Add llama3 configurations with longer sequences
Browse files
inference-cache-config/trn1/llama3.json
CHANGED
|
@@ -6,6 +6,12 @@
|
|
| 6 |
"tensor_parallel_size": 2,
|
| 7 |
"instance_type" : "trn1"
|
| 8 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
{
|
| 10 |
"batch_size": 4,
|
| 11 |
"sequence_length": 4096,
|
|
@@ -18,6 +24,12 @@
|
|
| 18 |
"tensor_parallel_size": 2,
|
| 19 |
"instance_type" : "trn1"
|
| 20 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
{
|
| 22 |
"batch_size": 4,
|
| 23 |
"sequence_length": 4096,
|
|
@@ -110,6 +122,12 @@
|
|
| 110 |
"tensor_parallel_size": 24,
|
| 111 |
"instance_type" : "trn1"
|
| 112 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 113 |
{
|
| 114 |
"batch_size": 8,
|
| 115 |
"sequence_length": 4096,
|
|
@@ -122,6 +140,12 @@
|
|
| 122 |
"tensor_parallel_size": 32,
|
| 123 |
"instance_type" : "trn1"
|
| 124 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 125 |
{
|
| 126 |
"batch_size": 8,
|
| 127 |
"sequence_length": 4096,
|
|
|
|
| 6 |
"tensor_parallel_size": 2,
|
| 7 |
"instance_type" : "trn1"
|
| 8 |
},
|
| 9 |
+
{
|
| 10 |
+
"batch_size": 1,
|
| 11 |
+
"sequence_length": 16384,
|
| 12 |
+
"tensor_parallel_size": 2,
|
| 13 |
+
"instance_type" : "trn1"
|
| 14 |
+
},
|
| 15 |
{
|
| 16 |
"batch_size": 4,
|
| 17 |
"sequence_length": 4096,
|
|
|
|
| 24 |
"tensor_parallel_size": 2,
|
| 25 |
"instance_type" : "trn1"
|
| 26 |
},
|
| 27 |
+
{
|
| 28 |
+
"batch_size": 1,
|
| 29 |
+
"sequence_length": 16384,
|
| 30 |
+
"tensor_parallel_size": 8,
|
| 31 |
+
"instance_type" : "trn1"
|
| 32 |
+
},
|
| 33 |
{
|
| 34 |
"batch_size": 4,
|
| 35 |
"sequence_length": 4096,
|
|
|
|
| 122 |
"tensor_parallel_size": 24,
|
| 123 |
"instance_type" : "trn1"
|
| 124 |
},
|
| 125 |
+
{
|
| 126 |
+
"batch_size": 1,
|
| 127 |
+
"sequence_length": 16384,
|
| 128 |
+
"tensor_parallel_size": 2,
|
| 129 |
+
"instance_type" : "trn1"
|
| 130 |
+
},
|
| 131 |
{
|
| 132 |
"batch_size": 8,
|
| 133 |
"sequence_length": 4096,
|
|
|
|
| 140 |
"tensor_parallel_size": 32,
|
| 141 |
"instance_type" : "trn1"
|
| 142 |
},
|
| 143 |
+
{
|
| 144 |
+
"batch_size": 1,
|
| 145 |
+
"sequence_length": 16384,
|
| 146 |
+
"tensor_parallel_size": 32,
|
| 147 |
+
"instance_type" : "trn1"
|
| 148 |
+
},
|
| 149 |
{
|
| 150 |
"batch_size": 8,
|
| 151 |
"sequence_length": 4096,
|