github.com/NVIDIA/aistore@v1.3.23-0.20240517131212-7df6609be51d/bench/microbenchmarks/dl/testrc (about) 1 # 2 # Sample test config file. Test config files are sourced by the control script, 3 # and all variables named rn50_* are processed for testing. Each such set of 4 # of tests begins with configuration assignments (node, gpupernode, [fillorder]) 5 # followed by dataloader test cases to run for that configuration. 6 # 7 # Columns for each test set are: 8 # 9 # model training|inference|both|- (- to disable) 10 # data-backend will be used as the --data-backend argument in main.py 11 # context if context is say xyz then will expand to value of ${config_xyz}, interpreted per dataloader type 12 # webdataset: AIS gateway 13 # datapath the final argument to main.py - the dataset filesystem path or bucket name 14 # batchsize the batch size to use on the main.py cmdline 15 # cast fp16, amp or none for --fp16, --amp or no model cast 16 # epochs number of epochs to time for 17 # iterations maximum number of iterations per epochs 18 # extra extra for backend; for webdataset this is bucket name 19 # 20 # Note: the quotes are essential! 21 # 22 23 declare -a enabled_sets=( 24 "rn50_red" 25 "rn50_pink" 26 "rn50_bluelite" 27 "rn50_green" 28 "rn50_yellow" 29 ) 30 31 rn50_red="$(cat <<-EOM 32 config_worlds=1x1,1x2,1x4,1x8 33 config_fillorder=widthfirst 34 config_ais12=http://demo-ais-proxy:51080 35 config_ais1=http://10.132.181.10:51080 36 config_shardpat_std=\"imagenet_train-{0000..0147}.tar\" 37 config_shardpat_inflated_std=\"imagenet_train-{0000..9999}.tar\" 38 config_shardpat_augmented=\"imagenet_train-{0000..0147}-{000..019}.tar\" 39 config_shardpat_inflated_augmented=\"imagenet_train-{0000..9999}-{000..019}.tar\" 40 both pytorch - /data/imagenet 256 fp16 1 100 - 41 both dali-cpu - /data/imagenet 256 fp16 1 100 - 42 both dali-gpu - /data/imagenet 256 fp16 1 100 - 43 EOM 44 )" 45 46 rn50_pink="$(cat <<-EOM 47 config_worlds=1x1,1x2,1x4,1x8 48 config_fillorder=widthfirst 49 config_ais12=http://demo-ais-proxy:51080 50 config_ais1=http://10.132.181.10:51080 51 config_shardpat_std=\"imagenet_train-{0000..0147}.tar\" 52 config_shardpat_inflated_std=\"imagenet_train-{0000..9999}.tar\" 53 config_shardpat_augmented=\"imagenet_train-{0000..0147}-{000..019}.tar\" 54 config_shardpat_inflated_augmented=\"imagenet_train-{0000..9999}-{000..019}.tar\" 55 both pytorch - /data/imagenet-inflated 256 fp16 1 100 - 56 both dali-cpu - /data/imagenet-inflated 256 fp16 1 100 - 57 both dali-gpu - /data/imagenet-inflated 256 fp16 1 100 - 58 EOM 59 )" 60 61 rn50_bluelite="$(cat <<-EOM 62 config_worlds=1x1,1x2,1x4,1x8,2x8,4x8,8x8 63 config_fillorder=widthfirst 64 config_ais12=http://demo-ais-proxy:51080 65 config_ais1=http://10.132.181.10:51080 66 config_shardpat_std=\"imagenet_train-{0000..0147}.tar\" 67 config_shardpat_inflated_std=\"imagenet_train-{0000..9999}.tar\" 68 config_shardpat_augmented=\"imagenet_train-{0000..0147}-{000..019}.tar\" 69 config_shardpat_inflated_augmented=\"imagenet_train-{0000..9999}-{000..019}.tar\" 70 both pytorch - /nfs/imagenet 256 fp16 1 100 - 71 EOM 72 )" 73 74 rn50_green="$(cat <<-EOM 75 config_worlds=1x1,1x2,1x4,1x8,2x8,4x8,8x8 76 config_fillorder=widthfirst 77 config_ais12=http://demo-ais-proxy:51080 78 config_ais1=http://10.132.181.10:51080 79 config_shardpat_std=\"imagenet_train-{0000..0147}.tar\" 80 config_shardpat_inflated_std=\"imagenet_train-{0000..9999}.tar\" 81 config_shardpat_augmented=\"imagenet_train-{0000..0147}-{000..019}.tar\" 82 config_shardpat_inflated_augmented=\"imagenet_train-{0000..9999}-{000..019}.tar\" 83 both webdataset ais1 lpr-imagenet-uncompressed 256 fp16 1 100 shardpat_std 84 EOM 85 )" 86 87 rn50_yellow="$(cat <<-EOM 88 config_worlds=1x1,1x2,1x4,1x8,2x8,4x8,8x8 89 config_fillorder=widthfirst 90 config_ais12=http://demo-ais-proxy:51080 91 config_ais1=http://10.132.181.10:51080 92 config_shardpat_std=\"imagenet_train-{0000..0147}.tar\" 93 config_shardpat_inflated_std=\"imagenet_train-{0000..9999}.tar\" 94 config_shardpat_augmented=\"imagenet_train-{0000..0147}-{000..019}.tar\" 95 config_shardpat_inflated_augmented=\"imagenet_train-{0000..9999}-{000..019}.tar\" 96 both webdataset ais12 lpr-imagenet-inflated 256 fp16 1 100 shardpat_inflated_std 97 EOM 98 )" 99 100 rn50_testset1="$(cat <<-EOM 101 config_nodes=1,2,4,8 102 config_gpupernode=1,8 103 config_fillorder=widthfirst 104 config_ais12=http://demo-ais-proxy:51080 105 config_ais1=http://10.132.181.10:51080 106 config_shardpat_std=\"imagenet_train-{0000..0147}.tar\" 107 config_shardpat_inflated_std=\"imagenet_train-{0000..9999}.tar\" 108 config_shardpat_augmented=\"imagenet_train-{0000..0147}-{000..019}.tar\" 109 config_shardpat_inflated_augmented=\"imagenet_train-{0000..9999}-{000..019}.tar\" 110 - dali-cpu - /data/imagenet 64 fp16 1 100 - 111 - dali-gpu - /data/imagenet 64 fp16 1 100 - 112 - dali-cpu - /nfs/imagenet 64 fp16 1 100 - 113 - dali-gpu - /nfs/imagenet 64 fp16 1 100 - 114 - pytorch - /nfs/imagenet-inflated 256 fp16 1 100 - 115 both pytorch - /nfs/imagenet 256 fp16 1 100 - 116 both pytorch - /data/imagenet 256 fp16 1 100 - 117 both pytorch - /data/imagenet-inflated 256 fp16 1 100 - 118 both webdataset-pre ais12 lpr-imagenet-augmented-uncompressed 256 fp16 1 100 shardpat_augmented 119 both webdataset ais12 lpr-imagenet-uncompressed 256 fp16 1 100 shardpat_std 120 both webdataset-pre ais12 lpr-imagenet-augmented-replicated 256 fp16 1 100 shardpat_inflated_augmented 121 both webdataset ais1 lpr-imagenet-uncompressed 256 fp16 1 100 shardpat_std 122 both webdataset-pre ais1 lpr-imagenet-augmented-uncompressed 256 fp16 1 100 shardpat_augmented 123 both webdataset-pre ais1 lpr-imagenet-augmented-replicated 256 fp16 1 100 shardpat_inflated_augmented 124 - dali-cpu - /nfs/imagenet-inflated 64 fp16 1 100 - 125 - dali-gpu - /nfs/imagenet-inflated 64 fp16 1 100 - 126 EOM 127 )"