New Base IPU config
#2
by
thorinf
- opened
- ipu_config.json +9 -8
ipu_config.json
CHANGED
@@ -1,15 +1,16 @@
|
|
1 |
{
|
2 |
-
"optimizer_state_offchip":
|
3 |
"replicated_tensor_sharding": true,
|
|
|
4 |
"enable_half_partials": true,
|
5 |
"recompute_checkpoint_every_layer": false,
|
6 |
-
"device_iterations":
|
7 |
"inference_device_iterations": 4,
|
8 |
-
"replication_factor":
|
9 |
-
"inference_replication_factor":
|
10 |
-
"gradient_accumulation_steps":
|
11 |
"executable_cache_dir": "./exe_cache",
|
12 |
-
"ipus_per_replica":
|
13 |
-
"layers_per_ipu": [
|
14 |
-
"matmul_proportion": [0.
|
15 |
}
|
|
|
1 |
{
|
2 |
+
"optimizer_state_offchip": true,
|
3 |
"replicated_tensor_sharding": true,
|
4 |
+
"enable_half_first_order_momentum": true,
|
5 |
"enable_half_partials": true,
|
6 |
"recompute_checkpoint_every_layer": false,
|
7 |
+
"device_iterations": 2,
|
8 |
"inference_device_iterations": 4,
|
9 |
+
"replication_factor": 2,
|
10 |
+
"inference_replication_factor": 2,
|
11 |
+
"gradient_accumulation_steps": 16,
|
12 |
"executable_cache_dir": "./exe_cache",
|
13 |
+
"ipus_per_replica": 8,
|
14 |
+
"layers_per_ipu": [2, 1, 4, 3, 3, 3, 3, 0],
|
15 |
+
"matmul_proportion": [0.1, 0.1, 0.1, 0.3, 0.3, 0.3, 0.3, 0.1]
|
16 |
}
|