Update README.md
Browse files
README.md
CHANGED
|
@@ -40,6 +40,25 @@ Step Training Loss Validation Loss<br>
|
|
| 40 |
275 0.931000 0.960848<br>
|
| 41 |
300 0.932000 0.958946 <-- picked checkpoint <br>
|
| 42 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 43 |
### Model Description
|
| 44 |
|
| 45 |
|
|
|
|
| 40 |
275 0.931000 0.960848<br>
|
| 41 |
300 0.932000 0.958946 <-- picked checkpoint <br>
|
| 42 |
|
| 43 |
+
### Training Parameters
|
| 44 |
+
per_device_train_batch_size = 4,<br>
|
| 45 |
+
gradient_accumulation_steps = 16,<br>
|
| 46 |
+
num_train_epochs=3,<br>
|
| 47 |
+
warmup_steps = 5,<br>
|
| 48 |
+
learning_rate = 3e-5,<br>
|
| 49 |
+
logging_steps = 25,<br>
|
| 50 |
+
optim = "adamw_8bit",<br>
|
| 51 |
+
weight_decay = 0.01,<br>
|
| 52 |
+
lr_scheduler_type = "linear",<br>
|
| 53 |
+
seed = 3407,<br>
|
| 54 |
+
per_device_eval_batch_size = 2,<br>
|
| 55 |
+
eval_strategy="steps",<br>
|
| 56 |
+
eval_accumulation_steps = 32,<br>
|
| 57 |
+
eval_steps = 25,<br>
|
| 58 |
+
eval_delay = 0,<br>
|
| 59 |
+
save_strategy="steps",<br>
|
| 60 |
+
save_steps=50,<br>
|
| 61 |
+
|
| 62 |
### Model Description
|
| 63 |
|
| 64 |
|