_block,torch.float16,1,256,7168,256,8,128,128,ck,1073741824,221.7238
gfx936,int8_w8a8_block,torch.float16,2,256,7168,256,8,128,128,ck,1073741824,271.6008
gfx936,int8_w8a8_block,torch.float16,3,256,7168,256,8,128,128,ck,256,347.8034
gfx936,int8_w8a8_block,torch.float16,4,256,7168,256,8,128,128,ck,256,351.302
gfx936,int8_w8a8_block,torch.float16,5,256,7168,256,8,128,128,ck,256,358.7936
gfx936,int8_w8a8_block,torch.float16,6,256,7168,256,8,128,128,ck,256,449.8388
gfx936,int8_w8a8_block,torch.float16,7,256,7168,256,8,128,128,ck,256,447.4445
gfx936,int8_w8a8_block,torch.float16,8,256,7168,256,8,128,128,ck,256,454.0529
gfx936,int8_w8a8_block,torch.float16,9,256,7168,256,8,128,128,ck,256,470.7523
gfx936,int8_w8a8_block,torch.float16,10,256,7168,256,8,128,128,ck,256,496.542
gfx936,int8_w8a8_block,torch.float16,11,256,7168,256,8,128,128,ck,256,496.3151
gfx936,int8_w8a8_block,torch.float16,12,256,7168,256,8,128,128,ck,256,517.557
gfx936,int8_w8a8_block,torch.float16,13,256,7168,256,8,128,128,ck,256,710.8428
gfx936,int8_w8a8_block,torch.float16,14,256,7168,256,8,128,128,ck,256,719.7744
gfx936,int8_w8a8_block,torch.float16,15,256,7168,256,8,128,128,ck,256,708.893
gfx936,int8_w8a8_block,torch.float16,16,256,7168,256,8,128,128,ck,256,736.713
gfx936,int8_w8a8_block,torch.float16,17,256,7168,256,8,128,128,ck,256,723.91
gfx936,int8_w8a8_block,torch.float16,18,256,7168,256,8,128,128,ck,256,731.7118
gfx936,int8_w8a8_block,torch.float16,19,256,7168,256,8,128,128,ck,256,921.5805
gfx936,int8_w8a8_block,torch.float16,20,256,7168,256,8,128,128,ck,256,931.757
gfx936,int8_w8a8_block,torch.float16,21,256,7168,256,8,128,128,ck,256,925.7805
gfx936,int8_w8a8_block,torch.float16,22,256,7168,256,8,128,128,ck,256,933.0314
gfx936,int8_w8a8_block,torch.float16,23,256,7168,256,8,128,128,ck,256,946.4677
gfx936,int8_w8a8_block,torch.float16,24,256,7168,256,8,128,128,ck,256,966.1805
gfx936,int8_w8a8_block,torch.float16,25,256,7168,256,8,128,128,ck,256,957.8053
gfx936,int8_w8a8_block,torch.float16,26,256,7168,256,8,128,128,ck,256,994.1703
gfx936,int8_w8a8_block,torch.float16,27,256,7168,256,8,128,128,ck,256,1027.6255
gfx936,int8_w8a8_block,torch.float16,28,256,7168,256,8,128,128,ck,256,986.7748
gfx936,int8_w8a8_block,torch.float16,29,256,7168,256,8,128,128,ck,256,1012.4916
gfx936,int8_w8a8_block,torch.float16,30,256,7168,256,8,128,128,ck,256,1031.3195
gfx936,int8_w8a8_block,torch.float16,31,256,7168,256,8,128,128,ck,256,1210.7522
gfx936,int8_w8a8_block,torch.float16,32,256,7168,256,8,128,128,ck,256,1213.3539
gfx936,int8_w8a8_block,torch.float16,34,256,7168,256,8,128,128,ck,256,1217.8463
gfx936,int8_w8a8_block,torch.float16,36,256,7168,256,8,128,128,ck,256,1229.2845
gfx936,int8_w8a8_block,torch.float16,40,256,7168,256,8,128,128,ck,256,1246.6734
gfx936,int8_w8a8_block,torch.float16,44,256,7168,256,8,128,128,ck,256,1268.3449
gfx936,int8_w8a8_block,torch.float16,48,256,7168,256,8,128,128,ck,256,1448.0706
gfx936,int8_w8a8_block,torch.float16,56,256,7168,256,8,128,128,ck,256,1448.6066
gfx936,int8_w8a8_block,torch.float16,64,256,7168,256,8,128,128,ck,256,1533.3137
gfx936,int8_w8a8_block,torch.float16,68,256,7168,256,8,128,128,ck,256,1533.2071
gfx936,int8_w8a8_block,torch.float16,72,256,7168,256,8,128,128,ck,256,1543.6504
gfx936,int8_w8a8_block,torch.float16,80,256,7168,256,8,128,128,ck,256,1557.7885
gfx936,int8_w8a8_block,torch.float16,88,256,7168,256,8,128,128,ck,256,1569.5416
gfx936,int8_w8a8_block,torch.float16,96,256,7168,256,8,128,128,ck,256,1748.5222
gfx936,int8_w8a8_block,torch.float16,104,256,7168,256,8,128,128,ck,256,1780.6233
gfx936,int8_w8a8_block,torch.float16,112,256,7168,256,8,128,128,ck,256,1788.6033
gfx936,int8_w8a8_block,torch.float16,128,256,7168,256,8,128,128,ck,256,1808.0838
gfx936,int8_w8a8_block,torch.float16,144,256,7168,256,8,128,128,ck,256,1837.7577
gfx936,int8_w8a8_block,torch.float16,160,256,7168,256,8,128,128,ck,256,1852.4556
gfx936,int8_w8a8_block,torch.float16,192,256,7168,256,8,128,128,ck,256,1883.3793
gfx936,int8_w8a8_block,torch.float16,224,256,7168,256,8,128,128,ck,256,1923.4614
gfx936,int8_w8a8_block,torch.float16,256,256,7168,256,8,128,128,ck,256,1958.9242
gfx936,int8_w8a8_block,torch.float16,320,256,7168,256,8,128,128,ck,256,2031.9791
gfx936,int8_w8a8_block,torch.float16,384,256,7168,256,8,128,128,ck,256,2129.4956
gfx936,int8_w8a8_block,torch.float16,448,256,7168,256,8,128,128,ck,256,2504.2611
gfx936,int8_w8a8_block,torch.float16,512,256,7168,256,8,128,128,ck,256,2790.1238
gfx936,int8_w8a8_block,torch.float16,768,256,7168,256,8,128,128,ck,1073741824,3529.9268
gfx936,int8_w8a8_block,torch.float16,1024,256,7168,256,8,128,128,ck,256,4366.3522
gfx936,int8_w8a8_block,torch.float16,1024,256,7168,256,8,128,128,ck,256,4391.4689
gfx936,int8_w8a8_block,torch.float16,1152,256,7168,256,8,128,128,ck,256,4889.4998
gfx936,int8_w8a8_block,torch.float16,1,2304,7168,256,8,128,128,ck,256,560.4577
gfx936,int8_w8a8_block,torch.float16,2,2304,7168,256,8,128,128,ck,1073741824,922.3036
gfx936,int8_w8a8_block,torch.float16,3,2304,7168,256,8,128,128,ck,256,1211.5382
gfx936,int8_w8a8_block,torch.float16,4,2304,7168,256,8,128,128,ck,256,1609.7351
gfx936,int8_w8a8_block,torch.float16,5,2304,7168,256,8,128,128,ck,256,2033.1529
gfx936,int8_w8a8_block,torch.float16,6,2304,7168,256,8,128,128,ck,256,2095.9823
gfx936,int8_w8a8_block,torch.float16,7,2304,7168,256,8,128,128,ck,256,2525.8506
gfx936,int8_w8a8_block,torch.float16,8,2304,7168,256,8,128,128,ck,256,3121.4785
gfx936,int8_w8a8_block,torch.float16,9,2304,7168,256,8,128,128,ck,256,3602.7825
gfx936,int8_w8a8_block,torch.float16,10,2304,7168,256,8,128,128,ck,256,3449.7641
gfx936,int8_w8a8_block,torch.float16,11,2304,7168,256,8,128,128,ck,256,4044.0698
gfx936,int8_w8a8_block,torch.float16,12,2304,7168,256,8,128,128,ck,256,4134.7811
gfx936,int8_w8a8_block,torch.float16,13,2304,7168,256,8,128,128,ck,256,4802.8215
gfx936,int8_w8a8_block,torch.float16,14,2304,7168,256,8,128,128,ck,256,4719.2344
gfx936,int8_w8a8_block,torch.float16,15,2304,7168,256,8,128,128,ck,256,4980.6316
gfx936,int8_w8a8_block,torch.float16,16,2304,7168,256,8,128,128,ck,256,5250.5172
gfx936,int8_w8a8_block,torch.float16,17,2304,7168,256,8,128,128,ck,256,5673.4436
gfx936,int8_w8a8_block,torch.float16,18,2304,7168,256,8,128,128,ck,256,5719.9952
gfx936,int8_w8a8_block,torch.float16,19,2304,7168,256,8,128,128,ck,256,6126.3857
gfx936,int8_w8a8_block,torch.float16,20,2304,7168,256,8,128,128,ck,256,6389.1556
gfx936,int8_w8a8_block,torch.float16,21,2304,7168,256,8,128,128,ck,256,6198.0488
gfx936,int8_w8a8_block,torch.float16,22,2304,7168,256,8,128,128,ck,256,6704.2199
gfx936,int8_w8a8_block,torch.float16,23,2304,7168,256,8,128,128,ck,256,6957.5245
gfx936,int8_w8a8_block,torch.float16,24,2304,7168,256,8,128,128,ck,256,6653.9634
gfx936,int8_w8a8_block,torch.float16,25,2304,7168,256,8,128,128,ck,256,7419.4212
gfx936,int8_w8a8_block,torch.float16,26,2304,7168,256,8,128,128,ck,256,7740.734
gfx936,int8_w8a8_block,torch.float16,27,2304,7168,256,8,128,128,ck,256,7583.9258
gfx936,int8_w8a8_block,torch.float16,28,2304,7168,256,8,128,128,ck,256,8067.6719
gfx936,int8_w8a8_block,torch.float16,29,2304,7168,256,8,128,128,ck,256,7827.3022
gfx936,int8_w8a8_block,torch.float16,30,2304,7168,256,8,128,128,ck,256,8226.0128
gfx936,int8_w8a8_block,torch.float16,31,2304,7168,256,8,128,128,ck,256,8765.0934
gfx936,int8_w8a8_block,torch.float16,32,2304,7168,256,8,128,128,ck,256,8287.4442
gfx936,int8_w8a8_block,torch.float16,34,2304,7168,256,8,128,128,ck,256,9189.8303
gfx936,int8_w8a8_block,torch.float16,36,2304,7168,256,8,128,128,ck,256,9010.0497
gfx936,int8_w8a8_block,torch.float16,40,2304,7168,256,8,128,128,ck,256,9991.6976
gfx936,int8_w8a8_block,torch.float16,44,2304,7168,256,8,128,128,ck,256,10406.3407
gfx936,int8_w8a8_block,torch.float16,48,2304,7168,256,8,128,128,ck,256,10457.3132
gfx936,int8_w8a8_block,torch.float16,56,2304,7168,256,8,128,128,ck,256,10918.0446
gfx936,int8_w8a8_block,torch.float16,64,2304,7168,256,8,128,128,ck,256,12263.38
gfx936,int8_w8a8_block,torch.float16,68,2304,7168,256,8,128,128,ck,256,12388.0282
gfx936,int8_w8a8_block,torch.float16,72,2304,7168,256,8,128,128,ck,256,12684.0274
gfx936,int8_w8a8_block,torch.float16,80,2304,7168,256,8,128,128,ck,256,12786.4017
gfx936,int8_w8a8_block,torch.float16,88,2304,7168,256,8,128,128,ck,256,12816.288
gfx936,int8_w8a8_block,torch.float16,96,2304,7168,256,8,128,128,ck,256,13388.8676
gfx936,int8_w8a8_block,torch.float16,104,2304,7168,256,8,128,128,ck,256,13664.4714
gfx936,int8_w8a8_block,torch.float16,112,2304,7168,256,8,128,128,ck,1073741824,13843.4207
gfx936,int8_w8a8_block,torch.float16,128,2304,7168,256,8,128,128,ck,1073741824,14062.0561
gfx936,int8_w8a8_block,torch.float16,144,2304,7168,256,8,128,128,ck,1073741824,13992.1954
gfx936,int8_w8a8_block,torch.float16,160,2304,7168,256,8,128,128,ck,1073741824,14250.3336
gfx936,int8_w8a8_block,torch.float16,192,2304,7168,256,8,128,128,ck,1073741824,14385.9878
gfx936,int8_w8a8_block,torch.float16,224,2304,7168,256,8,128,128,ck,1073741824,14440.0596
gfx936,int8_w8a8_block,torch.float16,256,2304,7168,256,8,128,128,ck,1073741824,14530.2234
gfx936,int8_w8a8_block,torch.float16,320,2304,7168,256,8,128,128,ck,1073741824,14700.4377
gfx936,int8_w8a8_block,torch.float16,384,2304,7168,256,8,128,128,ck,1073741824,14808.9845
gfx936,int8_w8a8_block,torch.float16,448,2304,7168,256,8,128,128,ck,1073741824,14983.8729
gfx936,int8_w8a8_block,torch.float16,512,2304,7168,256,8,128,128,ck,1073741824,15144.0743
gfx936,int8_w8a8_block,torch.float16,768,2304,7168,256,8,128,128,ck,1073741824,16370.4801
gfx936,int8_w8a8_block,torch.float16,1024,2304,7168,256,8,128,128,ck,1073741824,22165.7068
gfx936,int8_w8a8_block,torch.float16,1024,2304,7168,256,8,128,128,ck,1073741824,22200.5112
gfx936,int8_w8a8_block,torch.float16,1152,2304,7168,256,8,128,128,ck,1073741824,26188.6939
gfx936,int8_w8a8_block,torch.float16,1,1536,4096,128,8,128,128,ck,256,291.658
gfx936,int8_w8a8_block,torch.float16,2,1536,4096,128,8,128,128,ck,256,409.0957
gfx936,int8_w8a8_block,torch.float16,3,1536,4096,128,8,128,128,ck,256,545.9305
gfx936,int8_w8a8_block,torch.float16,4,1536,4096,128,8,128,128,ck,256,680.0244
gfx936,int8_w8a8_block,torch.float16,5,1536,4096,128,8,128,128,ck,256,824.0053
gfx936,int8_w8a8_block,torch.float16,6,1536,4096,128,8,128,128,ck,256,961.9745
gfx936,int8_w8a8_block,torch.float16,7,1536,4096,128,8,128,128,ck,256,993.5245
gfx936,int8_w8a8_block,torch.float16,8,1536,4096,128,8,128,128,ck,256,1152.4533
gfx936,int8_w8a8_block,torch.float16,9,1536,4096,128,8,128,128,ck,256,1160.7529
gfx936,int8_w8a8_block,torch.float16,10,1536,4096,128,8,128,128,ck,256,1266.7067
gfx936,int8_w8a8_block,torch.float16,11,1536,4096,128,8,128,128,ck,256,1295.2701
gfx936,int8_w8a8_block,torch.float16,12,1536,4096,128,8,128,128,ck,256,1440.103
gfx936,int8_w8a8_block,torch.float16,13,1536,4096,128,8,128,128,ck,256,1697.6591
gfx936,int8_w8a8_block,torch.float16,14,1536,4096,128,8,128,128,ck,256,1745.6223
gfx936,int8_w8a8_block,torch.float16,15,1536,4096,128,8,128,128,ck,256,1710.5023
gfx936,int8_w8a8_block,torch.float16,16,1536,4096,128,8,128,128,ck,256,1852.1887
gfx936,int8_w8a8_block,torch.float16,17,1536,4096,128,8,128,128,ck,256,1831.4358
gfx936,int8_w8a8_block,torch.float16,18,1536,4096,128,8,128,128,ck,256,1838.3458
gfx936,int8_w8a8_block,torch.float16,19,1536,4096,128,8,128,128,ck,256,2021.7323000000001
gfx936,int8_w8a8_block,torch.float16,20,1536,4096,128,8,128,128,ck,256,2118.7607
gfx936,int8_w8a8_block,torch.float16,21,1536,4096,128,8,128,128,ck,256,2139.1852
gfx936,int8_w8a8_block,torch.float16,22,1536,4096,128,8,128,128,ck,256,2195.0414
gfx936,int8_w8a8_block,torch.float16,23,1536,4096,128,8,128,128,ck,256,2192.328
gfx936,int8_w8a8_block,torch.float16,24,1536,4096,128,8,128,128,ck,256,2440.6246
gfx936,int8_w8a8_block,torch.float16,25,1536,4096,128,8,128,128,ck,256,2200.6248
gfx936,int8_w8a8_block,torch.float16,26,1536,4096,128,8,128,128,ck,256,2218.4566
gfx936,int8_w8a8_block,torch.float16,27,1536,4096,128,8,128,128,ck,256,2453.0879
gfx936,int8_w8a8_block,torch.float16,28,1536,4096,128,8,128,128,ck,256,2350.3684
gfx936,int8_w8a8_block,torch.float16,29,1536,4096,128,8,128,128,ck,256,2476.5557
gfx936,int8_w8a8_block,torch.float16,30,1536,4096,128,8,128,128,ck,256,2487.2788
gfx936,int8_w8a8_block,torch.float16,31,1536,4096,128,8,128,128,ck,256,2440.2945
gfx936,int8_w8a8_block,torch.float16,32,1536,4096,128,8,128,128,ck,256,2470.9654
gfx936,int8_w8a8_block,torch.float16,34,1536,4096,128,8,128,128,ck,256,2640.585
gfx936,int8_w8a8_block,torch.float16,36,1536,4096,128,8,128,128,ck,256,2472.0041
gfx936,int8_w8a8_block,torch.float16,40,1536,4096,128,8,128,128,ck,256,2689.1661
gfx936,int8_w8a8_block,torch.float16,44,1536,4096,128,8,128,128,ck,256,2760.7251
gfx936,int8_w8a8_block,torch.float16,48,1536,4096,128,8,128,128,ck,256,2789.2965
gfx936,int8_w8a8_block,torch.float16,56,1536,4096,128,8,128,128,ck,256,2911.2359
gfx936,int8_w8a8_block,torch.float16,64,1536,4096,128,8,128,128,ck,256,2866.2659
gfx936,int8_w8a8_block,torch.float16,68,1536,4096,128,8,128,128,ck,256,2932.9077
gfx936,int8_w8a8_block,torch.float16,72,1536,4096,128,8,128,128,ck,256,2849.2279
gfx936,int8_w8a8_block,torch.float16,80,1536,4096,128,8,128,128,ck,256,2902.7918
gfx936,int8_w8a8_block,torch.float16,88,1536,4096,128,8,128,128,ck,1073741824,2991.5017
gfx936,int8_w8a8_block,torch.float16,96,1536,4096,128,8,128,128,ck,1073741824,3009.2322
gfx936,int8_w8a8_block,torch.float16,104,1536,4096,128,8,128,128,ck,1073741824,3010.9726
gfx936,int8_w8a8_block,torch.float16,112,1536,4096,128,8,128,128,ck,1073741824,3021.8461
gfx936,int8_w8a8_block,torch.float16,128,1536,4096,128,8,128,128,ck,1073741824,3042.2445
gfx936,int8_w8a8_block,torch.float16,144,1536,4096,128,8,128,128,ck,1073741824,3091.5212
gfx936,int8_w8a8_block,torch.float16,160,1536,4096,128,8,128,128,ck,1073741824,3116.5366
gfx936,int8_w8a8_block,torch.float16,192,1536,4096,128,8,128,128,ck,1073741824,3152.8336
gfx936,int8_w8a8_block,torch.float16,224,1536,4096,128,8,128,128,ck,1073741824,3192.8889
gfx936,int8_w8a8_block,torch.float16,256,1536,4096,128,8,128,128,ck,1073741824,3223.7425
gfx936,int8_w8a8_block,torch.float16,320,1536,4096,128,8,128,128,ck,1073741824,3282.5084
gfx936,int8_w8a8_block,torch.float16,384,1536,4096,128,8,128,128,ck,1073741824,3431.9543
gfx936,int8_w8a8_block,torch.float16,448,1536,4096,128,8,128,128,ck,1073741824,3855.0492
gfx936,int8_w8a8_block,torch.float16,512,1536,4096,128,8,128,128,ck,1073741824,4436.9874
gfx936,int8_w8a8_block,torch.float16,768,1536,4096,128,8,128,128,ck,1073741824,5567.0914
gfx936,int8_w8a8_block,torch.float16,1024,1536,4096,128,8,128,128,ck,1073741824,6996.0452
gfx936,int8_w8a8_block,torch.float16,1024,1536,4096,128,8,128,128,ck,1073741824,6939.3137
gfx936,int8_w8a8_block,torch.float16,1152,1536,4096,128,8,128,128,ck,1073741824,7784.7617
aiter/configs/ck_tune/untuned_fmoe_int8_w8a8_group.csv 0 → 100644
View file @ 85762c1a
quant_type,indtype,token,inter_dim,model_dim,expert,topk,q_size_n,q_size_k
int8_w8a8_block,torch.float16,1,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,2,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,3,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,4,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,5,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,6,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,7,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,8,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,9,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,10,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,11,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,12,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,13,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,14,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,15,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,16,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,17,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,18,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,19,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,20,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,21,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,22,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,23,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,24,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,25,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,26,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,27,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,28,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,29,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,30,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,31,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,32,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,34,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,36,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,40,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,44,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,48,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,56,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,64,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,68,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,72,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,80,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,88,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,96,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,104,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,112,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,128,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,144,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,160,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,192,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,224,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,256,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,320,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,384,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,448,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,512,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,768,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,1024,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,1024,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,1152,256,7168,256,8,128,128
int8_w8a8_block,torch.float16,1,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,2,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,3,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,4,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,5,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,6,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,7,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,8,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,9,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,10,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,11,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,12,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,13,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,14,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,15,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,16,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,17,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,18,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,19,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,20,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,21,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,22,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,23,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,24,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,25,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,26,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,27,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,28,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,29,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,30,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,31,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,32,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,34,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,36,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,40,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,44,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,48,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,56,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,64,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,68,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,72,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,80,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,88,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,96,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,104,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,112,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,128,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,144,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,160,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,192,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,224,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,256,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,320,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,384,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,448,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,512,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,768,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,1024,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,1024,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,1152,2304,7168,256,8,128,128
int8_w8a8_block,torch.float16,1,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,2,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,3,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,4,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,5,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,6,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,7,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,8,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,9,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,10,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,11,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,12,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,13,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,14,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,15,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,16,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,17,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,18,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,19,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,20,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,21,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,22,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,23,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,24,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,25,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,26,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,27,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,28,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,29,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,30,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,31,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,32,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,34,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,36,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,40,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,44,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,48,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,56,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,64,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,68,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,72,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,80,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,88,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,96,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,104,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,112,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,128,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,144,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,160,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,192,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,224,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,256,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,320,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,384,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,448,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,512,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,768,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,1024,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,1024,1536,4096,128,8,128,128
int8_w8a8_block,torch.float16,1152,1536,4096,128,8,128,128
aiter/configs/tuned_fmoe.csv 0 → 100644
View file @ 85762c1a
token,model_dim,inter_dim,expert,topk,act_type,dtype,q_dtype_a,q_dtype_w,q_type,use_g1u1,doweight_stage1,block_m,ksplit,us,tag,err
256,6144,4096,8,2,ActivationType.Silu,torch.bfloat16,torch.float8_e4m3fnuz,torch.float8_e4m3fnuz,QuantType.per_128x128,1,0,32,0,270.8542,fmoe_stage1_bf16_pertokenFp8_blockscale_g1u1_32x512_pf2,1.6%
512,6144,4096,8,2,ActivationType.Silu,torch.bfloat16,torch.int8,torch.int8,QuantType.per_Tensor,1,0,64,0,346.93691000000007,fmoe_stage1_bf16_pertokenInt8_g1u1_64x256_pf3,0.0%
512,6144,4096,8,2,ActivationType.Silu,torch.bfloat16,torch.int8,torch.int8,QuantType.per_Token,1,0,64,0,340.65007,fmoe_stage1_bf16_pertokenInt8_g1u1_64x256_pf3,0.0%
512,6144,4096,8,2,ActivationType.Silu,torch.bfloat16,torch.float8_e4m3fnuz,torch.float8_e4m3fnuz,QuantType.per_Token,1,0,64,0,336.42205,fmoe_stage1_bf16_pertokenFp8_g1u1_64x256_pf3,0.0%
512,6144,4096,8,2,ActivationType.Silu,torch.bfloat16,torch.float8_e4m3fnuz,torch.int4,QuantType.per_Tensor,1,0,128,0,699.5049699999998,ck_128,0.1%
512,6144,4096,8,2,ActivationType.Gelu,torch.bfloat16,torch.int8,torch.int8,QuantType.per_Tensor,1,0,64,0,347.0561,fmoe_stage1_bf16_pertokenInt8_g1u1_64x256_pf3,0.0%
512,6144,4096,8,2,ActivationType.Gelu,torch.bfloat16,torch.int8,torch.int8,QuantType.per_Token,1,0,64,0,334.47364,fmoe_stage1_bf16_pertokenInt8_g1u1_64x256_pf3,0.0%
512,6144,4096,8,2,ActivationType.Gelu,torch.bfloat16,torch.float8_e4m3fnuz,torch.float8_e4m3fnuz,QuantType.per_Token,1,0,64,0,357.0689999999998,fmoe_stage1_bf16_pertokenFp8_g1u1_64x128_2tg_pf3,0.0%
512,6144,4096,8,2,ActivationType.Gelu,torch.bfloat16,torch.float8_e4m3fnuz,torch.int4,QuantType.per_Tensor,1,0,128,0,837.6014699999998,ck_128,0.0%
4,2304,1536,8,2,ActivationType.Gelu,torch.bfloat16,torch.float8_e4m3fnuz,torch.float8_e4m3fnuz,QuantType.per_Token,1,0,32,0,18.364460000000005,fmoe_stage1_bf16_pertokenFp8_g1u1_32x64_4tg_pf3,0.0%
4,2304,1536,8,2,ActivationType.Gelu,torch.bfloat16,torch.float8_e4m3fnuz,torch.float8_e4m3fnuz,QuantType.per_Token,1,1,32,0,19.47728,fmoe_stage1_bf16_pertokenFp8_doweight_g1u1_32x64_4tg_pf3,0.0%
aiter/configs/tuned_fmoe_asm.csv 0 → 100644
View file @ 85762c1a
arch,quant_type,indtype,token,inter_dim,model_dim,expert,topk,q_size_n,q_size_k,sol_type,sol_id,time_us
gfx938,no_quant,torch.float16,1,352,4096,128,8,0,0,asm,10008+20000,76.1135
gfx938,no_quant,torch.float16,2,352,4096,128,8,0,0,asm,10011+20002,116.9911
gfx938,no_quant,torch.float16,4,352,4096,128,8,0,0,asm,10002+20000,190.2544
gfx938,no_quant,torch.float16,6,352,4096,128,8,0,0,asm,10011+20000,255.6735
gfx938,no_quant,torch.float16,8,352,4096,128,8,0,0,asm,10002+20000,327.9644
gfx938,no_quant,torch.float16,10,352,4096,128,8,0,0,asm,10002+20000,381.3467
gfx938,no_quant,torch.float16,12,352,4096,128,8,0,0,asm,10002+20000,425.2932
gfx938,no_quant,torch.float16,14,352,4096,128,8,0,0,asm,10002+20000,466.8067
gfx938,no_quant,torch.float16,16,352,4096,128,8,0,0,asm,10002+20000,502.9911
gfx938,no_quant,torch.float16,20,352,4096,128,8,0,0,asm,10002+20000,559.3775
gfx938,no_quant,torch.float16,24,352,4096,128,8,0,0,asm,10002+20000,579.1763
gfx938,no_quant,torch.float16,28,352,4096,128,8,0,0,asm,10002+20000,598.1905
gfx938,no_quant,torch.float16,32,352,4096,128,8,0,0,asm,10002+20000,617.1032
gfx938,no_quant,torch.float16,36,352,4096,128,8,0,0,asm,10002+20000,634.1831
gfx938,no_quant,torch.float16,40,352,4096,128,8,0,0,asm,10002+20000,650.6875
gfx938,no_quant,torch.float16,44,352,4096,128,8,0,0,asm,10002+20000,673.797
gfx938,no_quant,torch.float16,48,352,4096,128,8,0,0,asm,10002+20000,688.9969
gfx938,no_quant,torch.float16,56,352,4096,128,8,0,0,asm,10002+20000,690.6337
gfx938,no_quant,torch.float16,64,352,4096,128,8,0,0,asm,10002+20000,705.25
gfx938,no_quant,torch.float16,68,352,4096,128,8,0,0,asm,10002+20000,692.6071
gfx938,no_quant,torch.float16,72,352,4096,128,8,0,0,asm,10002+20000,693.4728
gfx938,no_quant,torch.float16,80,352,4096,128,8,0,0,asm,10002+20000,693.0667
gfx938,no_quant,torch.float16,88,352,4096,128,8,0,0,asm,10002+20000,698.6174
gfx938,no_quant,torch.float16,96,352,4096,128,8,0,0,asm,10002+20000,701.0585
gfx938,no_quant,torch.float16,104,352,4096,128,8,0,0,asm,10002+20000,700.9929
gfx938,no_quant,torch.float16,112,352,4096,128,8,0,0,asm,10002+20000,702.7078
gfx938,no_quant,torch.float16,128,352,4096,128,8,0,0,asm,10002+20000,706.9581
gfx938,no_quant,torch.float16,144,352,4096,128,8,0,0,asm,10002+20000,729.0544
gfx938,no_quant,torch.float16,160,352,4096,128,8,0,0,asm,10002+20000,722.2728
gfx938,no_quant,torch.float16,192,352,4096,128,8,0,0,asm,10011+20000,742.4902
gfx938,no_quant,torch.float16,224,352,4096,128,8,0,0,asm,11004+21001,774.5392
gfx938,no_quant,torch.float16,256,352,4096,128,8,0,0,asm,11004+21001,780.4756
gfx938,no_quant,torch.float16,320,352,4096,128,8,0,0,asm,11004+21001,796.5904
gfx938,no_quant,torch.float16,384,352,4096,128,8,0,0,asm,11005+21001,804.2211
gfx938,no_quant,torch.float16,448,352,4096,128,8,0,0,asm,12005+22001,833.0702
gfx938,no_quant,torch.float16,512,352,4096,128,8,0,0,asm,12005+22001,847.1173
gfx938,no_quant,torch.float16,576,352,4096,128,8,0,0,asm,12005+22001,857.4557
gfx938,no_quant,torch.float16,640,352,4096,128,8,0,0,asm,12005+22001,867.1542
gfx938,no_quant,torch.float16,704,352,4096,128,8,0,0,asm,12005+22001,878.2925
gfx938,no_quant,torch.float16,768,352,4096,128,8,0,0,asm,12005+22001,890.8995
gfx938,no_quant,torch.float16,832,352,4096,128,8,0,0,asm,12005+22001,900.5198
gfx938,no_quant,torch.float16,896,352,4096,128,8,0,0,asm,12001+22001,958.5009
gfx938,no_quant,torch.float16,960,352,4096,128,8,0,0,asm,12003+22001,985.6514
gfx938,no_quant,torch.float16,1024,352,4096,128,8,0,0,asm,13001+23001,1062.3093
gfx938,no_quant,torch.float16,1152,352,4096,128,8,0,0,asm,13001+23001,1079.7408
gfx938,no_quant,torch.float16,1280,352,4096,128,8,0,0,asm,13001+23001,1090.4801
gfx938,no_quant,torch.float16,1408,352,4096,128,8,0,0,asm,13001+23001,1110.9976
gfx938,no_quant,torch.float16,1536,352,4096,128,8,0,0,asm,13001+23001,1123.6784
gfx938,no_quant,torch.float16,1664,352,4096,128,8,0,0,asm,13001+23001,1141.3393
gfx938,no_quant,torch.float16,1792,352,4096,128,8,0,0,asm,13001+23001,1163.7203
gfx938,no_quant,torch.float16,1920,352,4096,128,8,0,0,asm,13001+23001,1304.7715
gfx938,no_quant,torch.float16,2048,352,4096,128,8,0,0,asm,13001+23001,1584.7923
gfx938,no_quant,torch.float16,2304,352,4096,128,8,0,0,asm,13001+23001,1943.0033
gfx938,no_quant,torch.float16,2560,352,4096,128,8,0,0,asm,13001+23001,1988.7219
gfx938,no_quant,torch.float16,2816,352,4096,128,8,0,0,asm,13001+23001,2026.4655
gfx938,no_quant,torch.float16,3072,352,4096,128,8,0,0,asm,13001+23001,2056.3481
gfx938,no_quant,torch.float16,3328,352,4096,128,8,0,0,asm,13001+23001,2081.0787
gfx938,no_quant,torch.float16,3584,352,4096,128,8,0,0,asm,13001+23001,2113.5351
gfx938,no_quant,torch.float16,3840,352,4096,128,8,0,0,asm,13001+23001,2251.9895
gfx938,no_quant,torch.float16,4096,352,4096,128,8,0,0,asm,13001+23001,2568.3675
gfx938,no_quant,torch.float16,5120,352,4096,128,8,0,0,asm,13001+23001,3066.6487
gfx938,no_quant,torch.float16,6144,352,4096,128,8,0,0,asm,13001+23001,3581.1465
gfx938,no_quant,torch.float16,7168,352,4096,128,8,0,0,asm,13001+23001,4122.1716
gfx938,no_quant,torch.float16,7680,352,4096,128,8,0,0,asm,13001+23001,4222.0785
gfx938,no_quant,torch.float16,8192,352,4096,128,8,0,0,asm,13001+23001,4607.7508
gfx938,no_quant,torch.float16,10240,352,4096,128,8,0,0,asm,13001+23001,5676.0447
gfx938,no_quant,torch.float16,12288,352,4096,128,8,0,0,asm,13001+23001,6679.6272
gfx938,no_quant,torch.float16,14336,352,4096,128,8,0,0,asm,13001+23001,7789.8272
gfx938,no_quant,torch.float16,16384,352,4096,128,8,0,0,asm,13001+23001,8726.181
gfx938,no_quant,torch.float16,17408,352,4096,128,8,0,0,asm,13001+23001,9263.6738
gfx938,no_quant,torch.float16,24576,352,4096,128,8,0,0,asm,13001+23001,12830.282
gfx938,no_quant,torch.float16,32768,352,4096,128,8,0,0,asm,13001+23001,17031.4351
gfx938,no_quant,torch.float16,40960,352,4096,128,8,0,0,asm,13001+23001,21120.0199
gfx938,no_quant,torch.float16,49152,352,4096,128,8,0,0,asm,13001+23001,25184.3324
gfx938,no_quant,torch.float16,57344,352,4096,128,8,0,0,asm,13001+23001,29341.3454
gfx938,no_quant,torch.float16,65536,352,4096,128,8,0,0,asm,13001+23001,33437.6425
gfx938,no_quant,torch.float16,1,352,4096,129,9,0,0,asm,10008+20000,79.718
gfx938,no_quant,torch.float16,2,352,4096,129,9,0,0,asm,10011+20002,118.7885
gfx938,no_quant,torch.float16,4,352,4096,129,9,0,0,asm,10002+20000,196.5523
gfx938,no_quant,torch.float16,6,352,4096,129,9,0,0,asm,10011+20000,270.1234
gfx938,no_quant,torch.float16,8,352,4096,129,9,0,0,asm,10002+20000,322.2257
gfx938,no_quant,torch.float16,10,352,4096,129,9,0,0,asm,10002+20000,379.8174
gfx938,no_quant,torch.float16,12,352,4096,129,9,0,0,asm,10002+20000,416.2644
gfx938,no_quant,torch.float16,14,352,4096,129,9,0,0,asm,10002+20000,457.8191
gfx938,no_quant,torch.float16,16,352,4096,129,9,0,0,asm,10002+20000,493.3677
gfx938,no_quant,torch.float16,20,352,4096,129,9,0,0,asm,10002+20000,562.0455
gfx938,no_quant,torch.float16,24,352,4096,129,9,0,0,asm,10002+20000,609.2494
gfx938,no_quant,torch.float16,28,352,4096,129,9,0,0,asm,10002+20000,639.2872
gfx938,no_quant,torch.float16,32,352,4096,129,9,0,0,asm,10002+20000,650.7579
gfx938,no_quant,torch.float16,36,352,4096,129,9,0,0,asm,10002+20000,680.4902
gfx938,no_quant,torch.float16,40,352,4096,129,9,0,0,asm,10002+20000,686.082
gfx938,no_quant,torch.float16,48,352,4096,129,9,0,0,asm,10002+20000,712.4491
gfx938,no_quant,torch.float16,56,352,4096,129,9,0,0,asm,10002+20000,710.9639
gfx938,no_quant,torch.float16,64,352,4096,129,9,0,0,asm,10002+20000,730.5607
gfx938,no_quant,torch.float16,72,352,4096,129,9,0,0,asm,10002+20000,718.7957
gfx938,no_quant,torch.float16,80,352,4096,129,9,0,0,asm,10002+20000,720.0142
gfx938,no_quant,torch.float16,88,352,4096,129,9,0,0,asm,10002+20000,722.9229
gfx938,no_quant,torch.float16,96,352,4096,129,9,0,0,asm,10002+20000,723.0624
gfx938,no_quant,torch.float16,104,352,4096,129,9,0,0,asm,10002+20000,724.0839
gfx938,no_quant,torch.float16,112,352,4096,129,9,0,0,asm,10002+20000,735.3413
gfx938,no_quant,torch.float16,128,352,4096,129,9,0,0,asm,10002+20000,737.2243
gfx938,no_quant,torch.float16,144,352,4096,129,9,0,0,asm,10002+20000,754.1186
gfx938,no_quant,torch.float16,160,352,4096,129,9,0,0,asm,10002+20000,753.3679
gfx938,no_quant,torch.float16,192,352,4096,129,9,0,0,asm,10002+20000,788.8426
gfx938,no_quant,torch.float16,224,352,4096,129,9,0,0,asm,11004+21001,801.5606
gfx938,no_quant,torch.float16,256,352,4096,129,9,0,0,asm,11004+21001,807.7226
gfx938,no_quant,torch.float16,320,352,4096,129,9,0,0,asm,11004+21001,826.3646
gfx938,no_quant,torch.float16,384,352,4096,129,9,0,0,asm,12005+22001,854.2537
gfx938,no_quant,torch.float16,448,352,4096,129,9,0,0,asm,12005+22001,864.8998
gfx938,no_quant,torch.float16,512,352,4096,129,9,0,0,asm,12005+22001,883.4228
gfx938,no_quant,torch.float16,576,352,4096,129,9,0,0,asm,12005+22001,895.6977
gfx938,no_quant,torch.float16,640,352,4096,129,9,0,0,asm,12001+22001,908.3787
gfx938,no_quant,torch.float16,768,352,4096,129,9,0,0,asm,12001+22001,915.4966
gfx938,no_quant,torch.float16,960,352,4096,129,9,0,0,asm,13001+23001,1079.7907
gfx938,no_quant,torch.float16,1024,352,4096,129,9,0,0,asm,13001+23001,1088.9094
gfx938,no_quant,torch.float16,1280,352,4096,129,9,0,0,asm,13001+23001,1129.3092
gfx938,no_quant,torch.float16,1536,352,4096,129,9,0,0,asm,13001+23001,1169.2723
gfx938,no_quant,torch.float16,1920,352,4096,129,9,0,0,asm,13001+23001,1764.185
gfx938,no_quant,torch.float16,2048,352,4096,129,9,0,0,asm,13001+23001,1934.3223
gfx938,no_quant,torch.float16,2304,352,4096,129,9,0,0,asm,13001+23001,2006.923
gfx938,no_quant,torch.float16,2560,352,4096,129,9,0,0,asm,13001+23001,2035.540
gfx938,no_quant,torch.float16,2816,352,4096,129,9,0,0,asm,13001+23001,2068.5062
gfx938,no_quant,torch.float16,3072,352,4096,129,9,0,0,asm,13001+23001,2102.0827
gfx938,no_quant,torch.float16,3584,352,4096,129,9,0,0,asm,13001+23001,2447.431
gfx938,no_quant,torch.float16,3840,352,4096,129,9,0,0,asm,13001+23001,2805.2093
gfx938,no_quant,torch.float16,4096,352,4096,129,9,0,0,asm,13001+23001,3007.0664
gfx938,no_quant,torch.float16,4608,352,4096,129,9,0,0,asm,13001+23001,3123.0345
gfx938,no_quant,torch.float16,5120,352,4096,129,9,0,0,asm,13001+23001,3246.8705
gfx938,no_quant,torch.float16,5632,352,4096,129,9,0,0,asm,13001+23001,3764.9074
gfx938,no_quant,torch.float16,6144,352,4096,129,9,0,0,asm,13001+23001,4092.9607
gfx938,no_quant,torch.float16,6656,352,4096,129,9,0,0,asm,13001+23001,4175.6811
gfx938,no_quant,torch.float16,7168,352,4096,129,9,0,0,asm,13001+23001,4504.6947
gfx938,no_quant,torch.float16,7680,352,4096,129,9,0,0,asm,13001+23001,5027.1321
gfx938,no_quant,torch.float16,8192,352,4096,129,9,0,0,asm,13001+23001,5169.6112
gfx938,no_quant,torch.float16,10240,352,4096,129,9,0,0,asm,13001+23001,6238.8422
gfx938,no_quant,torch.float16,12288,352,4096,129,9,0,0,asm,13001+23001,7419.1773
gfx938,no_quant,torch.float16,16384,352,4096,129,9,0,0,asm,13001+23001,9753.8878
gfx938,no_quant,torch.float16,24576,352,4096,129,9,0,0,asm,13001+23001,14398.8284
gfx938,no_quant,torch.float16,32768,352,4096,129,9,0,0,asm,13001+23001,19058.7232
gfx938,no_quant,torch.float16,40960,352,4096,129,9,0,0,asm,13001+23001,23722.1115
gfx938,no_quant,torch.float16,49152,352,4096,129,9,0,0,asm,13001+23001,28329.0767
gfx938,no_quant,torch.float16,65536,352,4096,129,9,0,0,asm,13001+23001,37562.0269
gfx936,no_quant,torch.float16,1,256,3072,256,8,0,0,asm,10002+20000,55.456
gfx936,no_quant,torch.float16,2,256,3072,256,8,0,0,asm,10002+20000,86.6223
gfx936,no_quant,torch.float16,4,256,3072,256,8,0,0,asm,10002+20000,155.6412
gfx936,no_quant,torch.float16,6,256,3072,256,8,0,0,asm,10001+20000,212.4495
gfx936,no_quant,torch.float16,8,256,3072,256,8,0,0,asm,10001+20000,258.5125
gfx936,no_quant,torch.float16,12,256,3072,256,8,0,0,asm,10001+20000,349.5272
gfx936,no_quant,torch.float16,16,256,3072,256,8,0,0,asm,10001+20000,424.9797
gfx936,no_quant,torch.float16,24,256,3072,256,8,0,0,asm,10001+20000,569.7428
gfx936,no_quant,torch.float16,32,256,3072,256,8,0,0,asm,10001+20000,698.3014
gfx936,no_quant,torch.float16,36,256,3072,256,8,0,0,asm,10001+20000,731.8899
gfx936,no_quant,torch.float16,48,256,3072,256,8,0,0,asm,10001+20000,823.0646
gfx936,no_quant,torch.float16,56,256,3072,256,8,0,0,asm,10001+20000,869.7846
gfx936,no_quant,torch.float16,64,256,3072,256,8,0,0,asm,10001+20000,902.1634
gfx936,no_quant,torch.float16,72,256,3072,256,8,0,0,asm,10001+20000,939.9486
gfx936,no_quant,torch.float16,80,256,3072,256,8,0,0,asm,10001+20000,966.9634
gfx936,no_quant,torch.float16,88,256,3072,256,8,0,0,asm,10001+20000,994.9044
gfx936,no_quant,torch.float16,96,256,3072,256,8,0,0,asm,10001+20000,1011.3254
gfx936,no_quant,torch.float16,100,256,3072,256,8,0,0,asm,10001+20000,1012.3611
gfx936,no_quant,torch.float16,112,256,3072,256,8,0,0,asm,10001+20000,1031.8055
gfx936,no_quant,torch.float16,128,256,3072,256,8,0,0,asm,10001+20000,1058.5927
gfx936,no_quant,torch.float16,144,256,3072,256,8,0,0,asm,10001+20000,1068.6726
gfx936,no_quant,torch.float16,160,256,3072,256,8,0,0,asm,10001+20000,1079.586
gfx936,no_quant,torch.float16,192,256,3072,256,8,0,0,asm,10001+20000,1096.3268
gfx936,no_quant,torch.float16,224,256,3072,256,8,0,0,asm,10001+20000,1111.8469
gfx936,no_quant,torch.float16,256,256,3072,256,8,0,0,asm,10001+20000,1125.3208
gfx936,no_quant,torch.float16,320,256,3072,256,8,0,0,asm,10001+20000,1151.3166
gfx936,no_quant,torch.float16,384,256,3072,256,8,0,0,asm,10002+20000,1176.2598
gfx936,no_quant,torch.float16,448,256,3072,256,8,0,0,asm,10002+20000,1216.6051
gfx936,no_quant,torch.float16,512,256,3072,256,8,0,0,asm,12005+22001,1234.8115
gfx936,no_quant,torch.float16,640,256,3072,256,8,0,0,asm,12005+22001,1258.5252
gfx936,no_quant,torch.float16,768,256,3072,256,8,0,0,asm,12005+22001,1282.3315
gfx936,no_quant,torch.float16,896,256,3072,256,8,0,0,asm,12005+22001,1306.7272
gfx936,no_quant,torch.float16,1024,256,3072,256,8,0,0,asm,12005+22001,1334.2219
gfx936,no_quant,torch.float16,1280,256,3072,256,8,0,0,asm,12005+22001,1376.8577
gfx936,no_quant,torch.float16,1536,256,3072,256,8,0,0,asm,12005+22001,1437.6745
gfx936,no_quant,torch.float16,2048,256,3072,256,8,0,0,asm,13001+23001,1541.6658
gfx936,no_quant,torch.float16,2304,256,3072,256,8,0,0,asm,13001+23001,1576.0315
gfx936,no_quant,torch.float16,2560,256,3072,256,8,0,0,asm,13001+23001,1624.385
gfx936,no_quant,torch.float16,3072,256,3072,256,8,0,0,asm,13001+23001,1700.9494
gfx936,no_quant,torch.float16,3584,256,3072,256,8,0,0,asm,13001+23001,1818.2295
gfx936,no_quant,torch.float16,4096,256,3072,256,8,0,0,asm,13001+23001,1965.16
gfx936,no_quant,torch.float16,5120,256,3072,256,8,0,0,asm,13001+23001,2322.7428
gfx936,no_quant,torch.float16,6144,256,3072,256,8,0,0,asm,13001+23001,2458.768
gfx936,no_quant,torch.float16,7168,256,3072,256,8,0,0,asm,13001+23001,2695.1215
gfx936,no_quant,torch.float16,8192,256,3072,256,8,0,0,asm,13001+23001,3011.6433
gfx936,no_quant,torch.float16,10240,256,3072,256,8,0,0,asm,13001+23001,3578.2532
gfx936,no_quant,torch.float16,12288,256,3072,256,8,0,0,asm,13001+23001,4153.0568
gfx936,no_quant,torch.float16,16384,256,3072,256,8,0,0,asm,13001+23001,5263.1272
gfx936,no_quant,torch.float16,24576,256,3072,256,8,0,0,asm,13001+23001,7644.2697
gfx936,no_quant,torch.float16,32768,256,3072,256,8,0,0,asm,13001+23001,10052.4186
gfx936,no_quant,torch.float16,1,128,3072,256,8,0,0,asm,10000+20000,41.3928
gfx936,no_quant,torch.float16,2,128,3072,256,8,0,0,asm,10002+20000,57.1149
gfx936,no_quant,torch.float16,4,128,3072,256,8,0,0,asm,10002+20002,89.8055
gfx936,no_quant,torch.float16,6,128,3072,256,8,0,0,asm,10002+20002,119.0938
gfx936,no_quant,torch.float16,8,128,3072,256,8,0,0,asm,10002+20000,144.6431
gfx936,no_quant,torch.float16,12,128,3072,256,8,0,0,asm,10002+20000,194.622
gfx936,no_quant,torch.float16,16,128,3072,256,8,0,0,asm,10001+20000,235.6157
gfx936,no_quant,torch.float16,24,128,3072,256,8,0,0,asm,10001+20000,306.181
gfx936,no_quant,torch.float16,32,128,3072,256,8,0,0,asm,10001+20002,378.7229
gfx936,no_quant,torch.float16,36,128,3072,256,8,0,0,asm,10001+20002,388.2777
gfx936,no_quant,torch.float16,48,128,3072,256,8,0,0,asm,10001+20000,433.7514
gfx936,no_quant,torch.float16,56,128,3072,256,8,0,0,asm,10001+20000,457.8524
gfx936,no_quant,torch.float16,64,128,3072,256,8,0,0,asm,10001+20002,475.2419
gfx936,no_quant,torch.float16,72,128,3072,256,8,0,0,asm,10001+20002,493.6672
gfx936,no_quant,torch.float16,80,128,3072,256,8,0,0,asm,10001+20002,507.8987
gfx936,no_quant,torch.float16,88,128,3072,256,8,0,0,asm,10001+20002,524.2018
gfx936,no_quant,torch.float16,96,128,3072,256,8,0,0,asm,10001+20002,531.7892
gfx936,no_quant,torch.float16,100,128,3072,256,8,0,0,asm,10001+20002,532.3534
gfx936,no_quant,torch.float16,112,128,3072,256,8,0,0,asm,10001+20002,543.4018
gfx936,no_quant,torch.float16,128,128,3072,256,8,0,0,asm,10001+20002,556.3197
gfx936,no_quant,torch.float16,144,128,3072,256,8,0,0,asm,10001+20002,562.2986
gfx936,no_quant,torch.float16,160,128,3072,256,8,0,0,asm,10001+20000,569.8103
gfx936,no_quant,torch.float16,192,128,3072,256,8,0,0,asm,10001+20002,577.9871
gfx936,no_quant,torch.float16,224,128,3072,256,8,0,0,asm,10001+20002,588.0249
gfx936,no_quant,torch.float16,256,128,3072,256,8,0,0,asm,10001+20002,595.6291
gfx936,no_quant,torch.float16,320,128,3072,256,8,0,0,asm,10001+20002,610.5007
gfx936,no_quant,torch.float16,384,128,3072,256,8,0,0,asm,10002+20002,625.9365
gfx936,no_quant,torch.float16,448,128,3072,256,8,0,0,asm,11007+21001,636.3196
gfx936,no_quant,torch.float16,512,128,3072,256,8,0,0,asm,11007+21001,649.5155
gfx936,no_quant,torch.float16,640,128,3072,256,8,0,0,asm,11004+21001,690.1469
gfx936,no_quant,torch.float16,768,128,3072,256,8,0,0,asm,11006+21001,706.9048
gfx936,no_quant,torch.float16,896,128,3072,256,8,0,0,asm,11007+21001,707.7806
gfx936,no_quant,torch.float16,1024,128,3072,256,8,0,0,asm,11005+21001,737.069
gfx936,no_quant,torch.float16,1280,128,3072,256,8,0,0,asm,12005+22001,785.271
gfx936,no_quant,torch.float16,1536,128,3072,256,8,0,0,asm,12004+22001,845.0015
gfx936,no_quant,torch.float16,2048,128,3072,256,8,0,0,asm,12001+22001,926.7277
gfx936,no_quant,torch.float16,2304,128,3072,256,8,0,0,asm,13001+23001,967.7214
gfx936,no_quant,torch.float16,2560,128,3072,256,8,0,0,asm,13001+23001,993.0266
gfx936,no_quant,torch.float16,3072,128,3072,256,8,0,0,asm,13001+23001,1061.8433
gfx936,no_quant,torch.float16,3584,128,3072,256,8,0,0,asm,13001+23001,1178.6685
gfx936,no_quant,torch.float16,4096,128,3072,256,8,0,0,asm,13001+23001,1286.8873
gfx936,no_quant,torch.float16,5120,128,3072,256,8,0,0,asm,13001+23001,1546.3902
gfx936,no_quant,torch.float16,6144,128,3072,256,8,0,0,asm,13001+23001,1648.3774
gfx936,no_quant,torch.float16,7168,128,3072,256,8,0,0,asm,13001+23001,1815.4257
gfx936,no_quant,torch.float16,8192,128,3072,256,8,0,0,asm,13001+23001,2049.7412
gfx936,no_quant,torch.float16,10240,128,3072,256,8,0,0,asm,13001+23001,2439.6861
gfx936,no_quant,torch.float16,12288,128,3072,256,8,0,0,asm,13001+23001,2854.5487
gfx936,no_quant,torch.float16,16384,128,3072,256,8,0,0,asm,13001+23001,3669.7898
gfx936,no_quant,torch.float16,24576,128,3072,256,8,0,0,asm,13001+23001,5322.7565
gfx936,no_quant,torch.float16,32768,128,3072,256,8,0,0,asm,13001+23001,7028.0263
aiter/configs/tuned_fmoe_asm_shuffle.csv 0 → 100644
View file @ 85762c1a
arch,quant_type,indtype,token,inter_dim,model_dim,expert,topk,q_size_n,q_size_k,sol_type,sol_id,time_us
gfx938,no_quant,torch.float16,1,352,4096,128,8,0,0,asm,10009+20000,78.6408
gfx938,no_quant,torch.float16,2,352,4096,128,8,0,0,asm,10004+20002,117.1482
gfx938,no_quant,torch.float16,4,352,4096,128,8,0,0,asm,10006+20002,181.0486
gfx938,no_quant,torch.float16,6,352,4096,128,8,0,0,asm,10006+20000,236.5396
gfx938,no_quant,torch.float16,8,352,4096,128,8,0,0,asm,10007+20000,291.5917
gfx938,no_quant,torch.float16,10,352,4096,128,8,0,0,asm,10006+20000,336.7976
gfx938,no_quant,torch.float16,12,352,4096,128,8,0,0,asm,10006+20000,370.9924
gfx938,no_quant,torch.float16,16,352,4096,128,8,0,0,asm,10006+20000,436.4362
gfx938,no_quant,torch.float16,20,352,4096,128,8,0,0,asm,10006+20000,490.3447
gfx938,no_quant,torch.float16,24,352,4096,128,8,0,0,asm,10006+20000,508.1621
gfx938,no_quant,torch.float16,32,352,4096,128,8,0,0,asm,10007+20000,537.4327
gfx938,no_quant,torch.float16,36,352,4096,128,8,0,0,asm,10007+20000,556.2111
gfx938,no_quant,torch.float16,40,352,4096,128,8,0,0,asm,10006+20000,572.0756
gfx938,no_quant,torch.float16,48,352,4096,128,8,0,0,asm,10006+20000,606.7093
gfx938,no_quant,torch.float16,56,352,4096,128,8,0,0,asm,10006+20000,608.8057
gfx938,no_quant,torch.float16,64,352,4096,128,8,0,0,asm,10007+20000,618.06
gfx938,no_quant,torch.float16,72,352,4096,128,8,0,0,asm,10006+20000,612.0755
gfx938,no_quant,torch.float16,80,352,4096,128,8,0,0,asm,10006+20000,615.2016
gfx938,no_quant,torch.float16,88,352,4096,128,8,0,0,asm,10006+20000,620.6252
gfx938,no_quant,torch.float16,96,352,4096,128,8,0,0,asm,10006+20000,622.9226
gfx938,no_quant,torch.float16,112,352,4096,128,8,0,0,asm,10006+20000,625.9872
gfx938,no_quant,torch.float16,128,352,4096,128,8,0,0,asm,10007+20000,631.9154
gfx938,no_quant,torch.float16,144,352,4096,128,8,0,0,asm,10006+20000,648.1697
gfx938,no_quant,torch.float16,160,352,4096,128,8,0,0,asm,10006+20000,648.3789
gfx938,no_quant,torch.float16,192,352,4096,128,8,0,0,asm,10007+20000,667.0865
gfx938,no_quant,torch.float16,224,352,4096,128,8,0,0,asm,11004+21001,679.0988
gfx938,no_quant,torch.float16,256,352,4096,128,8,0,0,asm,11002+21001,691.788
gfx938,no_quant,torch.float16,320,352,4096,128,8,0,0,asm,11004+21001,704.5919
gfx938,no_quant,torch.float16,384,352,4096,128,8,0,0,asm,11004+21001,724.2637
gfx938,no_quant,torch.float16,512,352,4096,128,8,0,0,asm,12000+22001,774.9958
gfx938,no_quant,torch.float16,640,352,4096,128,8,0,0,asm,12000+22001,799.6931
gfx938,no_quant,torch.float16,768,352,4096,128,8,0,0,asm,12004+22001,822.5894
gfx938,no_quant,torch.float16,960,352,4096,128,8,0,0,asm,12005+22001,900.0783
gfx938,no_quant,torch.float16,1024,352,4096,128,8,0,0,asm,12005+22001,947.0566
gfx938,no_quant,torch.float16,1152,352,4096,128,8,0,0,asm,13001+23001,1047.4336
gfx938,no_quant,torch.float16,1280,352,4096,128,8,0,0,asm,13001+23001,1077.3083
gfx938,no_quant,torch.float16,1408,352,4096,128,8,0,0,asm,13001+23001,1088.9267
gfx938,no_quant,torch.float16,1536,352,4096,128,8,0,0,asm,13001+23001,1121.5625
gfx938,no_quant,torch.float16,2048,352,4096,128,8,0,0,asm,13001+23001,1410.5335
gfx938,no_quant,torch.float16,2304,352,4096,128,8,0,0,asm,12001+22001,1612.1491
gfx938,no_quant,torch.float16,2560,352,4096,128,8,0,0,asm,12001+22001,1651.5132
gfx938,no_quant,torch.float16,2816,352,4096,128,8,0,0,asm,12001+22001,1739.7137
gfx938,no_quant,torch.float16,3072,352,4096,128,8,0,0,asm,13001+23001,1789.1421
gfx938,no_quant,torch.float16,3840,352,4096,128,8,0,0,asm,13001+23001,1984.0983
gfx938,no_quant,torch.float16,4096,352,4096,128,8,0,0,asm,13001+23001,2248.1545
gfx938,no_quant,torch.float16,5120,352,4096,128,8,0,0,asm,13001+23001,2635.2017
gfx938,no_quant,torch.float16,6144,352,4096,128,8,0,0,asm,13001+23001,3099.4145
gfx938,no_quant,torch.float16,7168,352,4096,128,8,0,0,asm,13001+23001,3564.3297
gfx938,no_quant,torch.float16,7680,352,4096,128,8,0,0,asm,13001+23001,3648.124
gfx938,no_quant,torch.float16,8192,352,4096,128,8,0,0,asm,13001+23001,3963.4827
gfx938,no_quant,torch.float16,12288,352,4096,128,8,0,0,asm,13001+23001,5748.9863
gfx938,no_quant,torch.float16,16384,352,4096,128,8,0,0,asm,13001+23001,7546.4485
gfx938,no_quant,torch.float16,24576,352,4096,128,8,0,0,asm,13001+23001,11088.5202
gfx938,no_quant,torch.float16,32768,352,4096,128,8,0,0,asm,13001+23001,14713.3474
gfx938,no_quant,torch.float16,49152,352,4096,128,8,0,0,asm,13001+23001,21815.9769
gfx938,no_quant,torch.float16,65536,352,4096,128,8,0,0,asm,13001+23001,29012.8921
gfx938,no_quant,torch.float16,1,352,4096,129,9,0,0,asm,10004+20000,82.8296
gfx938,no_quant,torch.float16,2,352,4096,129,9,0,0,asm,10004+20002,117.517
gfx938,no_quant,torch.float16,4,352,4096,129,9,0,0,asm,10006+20002,183.4528
gfx938,no_quant,torch.float16,6,352,4096,129,9,0,0,asm,10006+20000,241.0813
gfx938,no_quant,torch.float16,8,352,4096,129,9,0,0,asm,10007+20000,279.1283
gfx938,no_quant,torch.float16,10,352,4096,129,9,0,0,asm,10006+20000,330.1271
gfx938,no_quant,torch.float16,12,352,4096,129,9,0,0,asm,10006+20000,359.8869
gfx938,no_quant,torch.float16,14,352,4096,129,9,0,0,asm,10006+20000,395.9155
gfx938,no_quant,torch.float16,16,352,4096,129,9,0,0,asm,10006+20000,424.8344
gfx938,no_quant,torch.float16,20,352,4096,129,9,0,0,asm,10007+20000,483.2763
gfx938,no_quant,torch.float16,24,352,4096,129,9,0,0,asm,10006+20000,521.8279
gfx938,no_quant,torch.float16,28,352,4096,129,9,0,0,asm,10006+20000,548.0873
gfx938,no_quant,torch.float16,32,352,4096,129,9,0,0,asm,10006+20000,558.2041
gfx938,no_quant,torch.float16,36,352,4096,129,9,0,0,asm,10006+20000,586.8205
gfx938,no_quant,torch.float16,40,352,4096,129,9,0,0,asm,10006+20000,588.6543
gfx938,no_quant,torch.float16,48,352,4096,129,9,0,0,asm,10006+20000,611.8296
gfx938,no_quant,torch.float16,56,352,4096,129,9,0,0,asm,10006+20000,614.9763
gfx938,no_quant,torch.float16,64,352,4096,129,9,0,0,asm,10006+20000,632.5423
gfx938,no_quant,torch.float16,72,352,4096,129,9,0,0,asm,10006+20000,624.0347
gfx938,no_quant,torch.float16,80,352,4096,129,9,0,0,asm,10006+20000,629.3024
gfx938,no_quant,torch.float16,88,352,4096,129,9,0,0,asm,10006+20000,631.3577
gfx938,no_quant,torch.float16,96,352,4096,129,9,0,0,asm,10006+20000,630.9926
gfx938,no_quant,torch.float16,104,352,4096,129,9,0,0,asm,10006+20000,633.9628
gfx938,no_quant,torch.float16,112,352,4096,129,9,0,0,asm,10007+20000,644.289
gfx938,no_quant,torch.float16,128,352,4096,129,9,0,0,asm,10007+20000,647.6202
gfx938,no_quant,torch.float16,144,352,4096,129,9,0,0,asm,10006+20000,659.6407
gfx938,no_quant,torch.float16,160,352,4096,129,9,0,0,asm,10006+20000,665.6427
gfx938,no_quant,torch.float16,192,352,4096,129,9,0,0,asm,11004+21001,682.6765
gfx938,no_quant,torch.float16,224,352,4096,129,9,0,0,asm,11004+21001,688.2437
gfx938,no_quant,torch.float16,256,352,4096,129,9,0,0,asm,11004+21001,694.6395
gfx938,no_quant,torch.float16,320,352,4096,129,9,0,0,asm,11002+21001,719.4435
gfx938,no_quant,torch.float16,384,352,4096,129,9,0,0,asm,11004+21001,752.6085
gfx938,no_quant,torch.float16,448,352,4096,129,9,0,0,asm,12000+22001,773.6054
gfx938,no_quant,torch.float16,512,352,4096,129,9,0,0,asm,12000+22001,791.353
gfx938,no_quant,torch.float16,576,352,4096,129,9,0,0,asm,12004+22001,804.8627
gfx938,no_quant,torch.float16,640,352,4096,129,9,0,0,asm,12004+22001,820.5673
gfx938,no_quant,torch.float16,768,352,4096,129,9,0,0,asm,12000+22001,862.1548
gfx938,no_quant,torch.float16,960,352,4096,129,9,0,0,asm,12001+22001,993.7481
gfx938,no_quant,torch.float16,1024,352,4096,129,9,0,0,asm,12001+22001,1051.5161
gfx938,no_quant,torch.float16,1280,352,4096,129,9,0,0,asm,13001+23001,1115.6184
gfx938,no_quant,torch.float16,1536,352,4096,129,9,0,0,asm,13001+23001,1145.0624
gfx938,no_quant,torch.float16,1920,352,4096,129,9,0,0,asm,12001+22001,1517.5169
gfx938,no_quant,torch.float16,2048,352,4096,129,9,0,0,asm,12001+22001,1602.4269
gfx938,no_quant,torch.float16,2304,352,4096,129,9,0,0,asm,12001+22001,1694.8448
gfx938,no_quant,torch.float16,2560,352,4096,129,9,0,0,asm,12001+22001,1768.7974
gfx938,no_quant,torch.float16,2816,352,4096,129,9,0,0,asm,13001+23001,1804.2844
gfx938,no_quant,torch.float16,3072,352,4096,129,9,0,0,asm,13001+23001,1848.514
gfx938,no_quant,torch.float16,3584,352,4096,129,9,0,0,asm,13001+23001,2137.4626
gfx938,no_quant,torch.float16,3840,352,4096,129,9,0,0,asm,13001+23001,2433.5947
gfx938,no_quant,torch.float16,4096,352,4096,129,9,0,0,asm,13001+23001,2581.0639
gfx938,no_quant,torch.float16,4608,352,4096,129,9,0,0,asm,13001+23001,2696.285
gfx938,no_quant,torch.float16,5120,352,4096,129,9,0,0,asm,13001+23001,2818.4465
gfx938,no_quant,torch.float16,5632,352,4096,129,9,0,0,asm,13001+23001,3244.0284
gfx938,no_quant,torch.float16,6144,352,4096,129,9,0,0,asm,13001+23001,3530.5874
gfx938,no_quant,torch.float16,6656,352,4096,129,9,0,0,asm,13001+23001,3601.098
gfx938,no_quant,torch.float16,7168,352,4096,129,9,0,0,asm,13001+23001,3889.9903
gfx938,no_quant,torch.float16,7680,352,4096,129,9,0,0,asm,13001+23001,4337.7228
gfx938,no_quant,torch.float16,8192,352,4096,129,9,0,0,asm,13001+23001,4443.3674
gfx938,no_quant,torch.float16,10240,352,4096,129,9,0,0,asm,13001+23001,5380.2134
gfx938,no_quant,torch.float16,12288,352,4096,129,9,0,0,asm,13001+23001,6382.2687
gfx938,no_quant,torch.float16,16384,352,4096,129,9,0,0,asm,13001+23001,8397.3437
gfx938,no_quant,torch.float16,24576,352,4096,129,9,0,0,asm,13001+23001,12379.6318
gfx938,no_quant,torch.float16,32768,352,4096,129,9,0,0,asm,13001+23001,16397.3012
gfx938,no_quant,torch.float16,40960,352,4096,129,9,0,0,asm,13001+23001,20398.5288
gfx938,no_quant,torch.float16,49152,352,4096,129,9,0,0,asm,13001+23001,24396.6972
gfx938,no_quant,torch.float16,65536,352,4096,129,9,0,0,asm,13001+23001,32435.0655
gfx936,no_quant,torch.float16,1,256,3072,256,8,0,0,asm,10006+20000,56.4327
gfx936,no_quant,torch.float16,2,256,3072,256,8,0,0,asm,10006+20000,85.2664
gfx936,no_quant,torch.float16,4,256,3072,256,8,0,0,asm,10004+20000,148.02
gfx936,no_quant,torch.float16,6,256,3072,256,8,0,0,asm,10001+20000,198.0409
gfx936,no_quant,torch.float16,8,256,3072,256,8,0,0,asm,10001+20000,237.9062
gfx936,no_quant,torch.float16,12,256,3072,256,8,0,0,asm,10001+20000,320.1882
gfx936,no_quant,torch.float16,16,256,3072,256,8,0,0,asm,10001+20000,388.3143
gfx936,no_quant,torch.float16,24,256,3072,256,8,0,0,asm,10001+20000,520.2353
gfx936,no_quant,torch.float16,32,256,3072,256,8,0,0,asm,10001+20000,634.3687
gfx936,no_quant,torch.float16,36,256,3072,256,8,0,0,asm,10001+20000,662.1298
gfx936,no_quant,torch.float16,48,256,3072,256,8,0,0,asm,10001+20000,745.2455
gfx936,no_quant,torch.float16,56,256,3072,256,8,0,0,asm,10001+20000,784.2686
gfx936,no_quant,torch.float16,64,256,3072,256,8,0,0,asm,10001+20000,822.5338
gfx936,no_quant,torch.float16,72,256,3072,256,8,0,0,asm,10001+20000,850.3484
gfx936,no_quant,torch.float16,80,256,3072,256,8,0,0,asm,10001+20000,875.1736
gfx936,no_quant,torch.float16,88,256,3072,256,8,0,0,asm,10001+20000,901.001
gfx936,no_quant,torch.float16,96,256,3072,256,8,0,0,asm,10001+20000,914.5758
gfx936,no_quant,torch.float16,100,256,3072,256,8,0,0,asm,10001+20000,917.5063
gfx936,no_quant,torch.float16,112,256,3072,256,8,0,0,asm,10001+20000,936.2515
gfx936,no_quant,torch.float16,128,256,3072,256,8,0,0,asm,10001+20000,967.6199
gfx936,no_quant,torch.float16,144,256,3072,256,8,0,0,asm,10001+20000,976.1083
gfx936,no_quant,torch.float16,160,256,3072,256,8,0,0,asm,10001+20000,987.5608
gfx936,no_quant,torch.float16,192,256,3072,256,8,0,0,asm,10001+20000,1007.7293
gfx936,no_quant,torch.float16,224,256,3072,256,8,0,0,asm,10007+20000,1009.22
gfx936,no_quant,torch.float16,256,256,3072,256,8,0,0,asm,10007+20000,1027.6957
gfx936,no_quant,torch.float16,320,256,3072,256,8,0,0,asm,10007+20000,1043.8052
gfx936,no_quant,torch.float16,384,256,3072,256,8,0,0,asm,10007+20000,1075.1736
gfx936,no_quant,torch.float16,448,256,3072,256,8,0,0,asm,10006+20000,1110.1966
gfx936,no_quant,torch.float16,512,256,3072,256,8,0,0,asm,11006+21001,1120.0746
gfx936,no_quant,torch.float16,640,256,3072,256,8,0,0,asm,12004+22001,1151.2998
gfx936,no_quant,torch.float16,768,256,3072,256,8,0,0,asm,12004+22001,1180.8324
gfx936,no_quant,torch.float16,896,256,3072,256,8,0,0,asm,12004+22001,1201.9524
gfx936,no_quant,torch.float16,1024,256,3072,256,8,0,0,asm,12004+22001,1221.1692
gfx936,no_quant,torch.float16,1280,256,3072,256,8,0,0,asm,12004+22001,1256.6892
gfx936,no_quant,torch.float16,1536,256,3072,256,8,0,0,asm,12004+22001,1313.9523
gfx936,no_quant,torch.float16,2048,256,3072,256,8,0,0,asm,13000+23001,1467.2068
gfx936,no_quant,torch.float16,2304,256,3072,256,8,0,0,asm,13000+23001,1512.9077
gfx936,no_quant,torch.float16,2560,256,3072,256,8,0,0,asm,13000+23001,1530.7098
gfx936,no_quant,torch.float16,3072,256,3072,256,8,0,0,asm,13000+23001,1615.5686
gfx936,no_quant,torch.float16,3584,256,3072,256,8,0,0,asm,13000+23001,1724.7306
gfx936,no_quant,torch.float16,4096,256,3072,256,8,0,0,asm,13001+23001,1918.8103
gfx936,no_quant,torch.float16,5120,256,3072,256,8,0,0,asm,13001+23001,2272.5277
gfx936,no_quant,torch.float16,6144,256,3072,256,8,0,0,asm,13001+23001,2343.0624
gfx936,no_quant,torch.float16,7168,256,3072,256,8,0,0,asm,13001+23001,2560.4684
gfx936,no_quant,torch.float16,8192,256,3072,256,8,0,0,asm,13001+23001,2878.8006
gfx936,no_quant,torch.float16,10240,256,3072,256,8,0,0,asm,13001+23001,3414.3448
gfx936,no_quant,torch.float16,12288,256,3072,256,8,0,0,asm,13001+23001,3988.8697
gfx936,no_quant,torch.float16,16384,256,3072,256,8,0,0,asm,13001+23001,5057.6343
gfx936,no_quant,torch.float16,24576,256,3072,256,8,0,0,asm,13001+23001,7355.5588
gfx936,no_quant,torch.float16,32768,256,3072,256,8,0,0,asm,13001+23001,9745.8034
gfx936,no_quant,torch.float16,1,128,3072,256,8,0,0,asm,10005+20000,42.0834
gfx936,no_quant,torch.float16,2,128,3072,256,8,0,0,asm,10006+20000,57.5023
gfx936,no_quant,torch.float16,4,128,3072,256,8,0,0,asm,10003+20000,89.1148
gfx936,no_quant,torch.float16,6,128,3072,256,8,0,0,asm,10004+20000,115.9274
gfx936,no_quant,torch.float16,8,128,3072,256,8,0,0,asm,10004+20000,140.0622
gfx936,no_quant,torch.float16,12,128,3072,256,8,0,0,asm,10004+20000,183.4137
gfx936,no_quant,torch.float16,16,128,3072,256,8,0,0,asm,10001+20000,217.6199
gfx936,no_quant,torch.float16,24,128,3072,256,8,0,0,asm,10001+20000,280.0841
gfx936,no_quant,torch.float16,32,128,3072,256,8,0,0,asm,10001+20000,347.9859
gfx936,no_quant,torch.float16,36,128,3072,256,8,0,0,asm,10001+20002,354.8798
gfx936,no_quant,torch.float16,48,128,3072,256,8,0,0,asm,10004+20000,397.8103
gfx936,no_quant,torch.float16,56,128,3072,256,8,0,0,asm,10001+20000,417.8355
gfx936,no_quant,torch.float16,64,128,3072,256,8,0,0,asm,10001+20000,431.9071
gfx936,no_quant,torch.float16,72,128,3072,256,8,0,0,asm,10001+20000,449.6586
gfx936,no_quant,torch.float16,80,128,3072,256,8,0,0,asm,10001+20000,466.3575
gfx936,no_quant,torch.float16,88,128,3072,256,8,0,0,asm,10001+20000,476.4375
gfx936,no_quant,torch.float16,96,128,3072,256,8,0,0,asm,10001+20000,484.1006
gfx936,no_quant,torch.float16,100,128,3072,256,8,0,0,asm,10001+20000,483.6543
gfx936,no_quant,torch.float16,112,128,3072,256,8,0,0,asm,10001+20000,496.6479
gfx936,no_quant,torch.float16,128,128,3072,256,8,0,0,asm,10001+20000,506.4332
gfx936,no_quant,torch.float16,144,128,3072,256,8,0,0,asm,10001+20000,514.7279
gfx936,no_quant,torch.float16,160,128,3072,256,8,0,0,asm,10001+20000,518.9469
gfx936,no_quant,torch.float16,192,128,3072,256,8,0,0,asm,10001+20002,528.2184
gfx936,no_quant,torch.float16,224,128,3072,256,8,0,0,asm,10001+20000,536.9594
gfx936,no_quant,torch.float16,256,128,3072,256,8,0,0,asm,10001+20002,542.1299
gfx936,no_quant,torch.float16,320,128,3072,256,8,0,0,asm,10001+20002,560.412
gfx936,no_quant,torch.float16,384,128,3072,256,8,0,0,asm,10007+20002,578.4752
gfx936,no_quant,torch.float16,448,128,3072,256,8,0,0,asm,10006+20002,595.0982
gfx936,no_quant,torch.float16,512,128,3072,256,8,0,0,asm,11006+21001,614.9719
gfx936,no_quant,torch.float16,640,128,3072,256,8,0,0,asm,11006+21001,643.1908
gfx936,no_quant,torch.float16,768,128,3072,256,8,0,0,asm,11006+21001,656.4455
gfx936,no_quant,torch.float16,896,128,3072,256,8,0,0,asm,11006+21001,669.5487
gfx936,no_quant,torch.float16,1024,128,3072,256,8,0,0,asm,11006+21001,709.4055
gfx936,no_quant,torch.float16,1280,128,3072,256,8,0,0,asm,12004+22001,751.9485
gfx936,no_quant,torch.float16,1536,128,3072,256,8,0,0,asm,12004+22001,792.2685
gfx936,no_quant,torch.float16,2048,128,3072,256,8,0,0,asm,12004+22001,896.9672
gfx936,no_quant,torch.float16,2304,128,3072,256,8,0,0,asm,13000+23001,963.6365
gfx936,no_quant,torch.float16,2560,128,3072,256,8,0,0,asm,13000+23001,1003.0975
gfx936,no_quant,torch.float16,3072,128,3072,256,8,0,0,asm,13000+23001,1068.8068
gfx936,no_quant,torch.float16,3584,128,3072,256,8,0,0,asm,13000+23001,1157.8004
gfx936,no_quant,torch.float16,4096,128,3072,256,8,0,0,asm,12001+22001,1282.457
gfx936,no_quant,torch.float16,5120,128,3072,256,8,0,0,asm,12001+22001,1514.4903
gfx936,no_quant,torch.float16,6144,128,3072,256,8,0,0,asm,13001+23001,1667.9553
gfx936,no_quant,torch.float16,7168,128,3072,256,8,0,0,asm,13001+23001,1819.8287
gfx936,no_quant,torch.float16,8192,128,3072,256,8,0,0,asm,13001+23001,2046.742
gfx936,no_quant,torch.float16,10240,128,3072,256,8,0,0,asm,13001+23001,2443.8613
gfx936,no_quant,torch.float16,12288,128,3072,256,8,0,0,asm,13001+23001,2844.9048
gfx936,no_quant,torch.float16,16384,128,3072,256,8,0,0,asm,13001+23001,3597.2571
gfx936,no_quant,torch.float16,24576,128,3072,256,8,0,0,asm,13001+23001,5205.65
gfx936,no_quant,torch.float16,32768,128,3072,256,8,0,0,asm,13001+23001,6847.9883
aiter/configs/tuned_fmoe_asm_w4a16.csv 0 → 100644
View file @ 85762c1a
arch,quant_type,indtype,token,inter_dim,model_dim,expert,topk,q_size_n,q_size_k,sol_type,sol_id,time_us
gfx936,int4_w4a16,torch.float16,1,256,7168,256,8,0,0,asm,10000+20000,95.17214285714428
gfx936,int4_w4a16,torch.float16,2,256,7168,256,8,0,0,asm,10000+20000,109.64071428571371
gfx936,int4_w4a16,torch.float16,4,256,7168,256,8,0,0,asm,10000+20001,162.646428571428
gfx936,int4_w4a16,torch.float16,6,256,7168,256,8,0,0,asm,10000+20000,224.8634285714288
gfx936,int4_w4a16,torch.float16,8,256,7168,256,8,0,0,asm,10000+20000,245.8005714285721
gfx936,int4_w4a16,torch.float16,10,256,7168,256,8,0,0,asm,10000+20001,298.3948571428573
gfx936,int4_w4a16,torch.float16,12,256,7168,256,8,0,0,asm,10000+20001,304.8405714285706
gfx936,int4_w4a16,torch.float16,14,256,7168,256,8,0,0,asm,10000+20001,386.76042857142886
gfx936,int4_w4a16,torch.float16,16,256,7168,256,8,0,0,asm,10000+20001,421.7547142857137
gfx936,int4_w4a16,torch.float16,20,256,7168,256,8,0,0,asm,10000+20000,517.134571428571
gfx936,int4_w4a16,torch.float16,24,256,7168,256,8,0,0,asm,10000+20000,530.8487142857141
gfx936,int4_w4a16,torch.float16,28,256,7168,256,8,0,0,asm,10000+20001,571.7202857142856
gfx936,int4_w4a16,torch.float16,32,256,7168,256,8,0,0,asm,10000+20001,568.383142857143
gfx936,int4_w4a16,torch.float16,36,256,7168,256,8,0,0,asm,10000+20001,652.1515714285714
gfx936,int4_w4a16,torch.float16,40,256,7168,256,8,0,0,asm,10000+20001,700.4257142857138
gfx936,int4_w4a16,torch.float16,44,256,7168,256,8,0,0,asm,10000+20001,705.5685714285721
gfx936,int4_w4a16,torch.float16,48,256,7168,256,8,0,0,asm,10000+20001,696.4028571428581
gfx936,int4_w4a16,torch.float16,56,256,7168,256,8,0,0,asm,10000+20001,794.9171428571426
gfx936,int4_w4a16,torch.float16,64,256,7168,256,8,0,0,asm,10000+20001,814.9171428571414
gfx936,int4_w4a16,torch.float16,80,256,7168,256,8,0,0,asm,10000+20001,829.8885714285714
gfx936,int4_w4a16,torch.float16,96,256,7168,256,8,0,0,asm,10000+20001,911.3284285714284
gfx936,int4_w4a16,torch.float16,112,256,7168,256,8,0,0,asm,10000+20001,918.1169999999996
gfx936,int4_w4a16,torch.float16,128,256,7168,256,8,0,0,asm,10000+20001,926.8941428571424
gfx936,int4_w4a16,torch.float16,160,256,7168,256,8,0,0,asm,10000+20001,941.3170000000003
gfx936,int4_w4a16,torch.float16,192,256,7168,256,8,0,0,asm,10000+20001,958.139857142858
gfx936,int4_w4a16,torch.float16,224,256,7168,256,8,0,0,asm,10000+20001,983.6941428571432
gfx936,int4_w4a16,torch.float16,256,256,7168,256,8,0,0,asm,10000+20001,988.4255714285724
gfx936,int4_w4a16,torch.float16,320,256,7168,256,8,0,0,asm,10000+20001,1010.7332857142854
gfx936,int4_w4a16,torch.float16,384,256,7168,256,8,0,0,asm,10000+20001,1094.2761428571423
gfx936,int4_w4a16,torch.float16,448,256,7168,256,8,0,0,asm,10000+20001,1153.590142857143
gfx936,int4_w4a16,torch.float16,512,256,7168,256,8,0,0,asm,10000+20001,1306.458571428572
gfx936,int4_w4a16,torch.float16,576,256,7168,256,8,0,0,asm,11001+21001,1410.2527142857143
gfx936,int4_w4a16,torch.float16,640,256,7168,256,8,0,0,asm,11001+21001,1434.2071428571428
gfx936,int4_w4a16,torch.float16,704,256,7168,256,8,0,0,asm,11001+21001,1496.423999999999
gfx936,int4_w4a16,torch.float16,768,256,7168,256,8,0,0,asm,11001+21001,1520.1268571428573
gfx936,int4_w4a16,torch.float16,832,256,7168,256,8,0,0,asm,11001+21001,1607.280999999999
gfx936,int4_w4a16,torch.float16,896,256,7168,256,8,0,0,asm,11001+21001,1697.932285714286
gfx936,int4_w4a16,torch.float16,960,256,7168,256,8,0,0,asm,11001+21001,1793.5892857142867
gfx936,int4_w4a16,torch.float16,1024,256,7168,256,8,0,0,asm,11001+21001,1949.2920000000008
gfx936,int4_w4a16,torch.float16,1152,256,7168,256,8,0,0,asm,11001+21001,2259.9428571428566
gfx936,int4_w4a16,torch.float16,1280,256,7168,256,8,0,0,asm,11001+21001,2472.994
gfx936,int4_w4a16,torch.float16,1408,256,7168,256,8,0,0,asm,11001+21001,2536.4909999999995
gfx936,int4_w4a16,torch.float16,1536,256,7168,256,8,0,0,asm,11001+21001,2635.736571428572
gfx936,int4_w4a16,torch.float16,1664,256,7168,256,8,0,0,asm,11001+21001,2727.5991428571456
gfx936,int4_w4a16,torch.float16,1792,256,7168,256,8,0,0,asm,11001+21001,2847.301857142858
gfx936,int4_w4a16,torch.float16,1920,256,7168,256,8,0,0,asm,11001+21001,3030.9815714285733
gfx936,int4_w4a16,torch.float16,2048,256,7168,256,8,0,0,asm,11001+21001,3253.38114285714
gfx936,int4_w4a16,torch.float16,2304,256,7168,256,8,0,0,asm,11001+21001,3691.437714285715
gfx936,int4_w4a16,torch.float16,2560,256,7168,256,8,0,0,asm,11001+21001,3961.105857142853
gfx936,int4_w4a16,torch.float16,2816,256,7168,256,8,0,0,asm,11001+21001,4214.888285714294
gfx936,int4_w4a16,torch.float16,3072,256,7168,256,8,0,0,asm,11001+21001,4598.590571428569
gfx936,int4_w4a16,torch.float16,3328,256,7168,256,8,0,0,asm,11001+21001,4923.961571428568
gfx936,int4_w4a16,torch.float16,3584,256,7168,256,8,0,0,asm,11001+21001,5283.663857142858
gfx936,int4_w4a16,torch.float16,3840,256,7168,256,8,0,0,asm,11001+21001,5579.549000000005
gfx936,int4_w4a16,torch.float16,4096,256,7168,256,8,0,0,asm,11001+21001,5959.091142857148
gfx936,int4_w4a16,torch.float16,4608,256,7168,256,8,0,0,asm,11001+21001,6573.558857142861
gfx936,int4_w4a16,torch.float16,5120,256,7168,256,8,0,0,asm,11001+21001,7240.826571428568
gfx936,int4_w4a16,torch.float16,5632,256,7168,256,8,0,0,asm,11001+21001,7880.962714285714
gfx936,int4_w4a16,torch.float16,6144,256,7168,256,8,0,0,asm,11001+21001,8558.858571428571
gfx936,int4_w4a16,torch.float16,6656,256,7168,256,8,0,0,asm,11001+21001,9222.514857142862
gfx936,int4_w4a16,torch.float16,7168,256,7168,256,8,0,0,asm,11001+21001,9888.730857142857
gfx936,int4_w4a16,torch.float16,7680,256,7168,256,8,0,0,asm,11001+21001,10493.644714285716
gfx936,int4_w4a16,torch.float16,8192,256,7168,256,8,0,0,asm,11001+21001,11209.300285714282
gfx936,int4_w4a16,torch.float16,10240,256,7168,256,8,0,0,asm,11001+21001,13796.542142857152
gfx936,int4_w4a16,torch.float16,12288,256,7168,256,8,0,0,asm,11001+21001,16473.77242857143
gfx936,int4_w4a16,torch.float16,14336,256,7168,256,8,0,0,asm,11001+21001,19168.557142857146
gfx936,int4_w4a16,torch.float16,16384,256,7168,256,8,0,0,asm,11001+21001,21783.615857142864
gfx936,int4_w4a16,torch.float16,17408,256,7168,256,8,0,0,asm,11001+21001,23135.63657142856
gfx936,int4_w4a16,torch.float16,24576,256,7168,256,8,0,0,asm,11001+21001,32335.211
gfx936,int4_w4a16,torch.float16,32768,256,7168,256,8,0,0,asm,11001+21001,42933.38957142859
gfx938,int4_w4a16,torch.float16,1,256,7168,256,8,0,0,asm,10000+20000,105.43471428539071
gfx938,int4_w4a16,torch.float16,2,256,7168,256,8,0,0,asm,10000+20000,127.08042857237159
gfx938,int4_w4a16,torch.float16,4,256,7168,256,8,0,0,asm,10000+20001,192.08599999959446
gfx938,int4_w4a16,torch.float16,6,256,7168,256,8,0,0,asm,10000+20000,284.08557142796263
gfx938,int4_w4a16,torch.float16,8,256,7168,256,8,0,0,asm,10000+20002,344.4512857141838
gfx938,int4_w4a16,torch.float16,10,256,7168,256,8,0,0,asm,10000+20000,440.88528571384296
gfx938,int4_w4a16,torch.float16,12,256,7168,256,8,0,0,asm,10000+20001,501.20514285670856
gfx938,int4_w4a16,torch.float16,14,256,7168,256,8,0,0,asm,10000+20000,596.9989999990378
gfx938,int4_w4a16,torch.float16,16,256,7168,256,8,0,0,asm,10000+20002,605.4334285715928
gfx938,int4_w4a16,torch.float16,20,256,7168,256,8,0,0,asm,10000+20001,750.7557142856531
gfx938,int4_w4a16,torch.float16,24,256,7168,256,8,0,0,asm,10000+20001,756.40142857137
gfx938,int4_w4a16,torch.float16,28,256,7168,256,8,0,0,asm,10000+20001,825.547142857087
gfx938,int4_w4a16,torch.float16,32,256,7168,256,8,0,0,asm,10000+20001,836.5414285711678
gfx938,int4_w4a16,torch.float16,36,256,7168,256,8,0,0,asm,10000+20002,911.303857142904
gfx938,int4_w4a16,torch.float16,40,256,7168,256,8,0,0,asm,10000+20002,967.166571428567
gfx938,int4_w4a16,torch.float16,44,256,7168,256,8,0,0,asm,10000+20001,1061.2692857140541
gfx938,int4_w4a16,torch.float16,48,256,7168,256,8,0,0,asm,10000+20001,1066.9377142857495
gfx938,int4_w4a16,torch.float16,56,256,7168,256,8,0,0,asm,10000+20001,1116.629000000057
gfx938,int4_w4a16,torch.float16,64,256,7168,256,8,0,0,asm,10000+20002,1193.4288571427876
gfx938,int4_w4a16,torch.float16,80,256,7168,256,8,0,0,asm,10000+20002,1217.9315714285476
gfx938,int4_w4a16,torch.float16,96,256,7168,256,8,0,0,asm,10000+20002,1250.1371428567384
gfx938,int4_w4a16,torch.float16,112,256,7168,256,8,0,0,asm,10000+20002,1273.9772857142877
gfx938,int4_w4a16,torch.float16,128,256,7168,256,8,0,0,asm,10000+20002,1276.6285714285416
gfx938,int4_w4a16,torch.float16,160,256,7168,256,8,0,0,asm,10000+20002,1278.320142857198
gfx938,int4_w4a16,torch.float16,192,256,7168,256,8,0,0,asm,10000+20002,1289.4514285716493
gfx938,int4_w4a16,torch.float16,224,256,7168,256,8,0,0,asm,10000+20002,1295.0284285714984
gfx938,int4_w4a16,torch.float16,256,256,7168,256,8,0,0,asm,10000+20001,1408.8795714286555
gfx938,int4_w4a16,torch.float16,320,256,7168,256,8,0,0,asm,10000+20001,1426.959571428597
gfx938,int4_w4a16,torch.float16,384,256,7168,256,8,0,0,asm,10000+20001,1486.8908571428951
gfx938,int4_w4a16,torch.float16,448,256,7168,256,8,0,0,asm,10000+20002,1611.6447142859522
gfx938,int4_w4a16,torch.float16,512,256,7168,256,8,0,0,asm,10000+20002,1923.9182857143958
gfx938,int4_w4a16,torch.float16,576,256,7168,256,8,0,0,asm,10000+20002,2170.1348571429094
gfx938,int4_w4a16,torch.float16,640,256,7168,256,8,0,0,asm,10000+20001,2377.3797142855556
gfx938,int4_w4a16,torch.float16,704,256,7168,256,8,0,0,asm,10000+20002,2500.55657142893
gfx938,int4_w4a16,torch.float16,768,256,7168,256,8,0,0,asm,10000+20002,2559.2078571429342
gfx938,int4_w4a16,torch.float16,832,256,7168,256,8,0,0,asm,10000+20001,2760.692999999305
gfx938,int4_w4a16,torch.float16,896,256,7168,256,8,0,0,asm,10000+20002,2826.293000000263
gfx938,int4_w4a16,torch.float16,960,256,7168,256,8,0,0,asm,10000+20002,3070.77228571433
gfx938,int4_w4a16,torch.float16,1024,256,7168,256,8,0,0,asm,10000+20002,3187.7548571428715
gfx938,int4_w4a16,torch.float16,1152,256,7168,256,8,0,0,asm,10000+20002,3501.8795714285225
gfx938,int4_w4a16,torch.float16,1280,256,7168,256,8,0,0,asm,10000+20002,3815.9817142858437
gfx938,int4_w4a16,torch.float16,1408,256,7168,256,8,0,0,asm,10000+20002,4135.843714285708
gfx938,int4_w4a16,torch.float16,1536,256,7168,256,8,0,0,asm,10000+20002,4453.054285714236
gfx938,int4_w4a16,torch.float16,1664,256,7168,256,8,0,0,asm,10000+20002,4908.824428571521
gfx938,int4_w4a16,torch.float16,1792,256,7168,256,8,0,0,asm,10000+20002,5088.504000000057
gfx938,int4_w4a16,torch.float16,1920,256,7168,256,8,0,0,asm,10000+20002,5400.8005714285455
gfx938,int4_w4a16,torch.float16,2048,256,7168,256,8,0,0,asm,10000+20002,5717.8740000000025
gfx938,int4_w4a16,torch.float16,2304,256,7168,256,8,0,0,asm,10000+20002,6355.015000000048
gfx938,int4_w4a16,torch.float16,2560,256,7168,256,8,0,0,asm,10000+20002,7130.669857142909
gfx938,int4_w4a16,torch.float16,2816,256,7168,256,8,0,0,asm,10000+20002,7623.331428571405
gfx938,int4_w4a16,torch.float16,3072,256,7168,256,8,0,0,asm,10000+20002,8248.769285714088
gfx938,int4_w4a16,torch.float16,3328,256,7168,256,8,0,0,asm,10000+20002,8877.841571428573
gfx938,int4_w4a16,torch.float16,3584,256,7168,256,8,0,0,asm,10000+20002,9516.399999999574
gfx938,int4_w4a16,torch.float16,3840,256,7168,256,8,0,0,asm,10000+20002,10147.803285714293
gfx938,int4_w4a16,torch.float16,4096,256,7168,256,8,0,0,asm,10000+20002,10784.464142857187
gfx938,int4_w4a16,torch.float16,4608,256,7168,256,8,0,0,asm,10000+20002,12002.792285714431
gfx938,int4_w4a16,torch.float16,5120,256,7168,256,8,0,0,asm,10000+20002,13309.120142857179
gfx938,int4_w4a16,torch.float16,5632,256,7168,256,8,0,0,asm,10000+20002,14576.659714285517
gfx938,int4_w4a16,torch.float16,6144,256,7168,256,8,0,0,asm,10000+20002,15842.21071428606
gfx938,int4_w4a16,torch.float16,6656,256,7168,256,8,0,0,asm,10000+20002,17109.932714285194
gfx938,int4_w4a16,torch.float16,7168,256,7168,256,8,0,0,asm,10000+20002,18361.67728571487
gfx938,int4_w4a16,torch.float16,7680,256,7168,256,8,0,0,asm,10000+20002,19628.737285714065
gfx938,int4_w4a16,torch.float16,8192,256,7168,256,8,0,0,asm,10000+20002,20897.122571428772
gfx938,int4_w4a16,torch.float16,10240,256,7168,256,8,0,0,asm,10000+20002,25957.222142856703
gfx938,int4_w4a16,torch.float16,12288,256,7168,256,8,0,0,asm,10000+20002,31018.37357142857
gfx938,int4_w4a16,torch.float16,14336,256,7168,256,8,0,0,asm,10000+20002,36075.15457142866
gfx938,int4_w4a16,torch.float16,16384,256,7168,256,8,0,0,asm,10000+20002,41135.762714286466
gfx938,int4_w4a16,torch.float16,17408,256,7168,256,8,0,0,asm,10000+20002,43667.20685714267
gfx938,int4_w4a16,torch.float16,24576,256,7168,256,8,0,0,asm,10000+20002,61364.23214285701
gfx938,int4_w4a16,torch.float16,32768,256,7168,256,8,0,0,asm,10000+20002,81594.00357142859
gfx938,int4_w4a16,torch.float16,1,256,7168,384,8,0,0,asm,10000+20000,108.1967
gfx938,int4_w4a16,torch.float16,2,256,7168,384,8,0,0,asm,10000+20000,126.5040
gfx938,int4_w4a16,torch.float16,4,256,7168,384,8,0,0,asm,10000+20001,196.5924
gfx938,int4_w4a16,torch.float16,6,256,7168,384,8,0,0,asm,10000+20000,290.6636
gfx938,int4_w4a16,torch.float16,8,256,7168,384,8,0,0,asm,10000+20002,348.1119
gfx938,int4_w4a16,torch.float16,10,256,7168,384,8,0,0,asm,10000+20000,445.9054
gfx938,int4_w4a16,torch.float16,12,256,7168,384,8,0,0,asm,10000+20001,511.1852
gfx938,int4_w4a16,torch.float16,14,256,7168,384,8,0,0,asm,10000+20001,519.2441
gfx938,int4_w4a16,torch.float16,16,256,7168,384,8,0,0,asm,10000+20002,616.4397
gfx938,int4_w4a16,torch.float16,20,256,7168,384,8,0,0,asm,10000+20001,753.6237
gfx938,int4_w4a16,torch.float16,24,256,7168,384,8,0,0,asm,10000+20001,817.4551
gfx938,int4_w4a16,torch.float16,28,256,7168,384,8,0,0,asm,10000+20002,912.1243
gfx938,int4_w4a16,torch.float16,32,256,7168,384,8,0,0,asm,10000+20002,963.9222
gfx938,int4_w4a16,torch.float16,36,256,7168,384,8,0,0,asm,10000+20001,1108.6964
gfx938,int4_w4a16,torch.float16,40,256,7168,384,8,0,0,asm,10000+20002,1200.7804
gfx938,int4_w4a16,torch.float16,44,256,7168,384,8,0,0,asm,10000+20002,1252.2244
gfx938,int4_w4a16,torch.float16,48,256,7168,384,8,0,0,asm,10000+20002,1264.3423
gfx938,int4_w4a16,torch.float16,56,256,7168,384,8,0,0,asm,10000+20001,1378.8346
gfx938,int4_w4a16,torch.float16,64,256,7168,384,8,0,0,asm,10000+20002,1498.6154
gfx938,int4_w4a16,torch.float16,80,256,7168,384,8,0,0,asm,10000+20002,1581.3351
gfx938,int4_w4a16,torch.float16,96,256,7168,384,8,0,0,asm,10000+20001,1686.9685
gfx938,int4_w4a16,torch.float16,112,256,7168,384,8,0,0,asm,10000+20001,1739.7936
gfx938,int4_w4a16,torch.float16,128,256,7168,384,8,0,0,asm,10000+20002,1813.1493
gfx938,int4_w4a16,torch.float16,160,256,7168,384,8,0,0,asm,10000+20002,1869.6711
gfx938,int4_w4a16,torch.float16,192,256,7168,384,8,0,0,asm,10000+20002,1879.8691
gfx938,int4_w4a16,torch.float16,224,256,7168,384,8,0,0,asm,10000+20002,1898.5048
gfx938,int4_w4a16,torch.float16,256,256,7168,384,8,0,0,asm,10000+20002,1908.3069
gfx938,int4_w4a16,torch.float16,320,256,7168,384,8,0,0,asm,10000+20002,1914.6057
gfx938,int4_w4a16,torch.float16,384,256,7168,384,8,0,0,asm,10000+20001,2022.0243
gfx938,int4_w4a16,torch.float16,448,256,7168,384,8,0,0,asm,10000+20001,2031.7760
gfx938,int4_w4a16,torch.float16,512,256,7168,384,8,0,0,asm,10000+20001,2056.854
gfx938,int4_w4a16,torch.float16,576,256,7168,384,8,0,0,asm,10000+20002,2222.1249
gfx938,int4_w4a16,torch.float16,640,256,7168,384,8,0,0,asm,10000+20001,2354.6466
gfx938,int4_w4a16,torch.float16,768,256,7168,384,8,0,0,asm,10000+20002,2802.9235
gfx938,int4_w4a16,torch.float16,896,256,7168,384,8,0,0,asm,10000+20002,3359.1829
gfx938,int4_w4a16,torch.float16,1024,256,7168,384,8,0,0,asm,10000+20002,3622.8789
gfx938,int4_w4a16,torch.float16,1280,256,7168,384,8,0,0,asm,10000+20002,4052.1077
gfx938,int4_w4a16,torch.float16,1536,256,7168,384,8,0,0,asm,10000+20002,4672.0215
gfx938,int4_w4a16,torch.float16,2048,256,7168,384,8,0,0,asm,10000+20002,5924.0686
gfx938,int4_w4a16,torch.float16,2560,256,7168,384,8,0,0,asm,10000+20002,7228.8482
gfx938,int4_w4a16,torch.float16,3072,256,7168,384,8,0,0,asm,10000+20002,8496.5672
gfx938,int4_w4a16,torch.float16,3584,256,7168,384,8,0,0,asm,10000+20002,9699.309
gfx938,int4_w4a16,torch.float16,3840,256,7168,384,8,0,0,asm,10000+20002,10366.0861
gfx938,int4_w4a16,torch.float16,4096,256,7168,384,8,0,0,asm,10000+20002,10946.9519
gfx938,int4_w4a16,torch.float16,4608,256,7168,384,8,0,0,asm,10000+20002,12245.0451
gfx938,int4_w4a16,torch.float16,5120,256,7168,384,8,0,0,asm,10000+20002,13491.7789
gfx938,int4_w4a16,torch.float16,6144,256,7168,384,8,0,0,asm,10000+20002,15987.2496
gfx938,int4_w4a16,torch.float16,7168,256,7168,384,8,0,0,asm,10000+20002,18455.4978
gfx938,int4_w4a16,torch.float16,7680,256,7168,384,8,0,0,asm,10000+20002,19750.3316
gfx938,int4_w4a16,torch.float16,8192,256,7168,384,8,0,0,asm,10000+20002,21000.2742
gfx938,int4_w4a16,torch.float16,10240,256,7168,384,8,0,0,asm,10000+20002,26001.247
gfx938,int4_w4a16,torch.float16,12288,256,7168,384,8,0,0,asm,10000+20002,30978.9432
gfx938,int4_w4a16,torch.float16,16384,256,7168,384,8,0,0,asm,10000+20002,40987.1437
gfx938,int4_w4a16,torch.float16,24576,256,7168,384,8,0,0,asm,10000+20002,60960.2526
gfx938,int4_w4a16,torch.float16,32768,256,7168,384,8,0,0,asm,10000+20002,80978.0681
aiter/configs/tuned_fmoe_asm_w4a8_group.csv 0 → 100644
View file @ 85762c1a
arch,quant_type,indtype,token,inter_dim,model_dim,expert,topk,q_size_n,q_size_k,sol_type,sol_id,time_us
gfx936,int4_w4a8,torch.float16,1,256,7168,256,8,0,0,asm,10001+20100,121.2912857142849
gfx936,int4_w4a8,torch.float16,2,256,7168,256,8,0,0,asm,10001+20000,137.86271428571425
gfx936,int4_w4a8,torch.float16,4,256,7168,256,8,0,0,asm,10002+20101,181.22271428571534
gfx936,int4_w4a8,torch.float16,6,256,7168,256,8,0,0,asm,10001+20100,243.78257142857066
gfx936,int4_w4a8,torch.float16,8,256,7168,256,8,0,0,asm,10001+20101,270.7997142857147
gfx936,int4_w4a8,torch.float16,10,256,7168,256,8,0,0,asm,10001+20101,305.10828571428607
gfx936,int4_w4a8,torch.float16,12,256,7168,256,8,0,0,asm,10002+20101,376.28528571428666
gfx936,int4_w4a8,torch.float16,14,256,7168,256,8,0,0,asm,10001+20101,385.5881428571429
gfx936,int4_w4a8,torch.float16,16,256,7168,256,8,0,0,asm,10002+20101,415.09671428571426
gfx936,int4_w4a8,torch.float16,20,256,7168,256,8,0,0,asm,10001+20101,496.4679999999984
gfx936,int4_w4a8,torch.float16,24,256,7168,256,8,0,0,asm,10001+20101,504.7651428571422
gfx936,int4_w4a8,torch.float16,28,256,7168,256,8,0,0,asm,10001+20101,546.5937142857139
gfx936,int4_w4a8,torch.float16,32,256,7168,256,8,0,0,asm,10001+20101,552.2165714285693
gfx936,int4_w4a8,torch.float16,36,256,7168,256,8,0,0,asm,10001+20101,629.9535714285719
gfx936,int4_w4a8,torch.float16,40,256,7168,256,8,0,0,asm,10002+20101,665.0164285714278
gfx936,int4_w4a8,torch.float16,44,256,7168,256,8,0,0,asm,10001+20101,673.3135714285709
gfx936,int4_w4a8,torch.float16,48,256,7168,256,8,0,0,asm,10001+20101,732.5589999999995
gfx936,int4_w4a8,torch.float16,56,256,7168,256,8,0,0,asm,10001+20101,769.8848571428576
gfx936,int4_w4a8,torch.float16,64,256,7168,256,8,0,0,asm,10001+20101,783.5077142857135
gfx936,int4_w4a8,torch.float16,72,256,7168,256,8,0,0,asm,10002+20101,816.6962857142868
gfx936,int4_w4a8,torch.float16,80,256,7168,256,8,0,0,asm,10001+20101,812.2162857142861
gfx936,int4_w4a8,torch.float16,96,256,7168,256,8,0,0,asm,10001+20101,834.7534285714281
gfx936,int4_w4a8,torch.float16,112,256,7168,256,8,0,0,asm,10001+20101,888.2618571428568
gfx936,int4_w4a8,torch.float16,128,256,7168,256,8,0,0,asm,10001+20101,894.7989999999998
gfx936,int4_w4a8,torch.float16,160,256,7168,256,8,0,0,asm,10002+20101,916.1018571428574
gfx936,int4_w4a8,torch.float16,192,256,7168,256,8,0,0,asm,10002+20101,937.0160000000004
gfx936,int4_w4a8,torch.float16,224,256,7168,256,8,0,0,asm,10002+20101,980.3990000000005
gfx936,int4_w4a8,torch.float16,256,256,7168,256,8,0,0,asm,10002+20101,989.221714285714
gfx936,int4_w4a8,torch.float16,320,256,7168,256,8,0,0,asm,10002+20101,1032.2160000000001
gfx936,int4_w4a8,torch.float16,384,256,7168,256,8,0,0,asm,10002+20101,1075.0731428571432
gfx936,int4_w4a8,torch.float16,448,256,7168,256,8,0,0,asm,10002+20101,1112.9245714285714
gfx936,int4_w4a8,torch.float16,512,256,7168,256,8,0,0,asm,11001+21101,1172.170142857143
gfx936,int4_w4a8,torch.float16,576,256,7168,256,8,0,0,asm,11001+21101,1200.741428571429
gfx936,int4_w4a8,torch.float16,640,256,7168,256,8,0,0,asm,11001+21101,1236.4672857142875
gfx936,int4_w4a8,torch.float16,704,256,7168,256,8,0,0,asm,11001+21101,1269.9985714285708
gfx936,int4_w4a8,torch.float16,768,256,7168,256,8,0,0,asm,11001+21101,1321.54142857143
gfx936,int4_w4a8,torch.float16,832,256,7168,256,8,0,0,asm,11001+21101,1423.6670000000001
gfx936,int4_w4a8,torch.float16,896,256,7168,256,8,0,0,asm,11001+21101,1467.1412857142861
gfx936,int4_w4a8,torch.float16,960,256,7168,256,8,0,0,asm,11001+21101,1529.1754285714298
gfx936,int4_w4a8,torch.float16,1024,256,7168,256,8,0,0,asm,11001+21101,1651.9409999999991
gfx936,int4_w4a8,torch.float16,1152,256,7168,256,8,0,0,asm,12000+22101,1841.1522857142859
gfx936,int4_w4a8,torch.float16,1280,256,7168,256,8,0,0,asm,12000+22101,1928.740714285715
gfx936,int4_w4a8,torch.float16,1408,256,7168,256,8,0,0,asm,12000+22101,2031.506428571427
gfx936,int4_w4a8,torch.float16,1536,256,7168,256,8,0,0,asm,12000+22100,2093.7921428571417
gfx936,int4_w4a8,torch.float16,1664,256,7168,256,8,0,0,asm,12000+22100,2159.643428571429
gfx936,int4_w4a8,torch.float16,1792,256,7168,256,8,0,0,asm,12000+22100,2340.2147142857125
gfx936,int4_w4a8,torch.float16,1920,256,7168,256,8,0,0,asm,12000+22101,2536.511571428572
gfx936,int4_w4a8,torch.float16,2048,256,7168,256,8,0,0,asm,12000+22101,2711.5742857142855
gfx936,int4_w4a8,torch.float16,2304,256,7168,256,8,0,0,asm,12000+22101,3163.2538571428563
gfx936,int4_w4a8,torch.float16,2560,256,7168,256,8,0,0,asm,11001+21101,3370.6592857142864
gfx936,int4_w4a8,torch.float16,2816,256,7168,256,8,0,0,asm,12000+22101,3553.2875714285706
gfx936,int4_w4a8,torch.float16,3072,256,7168,256,8,0,0,asm,12000+22101,3640.8532857142886
gfx936,int4_w4a8,torch.float16,3328,256,7168,256,8,0,0,asm,12000+22101,3754.5674285714254
gfx936,int4_w4a8,torch.float16,3584,256,7168,256,8,0,0,asm,12000+22101,4066.704285714291
gfx936,int4_w4a8,torch.float16,3840,256,7168,256,8,0,0,asm,12000+22101,4285.606857142853
gfx936,int4_w4a8,torch.float16,4096,256,7168,256,8,0,0,asm,12000+22101,4511.504
gfx936,int4_w4a8,torch.float16,4608,256,7168,256,8,0,0,asm,12000+22101,5246.771714285715
gfx936,int4_w4a8,torch.float16,5120,256,7168,256,8,0,0,asm,12000+22101,5601.42285714286
gfx936,int4_w4a8,torch.float16,5632,256,7168,256,8,0,0,asm,12000+22101,5841.239571428574
gfx936,int4_w4a8,torch.float16,6144,256,7168,256,8,0,0,asm,12000+22101,6483.31914285714
gfx936,int4_w4a8,torch.float16,6656,256,7168,256,8,0,0,asm,12000+22101,7094.564285714291
gfx936,int4_w4a8,torch.float16,7168,256,7168,256,8,0,0,asm,12000+22101,7407.546857142855
gfx936,int4_w4a8,torch.float16,7680,256,7168,256,8,0,0,asm,12000+22101,7759.112571428579
gfx936,int4_w4a8,torch.float16,8192,256,7168,256,8,0,0,asm,12000+22101,8344.666142857135
gfx936,int4_w4a8,torch.float16,10240,256,7168,256,8,0,0,asm,12000+22101,10278.127285714294
gfx936,int4_w4a8,torch.float16,12288,256,7168,256,8,0,0,asm,12000+22101,12087.954142857148
gfx936,int4_w4a8,torch.float16,14336,256,7168,256,8,0,0,asm,12000+22101,13912.912571428593
gfx936,int4_w4a8,torch.float16,16384,256,7168,256,8,0,0,asm,12000+22101,15923.08257142857
gfx936,int4_w4a8,torch.float16,17408,256,7168,256,8,0,0,asm,13000+23101,17620.064000000006
gfx936,int4_w4a8,torch.float16,24576,256,7168,256,8,0,0,asm,13000+23101,22840.699571428548
gfx936,int4_w4a8,torch.float16,32768,256,7168,256,8,0,0,asm,13000+23101,29839.300142857148
gfx936,int4_w4a8,torch.float16,1,128,7168,256,8,0,0,asm,10001+20000,118.84557142857135
gfx936,int4_w4a8,torch.float16,8,128,7168,256,8,0,0,asm,10002+20000,210.50271428571457
gfx936,int4_w4a8,torch.float16,32,128,7168,256,8,0,0,asm,10002+20000,410.98257142857136
gfx936,int4_w4a8,torch.float16,48,128,7168,256,8,0,0,asm,10001+20000,486.6395714285716
gfx936,int4_w4a8,torch.float16,64,128,7168,256,8,0,0,asm,10001+20000,507.0281428571426
gfx936,int4_w4a8,torch.float16,80,128,7168,256,8,0,0,asm,10001+20000,530.9595714285716
gfx936,int4_w4a8,torch.float16,96,128,7168,256,8,0,0,asm,10002+20000,564.7651428571434
gfx936,int4_w4a8,torch.float16,128,128,7168,256,8,0,0,asm,10002+20000,588.8794285714288
gfx936,int4_w4a8,torch.float16,256,128,7168,256,8,0,0,asm,10002+20000,651.4622857142858
gfx936,int4_w4a8,torch.float16,512,128,7168,256,8,0,0,asm,11001+21000,816.7192857142861
gfx936,int4_w4a8,torch.float16,768,128,7168,256,8,0,0,asm,11001+21000,905.7248571428562
gfx936,int4_w4a8,torch.float16,1024,128,7168,256,8,0,0,asm,11001+21000,1194.478857142858
gfx936,int4_w4a8,torch.float16,2048,128,7168,256,8,0,0,asm,11001+21000,2014.7065714285711
gfx936,int4_w4a8,torch.float16,3072,128,7168,256,8,0,0,asm,13000+23000,2794.6144285714277
gfx936,int4_w4a8,torch.float16,4096,128,7168,256,8,0,0,asm,12000+22000,3567.93942857143
gfx936,int4_w4a8,torch.float16,5120,128,7168,256,8,0,0,asm,12000+22000,4365.652857142859
gfx936,int4_w4a8,torch.float16,6144,128,7168,256,8,0,0,asm,12000+22000,5152.532285714286
gfx936,int4_w4a8,torch.float16,8192,128,7168,256,8,0,0,asm,13000+23000,6526.58800000001
gfx936,int4_w4a8,torch.float16,10240,128,7168,256,8,0,0,asm,13000+23000,8007.180714285722
gfx936,int4_w4a8,torch.float16,12288,128,7168,256,8,0,0,asm,13000+23000,9478.585142857133
gfx936,int4_w4a8,torch.float16,16384,128,7168,256,8,0,0,asm,13000+23000,12350.97114285713
gfx936,int4_w4a8,torch.float16,24576,128,7168,256,8,0,0,asm,13000+23000,17921.823142857134
gfx936,int4_w4a8,torch.float16,32768,128,7168,256,8,0,0,asm,13000+23000,23625.54328571425
aiter/configs/tuned_fmoe_asm_w8a8_channel.csv 0 → 100644
View file @ 85762c1a
This diff is collapsed.
aiter/configs/tuned_fmoe_asm_w8a8_channel_shuffle.csv 0 → 100644
View file @ 85762c1a
This diff is collapsed.
aiter/configs/tuned_fmoe_asm_w8a8_group.csv 0 → 100644
View file @ 85762c1a
This diff is collapsed.
aiter/configs/tuned_fmoe_asm_w8a8_group_shuffle.csv 0 → 100644
View file @ 85762c1a
This diff is collapsed.
aiter/configs/tuned_fmoe_ck.csv 0 → 100644
View file @ 85762c1a
quant_type,indtype,token,inter_dim,model_dim,expert,topk,q_size_n,q_size_k,sol_type,sol_id,time_us
no_quant,torch.float16,1,256,8192,256,4,0,0,ck,576,309.8021428571405
no_quant,torch.float16,8,256,8192,256,4,0,0,ck,576,364.40785714285965
no_quant,torch.float16,16,256,8192,256,4,0,0,ck,576,705.001571428565
no_quant,torch.float16,24,256,8192,256,4,0,0,ck,272,1028.2664285714334
no_quant,torch.float16,32,256,8192,256,4,0,0,ck,576,1293.2290000000055
no_quant,torch.float16,48,256,8192,256,4,0,0,ck,272,1877.2707142857207
no_quant,torch.float16,64,256,8192,256,4,0,0,ck,576,2193.247142857142
no_quant,torch.float16,96,256,8192,256,4,0,0,ck,272,2471.966714285715
no_quant,torch.float16,128,256,8192,256,4,0,0,ck,576,2724.7889999999948
no_quant,torch.float16,256,256,8192,256,4,0,0,ck,576,3391.9192857142875
no_quant,torch.float16,512,256,8192,256,4,0,0,ck,576,3537.358999999999
no_quant,torch.float16,1024,256,8192,256,4,0,0,ck,272,5022.04200000001
no_quant,torch.float16,2048,256,8192,256,4,0,0,ck,272,6217.148285714277
no_quant,torch.float16,4096,256,8192,256,4,0,0,ck,272,8764.503571428571
no_quant,torch.float16,8192,256,8192,256,4,0,0,ck,272,16520.489142857128
no_quant,torch.float16,1,256,7168,256,8,0,0,ck,576,274.4422857142844
no_quant,torch.float16,8,256,7168,256,8,0,0,ck,576,584.9560000000014
no_quant,torch.float16,16,256,7168,256,8,0,0,ck,576,1056.360857142859
no_quant,torch.float16,24,256,7168,256,8,0,0,ck,576,1480.196999999999
no_quant,torch.float16,32,256,7168,256,8,0,0,ck,576,1867.970714285712
no_quant,torch.float16,48,256,7168,256,8,0,0,ck,576,2312.378428571425
no_quant,torch.float16,64,256,7168,256,8,0,0,ck,576,2382.001142857145
no_quant,torch.float16,96,256,7168,256,8,0,0,ck,576,2795.691714285705
no_quant,torch.float16,128,256,7168,256,8,0,0,ck,576,2820.2631428571426
no_quant,torch.float16,256,256,7168,256,8,0,0,ck,576,3055.348428571423
no_quant,torch.float16,512,256,7168,256,8,0,0,ck,272,4095.392285714287
no_quant,torch.float16,1024,256,7168,256,8,0,0,ck,272,5514.589714285707
no_quant,torch.float16,2048,256,7168,256,8,0,0,ck,272,7718.288428571438
no_quant,torch.float16,4096,256,7168,256,8,0,0,ck,272,14435.327142857148
no_quant,torch.float16,8192,256,7168,256,8,0,0,ck,272,27850.411142857163
aiter/configs/tuned_gemm.csv 0 → 100644
View file @ 85762c1a
M,N,K,bias,dtype,outdtype,scaleAB,libtype,solidx,soltimes,kernelName
aiter/configs/untuned_fmoe.csv 0 → 100644
View file @ 85762c1a
token,model_dim,inter_dim,expert,topk,act_type,dtype,q_dtype_a,q_dtype_w,q_type,use_g1u1,doweight_stage1
256,6144,4096,8,2,ActivationType.Silu,torch.bfloat16,torch.float8_e4m3fnuz,torch.float8_e4m3fnuz,QuantType.per_128x128,1,0
512,6144,4096,8,2,ActivationType.Silu,torch.bfloat16,torch.int8,torch.int8,QuantType.per_Tensor,1,0
512,6144,4096,8,2,ActivationType.Silu,torch.bfloat16,torch.int8,torch.int8,QuantType.per_Token,1,0
512,6144,4096,8,2,ActivationType.Silu,torch.bfloat16,torch.float8_e4m3fnuz,torch.float8_e4m3fnuz,QuantType.per_Token,1,0
512,6144,4096,8,2,ActivationType.Silu,torch.bfloat16,torch.float8_e4m3fnuz,torch.int4,QuantType.per_Tensor,1,0
512,6144,4096,8,2,ActivationType.Gelu,torch.bfloat16,torch.int8,torch.int8,QuantType.per_Tensor,1,0
512,6144,4096,8,2,ActivationType.Gelu,torch.bfloat16,torch.int8,torch.int8,QuantType.per_Token,1,0
512,6144,4096,8,2,ActivationType.Gelu,torch.bfloat16,torch.float8_e4m3fnuz,torch.float8_e4m3fnuz,QuantType.per_Token,1,0
512,6144,4096,8,2,ActivationType.Gelu,torch.bfloat16,torch.float8_e4m3fnuz,torch.int4,QuantType.per_Tensor,1,0
4,2304,1536,8,2,ActivationType.Gelu,torch.bfloat16,torch.float8_e4m3fnuz,torch.float8_e4m3fnuz,QuantType.per_Token,1,0
4,2304,1536,8,2,ActivationType.Gelu,torch.bfloat16,torch.float8_e4m3fnuz,torch.float8_e4m3fnuz,QuantType.per_Token,1,1
\ No newline at end of file
aiter/configs/untuned_fmoe_asm_w4a16.csv 0 → 100644
View file @ 85762c1a
quant_type,indtype,token,inter_dim,model_dim,expert,topk,q_size_n,q_size_k
int4_w4a16,torch.float16,1,256,7168,256,8,0,0
int4_w4a16,torch.float16,2,256,7168,256,8,0,0
int4_w4a16,torch.float16,4,256,7168,256,8,0,0
int4_w4a16,torch.float16,6,256,7168,256,8,0,0
int4_w4a16,torch.float16,8,256,7168,256,8,0,0
int4_w4a16,torch.float16,10,256,7168,256,8,0,0
int4_w4a16,torch.float16,12,256,7168,256,8,0,0
int4_w4a16,torch.float16,14,256,7168,256,8,0,0
int4_w4a16,torch.float16,16,256,7168,256,8,0,0
int4_w4a16,torch.float16,20,256,7168,256,8,0,0
int4_w4a16,torch.float16,24,256,7168,256,8,0,0
int4_w4a16,torch.float16,28,256,7168,256,8,0,0
int4_w4a16,torch.float16,32,256,7168,256,8,0,0
int4_w4a16,torch.float16,36,256,7168,256,8,0,0
int4_w4a16,torch.float16,40,256,7168,256,8,0,0
int4_w4a16,torch.float16,44,256,7168,256,8,0,0
int4_w4a16,torch.float16,48,256,7168,256,8,0,0
int4_w4a16,torch.float16,56,256,7168,256,8,0,0
int4_w4a16,torch.float16,64,256,7168,256,8,0,0
int4_w4a16,torch.float16,80,256,7168,256,8,0,0
int4_w4a16,torch.float16,96,256,7168,256,8,0,0
int4_w4a16,torch.float16,112,256,7168,256,8,0,0
int4_w4a16,torch.float16,128,256,7168,256,8,0,0
int4_w4a16,torch.float16,160,256,7168,256,8,0,0
int4_w4a16,torch.float16,192,256,7168,256,8,0,0
int4_w4a16,torch.float16,224,256,7168,256,8,0,0
int4_w4a16,torch.float16,256,256,7168,256,8,0,0
int4_w4a16,torch.float16,320,256,7168,256,8,0,0
int4_w4a16,torch.float16,384,256,7168,256,8,0,0
int4_w4a16,torch.float16,448,256,7168,256,8,0,0
int4_w4a16,torch.float16,512,256,7168,256,8,0,0
int4_w4a16,torch.float16,576,256,7168,256,8,0,0
int4_w4a16,torch.float16,640,256,7168,256,8,0,0
int4_w4a16,torch.float16,704,256,7168,256,8,0,0
int4_w4a16,torch.float16,768,256,7168,256,8,0,0
int4_w4a16,torch.float16,832,256,7168,256,8,0,0
int4_w4a16,torch.float16,896,256,7168,256,8,0,0
int4_w4a16,torch.float16,960,256,7168,256,8,0,0
int4_w4a16,torch.float16,1024,256,7168,256,8,0,0
int4_w4a16,torch.float16,1152,256,7168,256,8,0,0
int4_w4a16,torch.float16,1280,256,7168,256,8,0,0
int4_w4a16,torch.float16,1408,256,7168,256,8,0,0
int4_w4a16,torch.float16,1536,256,7168,256,8,0,0
int4_w4a16,torch.float16,1664,256,7168,256,8,0,0
int4_w4a16,torch.float16,1792,256,7168,256,8,0,0
int4_w4a16,torch.float16,1920,256,7168,256,8,0,0
int4_w4a16,torch.float16,2048,256,7168,256,8,0,0
int4_w4a16,torch.float16,2304,256,7168,256,8,0,0
int4_w4a16,torch.float16,2560,256,7168,256,8,0,0
int4_w4a16,torch.float16,2816,256,7168,256,8,0,0
int4_w4a16,torch.float16,3072,256,7168,256,8,0,0
int4_w4a16,torch.float16,3328,256,7168,256,8,0,0
int4_w4a16,torch.float16,3584,256,7168,256,8,0,0
int4_w4a16,torch.float16,3840,256,7168,256,8,0,0
int4_w4a16,torch.float16,4096,256,7168,256,8,0,0
int4_w4a16,torch.float16,4608,256,7168,256,8,0,0
int4_w4a16,torch.float16,5120,256,7168,256,8,0,0
int4_w4a16,torch.float16,5632,256,7168,256,8,0,0
int4_w4a16,torch.float16,6144,256,7168,256,8,0,0
int4_w4a16,torch.float16,6656,256,7168,256,8,0,0
int4_w4a16,torch.float16,7168,256,7168,256,8,0,0
int4_w4a16,torch.float16,7680,256,7168,256,8,0,0
int4_w4a16,torch.float16,8192,256,7168,256,8,0,0
int4_w4a16,torch.float16,10240,256,7168,256,8,0,0
int4_w4a16,torch.float16,12288,256,7168,256,8,0,0
int4_w4a16,torch.float16,14336,256,7168,256,8,0,0
int4_w4a16,torch.float16,16384,256,7168,256,8,0,0
int4_w4a16,torch.float16,17408,256,7168,256,8,0,0
int4_w4a16,torch.float16,24576,256,7168,256,8,0,0
int4_w4a16,torch.float16,32768,256,7168,256,8,0,0
\ No newline at end of file
aiter/configs/untuned_fmoe_asm_w4a8.csv 0 → 100644
View file @ 85762c1a
quant_type,indtype,token,inter_dim,model_dim,expert,topk,q_size_n,q_size_k
int4_w4a8,torch.float16,1,256,7168,256,8,0,0
int4_w4a8,torch.float16,2,256,7168,256,8,0,0
int4_w4a8,torch.float16,4,256,7168,256,8,0,0
int4_w4a8,torch.float16,6,256,7168,256,8,0,0
int4_w4a8,torch.float16,8,256,7168,256,8,0,0
int4_w4a8,torch.float16,10,256,7168,256,8,0,0
int4_w4a8,torch.float16,12,256,7168,256,8,0,0
int4_w4a8,torch.float16,14,256,7168,256,8,0,0
int4_w4a8,torch.float16,16,256,7168,256,8,0,0
int4_w4a8,torch.float16,20,256,7168,256,8,0,0
int4_w4a8,torch.float16,24,256,7168,256,8,0,0
int4_w4a8,torch.float16,28,256,7168,256,8,0,0
int4_w4a8,torch.float16,32,256,7168,256,8,0,0
int4_w4a8,torch.float16,36,256,7168,256,8,0,0
int4_w4a8,torch.float16,40,256,7168,256,8,0,0
int4_w4a8,torch.float16,44,256,7168,256,8,0,0
int4_w4a8,torch.float16,48,256,7168,256,8,0,0
int4_w4a8,torch.float16,56,256,7168,256,8,0,0
int4_w4a8,torch.float16,64,256,7168,256,8,0,0
int4_w4a8,torch.float16,72,256,7168,256,8,0,0
int4_w4a8,torch.float16,80,256,7168,256,8,0,0
int4_w4a8,torch.float16,96,256,7168,256,8,0,0
int4_w4a8,torch.float16,112,256,7168,256,8,0,0
int4_w4a8,torch.float16,128,256,7168,256,8,0,0
int4_w4a8,torch.float16,160,256,7168,256,8,0,0
int4_w4a8,torch.float16,192,256,7168,256,8,0,0
int4_w4a8,torch.float16,224,256,7168,256,8,0,0
int4_w4a8,torch.float16,256,256,7168,256,8,0,0
int4_w4a8,torch.float16,320,256,7168,256,8,0,0
int4_w4a8,torch.float16,384,256,7168,256,8,0,0
int4_w4a8,torch.float16,448,256,7168,256,8,0,0
int4_w4a8,torch.float16,512,256,7168,256,8,0,0
int4_w4a8,torch.float16,576,256,7168,256,8,0,0
int4_w4a8,torch.float16,640,256,7168,256,8,0,0
int4_w4a8,torch.float16,704,256,7168,256,8,0,0
int4_w4a8,torch.float16,768,256,7168,256,8,0,0
int4_w4a8,torch.float16,832,256,7168,256,8,0,0
int4_w4a8,torch.float16,896,256,7168,256,8,0,0
int4_w4a8,torch.float16,960,256,7168,256,8,0,0
int4_w4a8,torch.float16,1024,256,7168,256,8,0,0
int4_w4a8,torch.float16,1152,256,7168,256,8,0,0
int4_w4a8,torch.float16,1280,256,7168,256,8,0,0
int4_w4a8,torch.float16,1408,256,7168,256,8,0,0
int4_w4a8,torch.float16,1536,256,7168,256,8,0,0
int4_w4a8,torch.float16,1664,256,7168,256,8,0,0
int4_w4a8,torch.float16,1792,256,7168,256,8,0,0
int4_w4a8,torch.float16,1920,256,7168,256,8,0,0
int4_w4a8,torch.float16,2048,256,7168,256,8,0,0
int4_w4a8,torch.float16,2304,256,7168,256,8,0,0
int4_w4a8,torch.float16,2560,256,7168,256,8,0,0
int4_w4a8,torch.float16,2816,256,7168,256,8,0,0
int4_w4a8,torch.float16,3072,256,7168,256,8,0,0
int4_w4a8,torch.float16,3328,256,7168,256,8,0,0
int4_w4a8,torch.float16,3584,256,7168,256,8,0,0
int4_w4a8,torch.float16,3840,256,7168,256,8,0,0
int4_w4a8,torch.float16,4096,256,7168,256,8,0,0
int4_w4a8,torch.float16,4608,256,7168,256,8,0,0
int4_w4a8,torch.float16,5120,256,7168,256,8,0,0
int4_w4a8,torch.float16,5632,256,7168,256,8,0,0
int4_w4a8,torch.float16,6144,256,7168,256,8,0,0
int4_w4a8,torch.float16,6656,256,7168,256,8,0,0
int4_w4a8,torch.float16,7168,256,7168,256,8,0,0
int4_w4a8,torch.float16,7680,256,7168,256,8,0,0
int4_w4a8,torch.float16,8192,256,7168,256,8,0,0
int4_w4a8,torch.float16,10240,256,7168,256,8,0,0
int4_w4a8,torch.float16,12288,256,7168,256,8,0,0
int4_w4a8,torch.float16,14336,256,7168,256,8,0,0
int4_w4a8,torch.float16,16384,256,7168,256,8,0,0
int4_w4a8,torch.float16,17408,256,7168,256,8,0,0
int4_w4a8,torch.float16,24576,256,7168,256,8,0,0
int4_w4a8,torch.float16,32768,256,7168,256,8,0,0
int4_w4a8,torch.float16,1,128,7168,256,8,0,0
int4_w4a8,torch.float16,8,128,7168,256,8,0,0
int4_w4a8,torch.float16,32,128,7168,256,8,0,0
int4_w4a8,torch.float16,48,128,7168,256,8,0,0
int4_w4a8,torch.float16,64,128,7168,256,8,0,0
int4_w4a8,torch.float16,80,128,7168,256,8,0,0
int4_w4a8,torch.float16,96,128,7168,256,8,0,0
int4_w4a8,torch.float16,128,128,7168,256,8,0,0
int4_w4a8,torch.float16,256,128,7168,256,8,0,0
int4_w4a8,torch.float16,512,128,7168,256,8,0,0
int4_w4a8,torch.float16,768,128,7168,256,8,0,0
int4_w4a8,torch.float16,1024,128,7168,256,8,0,0
int4_w4a8,torch.float16,2048,128,7168,256,8,0,0
int4_w4a8,torch.float16,3072,128,7168,256,8,0,0
int4_w4a8,torch.float16,4096,128,7168,256,8,0,0
int4_w4a8,torch.float16,5120,128,7168,256,8,0,0
int4_w4a8,torch.float16,6144,128,7168,256,8,0,0
int4_w4a8,torch.float16,8192,128,7168,256,8,0,0
int4_w4a8,torch.float16,10240,128,7168,256,8,0,0
int4_w4a8,torch.float16,12288,128,7168,256,8,0,0
int4_w4a8,torch.float16,16384,128,7168,256,8,0,0
int4_w4a8,torch.float16,24576,128,7168,256,8,0,0
int4_w4a8,torch.float16,32768,128,7168,256,8,0,0
\ No newline at end of file
aiter/configs/untuned_fmoe_asm_w8a8.csv 0 → 100644
View file @ 85762c1a
quant_type,indtype,token,inter_dim,model_dim,expert,topk,q_size_n,q_size_k
f8_w8a8_channel,torch.float16,1,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,2,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,3,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,4,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,5,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,6,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,7,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,8,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,9,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,10,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,11,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,12,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,13,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,14,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,15,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,16,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,17,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,18,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,20,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,24,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,28,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,32,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,34,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,36,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,40,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,44,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,48,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,56,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,64,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,68,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,72,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,80,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,88,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,96,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,104,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,112,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,128,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,144,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,160,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,192,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,224,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,256,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,320,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,384,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,448,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,512,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,576,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,640,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,704,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,768,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,832,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,896,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,960,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,1024,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,1152,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,1280,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,1408,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,1536,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,1664,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,1792,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,1920,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,2048,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,2304,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,2560,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,2816,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,3072,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,3328,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,3584,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,3840,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,4096,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,4608,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,5120,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,5632,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,6144,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,6656,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,7168,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,7680,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,8192,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,10240,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,12288,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,14336,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,16384,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,17408,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,24576,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,32768,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,40960,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,49152,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,57344,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,65536,128,7168,256,8,0,0
f8_w8a8_channel,torch.float16,1,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,2,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,3,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,4,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,5,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,6,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,7,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,8,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,9,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,10,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,11,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,12,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,13,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,14,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,15,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,16,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,17,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,18,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,20,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,24,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,28,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,32,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,34,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,36,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,40,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,44,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,48,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,56,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,64,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,68,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,72,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,80,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,88,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,96,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,104,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,112,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,128,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,144,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,160,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,192,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,224,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,256,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,320,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,384,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,448,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,512,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,576,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,640,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,704,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,768,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,832,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,896,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,960,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,1024,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,1152,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,1280,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,1408,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,1536,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,1664,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,1792,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,1920,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,2048,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,2304,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,2560,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,2816,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,3072,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,3328,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,3584,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,3840,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,4096,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,4608,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,5120,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,5632,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,6144,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,6656,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,7168,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,7680,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,8192,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,10240,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,12288,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,14336,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,16384,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,17408,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,24576,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,32768,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,40960,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,49152,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,57344,256,7168,256,8,0,0
f8_w8a8_channel,torch.float16,65536,256,7168,256,8,0,0
\ No newline at end of file
aiter/configs/untuned_fmoe_asm_w8a8_group.csv 0 → 100644
View file @ 85762c1a
quant_type,indtype,token,inter_dim,model_dim,expert,topk,q_size_n,q_size_k
f8_w8a8_block,torch.float16,1,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,2,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,3,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,4,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,5,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,6,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,7,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,8,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,9,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,10,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,11,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,12,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,13,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,14,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,15,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,16,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,17,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,18,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,20,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,24,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,28,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,32,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,34,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,36,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,40,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,44,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,48,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,56,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,64,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,68,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,72,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,80,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,88,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,96,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,104,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,112,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,128,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,144,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,160,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,192,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,224,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,256,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,320,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,384,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,448,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,512,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,576,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,640,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,704,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,768,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,832,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,896,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,960,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,1024,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,1152,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,1280,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,1408,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,1536,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,1664,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,1792,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,1920,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,2048,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,2304,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,2560,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,2816,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,3072,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,3328,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,3584,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,3840,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,4096,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,4608,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,5120,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,5632,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,6144,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,6656,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,7168,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,7680,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,8192,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,10240,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,12288,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,14336,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,16384,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,17408,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,24576,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,32768,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,40960,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,49152,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,57344,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,65536,128,7168,256,8,0,0
f8_w8a8_block,torch.float16,1,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,2,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,3,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,4,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,5,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,6,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,7,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,8,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,9,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,10,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,11,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,12,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,13,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,14,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,15,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,16,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,17,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,18,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,20,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,24,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,28,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,32,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,34,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,36,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,40,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,44,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,48,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,56,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,64,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,68,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,72,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,80,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,88,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,96,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,104,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,112,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,128,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,144,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,160,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,192,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,224,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,256,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,320,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,384,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,448,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,512,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,576,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,640,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,704,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,768,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,832,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,896,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,960,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,1024,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,1152,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,1280,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,1408,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,1536,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,1664,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,1792,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,1920,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,2048,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,2304,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,2560,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,2816,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,3072,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,3328,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,3584,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,3840,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,4096,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,4608,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,5120,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,5632,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,6144,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,6656,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,7168,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,7680,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,8192,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,10240,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,12288,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,14336,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,16384,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,17408,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,24576,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,32768,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,40960,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,49152,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,57344,256,7168,256,8,0,0
f8_w8a8_block,torch.float16,65536,256,7168,256,8,0,0
\ No newline at end of file
aiter/configs/untuned_fmoe_ck.csv 0 → 100644
View file @ 85762c1a
quant_type,indtype,token,inter_dim,model_dim,expert,topk,q_size_n,q_size_k
no_quant,torch.float16,1,256,8192,256,4,0,0
no_quant,torch.float16,8,256,8192,256,4,0,0
no_quant,torch.float16,16,256,8192,256,4,0,0
no_quant,torch.float16,24,256,8192,256,4,0,0
no_quant,torch.float16,32,256,8192,256,4,0,0
no_quant,torch.float16,48,256,8192,256,4,0,0
no_quant,torch.float16,64,256,8192,256,4,0,0
no_quant,torch.float16,96,256,8192,256,4,0,0
no_quant,torch.float16,128,256,8192,256,4,0,0
no_quant,torch.float16,256,256,8192,256,4,0,0
no_quant,torch.float16,512,256,8192,256,4,0,0
no_quant,torch.float16,1024,256,8192,256,4,0,0
no_quant,torch.float16,2048,256,8192,256,4,0,0
no_quant,torch.float16,4096,256,8192,256,4,0,0
no_quant,torch.float16,8192,256,8192,256,4,0,0
no_quant,torch.float16,1,256,7168,256,8,0,0
no_quant,torch.float16,8,256,7168,256,8,0,0
no_quant,torch.float16,16,256,7168,256,8,0,0
no_quant,torch.float16,24,256,7168,256,8,0,0
no_quant,torch.float16,32,256,7168,256,8,0,0
no_quant,torch.float16,48,256,7168,256,8,0,0
no_quant,torch.float16,64,256,7168,256,8,0,0
no_quant,torch.float16,96,256,7168,256,8,0,0
no_quant,torch.float16,128,256,7168,256,8,0,0
no_quant,torch.float16,256,256,7168,256,8,0,0
no_quant,torch.float16,512,256,7168,256,8,0,0
no_quant,torch.float16,1024,256,7168,256,8,0,0
no_quant,torch.float16,2048,256,7168,256,8,0,0
no_quant,torch.float16,4096,256,7168,256,8,0,0
no_quant,torch.float16,8192,256,7168,256,8,0,0
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment