Update README.md
Browse files
README.md
CHANGED
|
@@ -23,31 +23,18 @@ After:
|
|
| 23 |
|
| 24 |
```
|
| 25 |
max_seq_length = 512
|
| 26 |
-
|
| 27 |
per_device_train_batch_size = 2,
|
| 28 |
-
|
| 29 |
gradient_accumulation_steps = 4,
|
| 30 |
-
|
| 31 |
warmup_steps = 2,
|
| 32 |
-
|
| 33 |
max_steps = 10,
|
| 34 |
-
|
| 35 |
learning_rate = 0.0005,
|
| 36 |
-
|
| 37 |
fp16 = not torch.cuda.is_bf16_supported(),
|
| 38 |
-
|
| 39 |
bf16 = torch.cuda.is_bf16_supported(),
|
| 40 |
-
|
| 41 |
logging_steps = 1,
|
| 42 |
-
|
| 43 |
optim = "adamw_8bit",
|
| 44 |
-
|
| 45 |
weight_decay = 0.01,
|
| 46 |
-
|
| 47 |
lr_scheduler_type = "linear",
|
| 48 |
-
|
| 49 |
seed = 3407,
|
| 50 |
-
|
| 51 |
output_dir = "outputs",
|
| 52 |
```
|
| 53 |
|
|
|
|
| 23 |
|
| 24 |
```
|
| 25 |
max_seq_length = 512
|
|
|
|
| 26 |
per_device_train_batch_size = 2,
|
|
|
|
| 27 |
gradient_accumulation_steps = 4,
|
|
|
|
| 28 |
warmup_steps = 2,
|
|
|
|
| 29 |
max_steps = 10,
|
|
|
|
| 30 |
learning_rate = 0.0005,
|
|
|
|
| 31 |
fp16 = not torch.cuda.is_bf16_supported(),
|
|
|
|
| 32 |
bf16 = torch.cuda.is_bf16_supported(),
|
|
|
|
| 33 |
logging_steps = 1,
|
|
|
|
| 34 |
optim = "adamw_8bit",
|
|
|
|
| 35 |
weight_decay = 0.01,
|
|
|
|
| 36 |
lr_scheduler_type = "linear",
|
|
|
|
| 37 |
seed = 3407,
|
|
|
|
| 38 |
output_dir = "outputs",
|
| 39 |
```
|
| 40 |
|