use longer sequence length for llama3 on trn2
Browse files
inference-cache-config/trn2/llama3.json
CHANGED
|
@@ -6,6 +6,12 @@
|
|
| 6 |
"tensor_parallel_size": 4,
|
| 7 |
"instance_type" : "trn2"
|
| 8 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
{
|
| 10 |
"batch_size": 8,
|
| 11 |
"sequence_length": 4096,
|
|
@@ -32,6 +38,12 @@
|
|
| 32 |
"tensor_parallel_size" : 4,
|
| 33 |
"instance_type" : "trn2"
|
| 34 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 35 |
{
|
| 36 |
"batch_size": 4,
|
| 37 |
"sequence_length": 4096,
|
|
@@ -52,6 +64,12 @@
|
|
| 52 |
"tensor_parallel_size" : 4,
|
| 53 |
"instance_type" : "trn2"
|
| 54 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 55 |
{
|
| 56 |
"batch_size": 64,
|
| 57 |
"sequence_length": 4096,
|
|
@@ -80,6 +98,12 @@
|
|
| 80 |
"tensor_parallel_size": 64,
|
| 81 |
"instance_type" : "trn2"
|
| 82 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 83 |
{
|
| 84 |
"batch_size": 32,
|
| 85 |
"sequence_length": 4096,
|
|
|
|
| 6 |
"tensor_parallel_size": 4,
|
| 7 |
"instance_type" : "trn2"
|
| 8 |
},
|
| 9 |
+
{
|
| 10 |
+
"batch_size": 1,
|
| 11 |
+
"sequence_length": 16384,
|
| 12 |
+
"tensor_parallel_size": 4,
|
| 13 |
+
"instance_type" : "trn2"
|
| 14 |
+
},
|
| 15 |
{
|
| 16 |
"batch_size": 8,
|
| 17 |
"sequence_length": 4096,
|
|
|
|
| 38 |
"tensor_parallel_size" : 4,
|
| 39 |
"instance_type" : "trn2"
|
| 40 |
},
|
| 41 |
+
{
|
| 42 |
+
"batch_size": 1,
|
| 43 |
+
"sequence_length": 16384,
|
| 44 |
+
"tensor_parallel_size" : 4,
|
| 45 |
+
"instance_type" : "trn2"
|
| 46 |
+
},
|
| 47 |
{
|
| 48 |
"batch_size": 4,
|
| 49 |
"sequence_length": 4096,
|
|
|
|
| 64 |
"tensor_parallel_size" : 4,
|
| 65 |
"instance_type" : "trn2"
|
| 66 |
},
|
| 67 |
+
{
|
| 68 |
+
"batch_size": 1,
|
| 69 |
+
"sequence_length": 16384,
|
| 70 |
+
"tensor_parallel_size" : 4,
|
| 71 |
+
"instance_type" : "trn2"
|
| 72 |
+
},
|
| 73 |
{
|
| 74 |
"batch_size": 64,
|
| 75 |
"sequence_length": 4096,
|
|
|
|
| 98 |
"tensor_parallel_size": 64,
|
| 99 |
"instance_type" : "trn2"
|
| 100 |
},
|
| 101 |
+
{
|
| 102 |
+
"batch_size": 1,
|
| 103 |
+
"sequence_length": 16384,
|
| 104 |
+
"tensor_parallel_size": 64,
|
| 105 |
+
"instance_type" : "trn2"
|
| 106 |
+
},
|
| 107 |
{
|
| 108 |
"batch_size": 32,
|
| 109 |
"sequence_length": 4096,
|