alibidaran
commited on
Commit
•
94e3e6e
1
Parent(s):
7e8b44c
Update README.md
Browse files
README.md
CHANGED
@@ -53,7 +53,20 @@ with torch.no_grad():
|
|
53 |
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
|
54 |
|
55 |
```
|
56 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
57 |
|
58 |
|
59 |
|
|
|
53 |
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
|
54 |
|
55 |
```
|
56 |
+
## Training Parameters
|
57 |
+
per_device_train_batch_size=1,
|
58 |
+
gradient_accumulation_steps=8,
|
59 |
+
warmup_steps=2,
|
60 |
+
#max_steps=200,
|
61 |
+
|
62 |
+
num_train_epochs=1,
|
63 |
+
learning_rate=2e-4,
|
64 |
+
fp16=True,
|
65 |
+
logging_steps=100,
|
66 |
+
output_dir="outputs",
|
67 |
+
optim="paged_adamw_8bit",
|
68 |
+
save_steps=500,
|
69 |
+
ddp_find_unused_parameters=False // for training on multiple GPU
|
70 |
|
71 |
|
72 |
|