Upload models/scripts/zero2_fp8.json with huggingface_hub
Browse files
models/scripts/zero2_fp8.json
ADDED
|
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"fp16": {
|
| 3 |
+
"enabled": "auto",
|
| 4 |
+
"loss_scale": 0,
|
| 5 |
+
"loss_scale_window": 1000,
|
| 6 |
+
"initial_scale_power": 16,
|
| 7 |
+
"hysteresis": 2,
|
| 8 |
+
"min_loss_scale": 1
|
| 9 |
+
},
|
| 10 |
+
"bf16": {
|
| 11 |
+
"enabled": "auto"
|
| 12 |
+
},
|
| 13 |
+
"msamp": {
|
| 14 |
+
"enabled": true,
|
| 15 |
+
"opt_level": "O3",
|
| 16 |
+
"use_te": false
|
| 17 |
+
},
|
| 18 |
+
"train_micro_batch_size_per_gpu": "auto",
|
| 19 |
+
"train_batch_size": "auto",
|
| 20 |
+
"gradient_accumulation_steps": "auto",
|
| 21 |
+
"zero_optimization": {
|
| 22 |
+
"stage": 2,
|
| 23 |
+
"overlap_comm": true,
|
| 24 |
+
"contiguous_gradients": true,
|
| 25 |
+
"sub_group_size": 1e9,
|
| 26 |
+
"reduce_bucket_size": "auto"
|
| 27 |
+
}
|
| 28 |
+
}
|