-
Notifications
You must be signed in to change notification settings - Fork 2
/
benchmarks_50346.out
198 lines (165 loc) · 8.66 KB
/
benchmarks_50346.out
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
1.13.1
[2023-11-16 17:46:38,010] [INFO] [distributed.py:36:init_distributed] Not using the DeepSpeed or torch.distributed launchers, attempting to detect MPI environment...
[2023-11-16 17:46:38,911] [INFO] [distributed.py:83:mpi_discovery] Discovered MPI settings of world_rank=0, local_rank=0, world_size=1, master_addr=26.0.148.159, master_port=6000
[2023-11-16 17:46:38,911] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2023-11-16 17:46:41,931] [INFO] [checkpointing.py:223:model_parallel_cuda_manual_seed] > initializing model parallel cuda seeds on global rank 0, model parallel rank 0, and data parallel rank 0 with model parallel seed: 3952 and data parallel seed: 1234
num_attention_heads: 128, hidden_size: 1024, train_micro_batch_size_per_gpu: 4, tensor_mp_size: 1, pipeline_mp_size: 1, dp_size: 1
Actual
------
Attention duration (in seconds): 0.0081
Attention throughput (in TFLOP/s): 17.027
MLP duration (in seconds): 0.0008
MLP throughput (in TFLOP/s): 182.309
Transformer duration (in seconds): 0.0119
Transformer throughput (in TFLOP/s): 23.121
========================================================================================================================
num_attention_heads: 128, hidden_size: 2048, train_micro_batch_size_per_gpu: 4, tensor_mp_size: 1, pipeline_mp_size: 1, dp_size: 1
Actual
------
Attention duration (in seconds): 0.0113
Attention throughput (in TFLOP/s): 36.332
MLP duration (in seconds): 0.0026
MLP throughput (in TFLOP/s): 213.031
Transformer duration (in seconds): 0.0137
Transformer throughput (in TFLOP/s): 70.045
========================================================================================================================
num_attention_heads: 128, hidden_size: 3072, train_micro_batch_size_per_gpu: 4, tensor_mp_size: 1, pipeline_mp_size: 1, dp_size: 1
Actual
------
Attention duration (in seconds): 0.0116
Attention throughput (in TFLOP/s): 70.844
MLP duration (in seconds): 0.0058
MLP throughput (in TFLOP/s): 214.599
Transformer duration (in seconds): 0.0190
Transformer throughput (in TFLOP/s): 108.363
========================================================================================================================
num_attention_heads: 128, hidden_size: 4096, train_micro_batch_size_per_gpu: 4, tensor_mp_size: 1, pipeline_mp_size: 1, dp_size: 1
Actual
------
Attention duration (in seconds): 0.0200
Attention throughput (in TFLOP/s): 68.625
MLP duration (in seconds): 0.0096
MLP throughput (in TFLOP/s): 228.121
Transformer duration (in seconds): 0.0317
Transformer throughput (in TFLOP/s): 112.557
========================================================================================================================
num_attention_heads: 128, hidden_size: 5120, train_micro_batch_size_per_gpu: 4, tensor_mp_size: 1, pipeline_mp_size: 1, dp_size: 1
Actual
------
Attention duration (in seconds): 0.0277
Attention throughput (in TFLOP/s): 74.417
MLP duration (in seconds): 0.0141
MLP throughput (in TFLOP/s): 243.298
Transformer duration (in seconds): 0.0424
Transformer throughput (in TFLOP/s): 129.550
========================================================================================================================
num_attention_heads: 128, hidden_size: 6144, train_micro_batch_size_per_gpu: 4, tensor_mp_size: 1, pipeline_mp_size: 1, dp_size: 1
Actual
------
Attention duration (in seconds): 0.0317
Attention throughput (in TFLOP/s): 91.151
MLP duration (in seconds): 0.0202
MLP throughput (in TFLOP/s): 244.987
Transformer duration (in seconds): 0.0474
Transformer throughput (in TFLOP/s): 165.404
========================================================================================================================
num_attention_heads: 128, hidden_size: 7168, train_micro_batch_size_per_gpu: 4, tensor_mp_size: 1, pipeline_mp_size: 1, dp_size: 1
Actual
------
Attention duration (in seconds): 0.0376
Attention throughput (in TFLOP/s): 102.440
MLP duration (in seconds): 0.0272
MLP throughput (in TFLOP/s): 247.391
Transformer duration (in seconds): 0.0532
Transformer throughput (in TFLOP/s): 198.935
========================================================================================================================
num_attention_heads: 128, hidden_size: 8192, train_micro_batch_size_per_gpu: 4, tensor_mp_size: 1, pipeline_mp_size: 1, dp_size: 1
Actual
------
Attention duration (in seconds): 0.0418
Attention throughput (in TFLOP/s): 118.484
MLP duration (in seconds): 0.0361
MLP throughput (in TFLOP/s): 243.402
Transformer duration (in seconds): 0.0680
Transformer throughput (in TFLOP/s): 202.184
========================================================================================================================
num_attention_heads: 128, hidden_size: 9216, train_micro_batch_size_per_gpu: 4, tensor_mp_size: 1, pipeline_mp_size: 1, dp_size: 1
Actual
------
Attention duration (in seconds): 0.0474
Attention throughput (in TFLOP/s): 130.496
MLP duration (in seconds): 0.0458
MLP throughput (in TFLOP/s): 243.321
Transformer duration (in seconds): 0.0861
Transformer throughput (in TFLOP/s): 201.131
========================================================================================================================
num_attention_heads: 128, hidden_size: 10240, train_micro_batch_size_per_gpu: 4, tensor_mp_size: 1, pipeline_mp_size: 1, dp_size: 1
Actual
------
Attention duration (in seconds): 0.0471
Attention throughput (in TFLOP/s): 160.579
MLP duration (in seconds): 0.0555
MLP throughput (in TFLOP/s): 247.716
Transformer duration (in seconds): 0.1013
Transformer throughput (in TFLOP/s): 210.391
========================================================================================================================
num_attention_heads: 128, hidden_size: 11264, train_micro_batch_size_per_gpu: 4, tensor_mp_size: 1, pipeline_mp_size: 1, dp_size: 1
Actual
------
Attention duration (in seconds): 0.0524
Attention throughput (in TFLOP/s): 173.033
MLP duration (in seconds): 0.0672
MLP throughput (in TFLOP/s): 247.515
Transformer duration (in seconds): 0.1203
Transformer throughput (in TFLOP/s): 213.573
========================================================================================================================
num_attention_heads: 128, hidden_size: 12288, train_micro_batch_size_per_gpu: 4, tensor_mp_size: 1, pipeline_mp_size: 1, dp_size: 1
Actual
------
Attention duration (in seconds): 0.0600
Attention throughput (in TFLOP/s): 178.688
MLP duration (in seconds): 0.0788
MLP throughput (in TFLOP/s): 251.122
Transformer duration (in seconds): 0.1387
Transformer throughput (in TFLOP/s): 220.040
========================================================================================================================
num_attention_heads: 128, hidden_size: 13312, train_micro_batch_size_per_gpu: 4, tensor_mp_size: 1, pipeline_mp_size: 1, dp_size: 1
Actual
------
Attention duration (in seconds): 0.0603
Attention throughput (in TFLOP/s): 207.420
MLP duration (in seconds): 0.0926
MLP throughput (in TFLOP/s): 250.856
Transformer duration (in seconds): 0.1587
Transformer throughput (in TFLOP/s): 225.228
========================================================================================================================
num_attention_heads: 128, hidden_size: 14336, train_micro_batch_size_per_gpu: 4, tensor_mp_size: 1, pipeline_mp_size: 1, dp_size: 1
Actual
------
Attention duration (in seconds): 0.0705
Attention throughput (in TFLOP/s): 204.652
MLP duration (in seconds): 0.1079
MLP throughput (in TFLOP/s): 249.742
Transformer duration (in seconds): 0.1833
Transformer throughput (in TFLOP/s): 225.752
========================================================================================================================
num_attention_heads: 128, hidden_size: 15360, train_micro_batch_size_per_gpu: 4, tensor_mp_size: 1, pipeline_mp_size: 1, dp_size: 1
Actual
------
Attention duration (in seconds): 0.0800
Attention throughput (in TFLOP/s): 206.148
MLP duration (in seconds): 0.1236
MLP throughput (in TFLOP/s): 250.092
Transformer duration (in seconds): 0.2085
Transformer throughput (in TFLOP/s): 227.465
========================================================================================================================
num_attention_heads: 128, hidden_size: 16384, train_micro_batch_size_per_gpu: 4, tensor_mp_size: 1, pipeline_mp_size: 1, dp_size: 1
Actual
------
Attention duration (in seconds): 0.0847
Attention throughput (in TFLOP/s): 220.792
MLP duration (in seconds): 0.1413
MLP throughput (in TFLOP/s): 249.029
Transformer duration (in seconds): 0.2326
Transformer throughput (in TFLOP/s): 231.596
========================================================================================================================