Isaak Carter Augustus
commited on
Commit
•
1aba652
1
Parent(s):
d39bf60
Update README.md
Browse files
README.md
CHANGED
|
@@ -12,6 +12,8 @@ tags:
|
|
| 12 |
- sft
|
| 13 |
---
|
| 14 |
|
|
|
|
|
|
|
| 15 |
# Uploaded model
|
| 16 |
|
| 17 |
- **Developed by:** Isaak-Carter
|
|
@@ -20,4 +22,12 @@ tags:
|
|
| 20 |
|
| 21 |
This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
|
| 22 |
|
| 23 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
- sft
|
| 13 |
---
|
| 14 |
|
| 15 |
+
## Still in Beta!!!
|
| 16 |
+
|
| 17 |
# Uploaded model
|
| 18 |
|
| 19 |
- **Developed by:** Isaak-Carter
|
|
|
|
| 22 |
|
| 23 |
This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
|
| 24 |
|
| 25 |
+
```json
|
| 26 |
+
"per_device_train_batch_size": 2,
|
| 27 |
+
"gradient_accumulation_steps": 8,
|
| 28 |
+
"max_steps": 100,
|
| 29 |
+
"learning_rate": 2e-4,
|
| 30 |
+
"optim": "adamw_8bit",
|
| 31 |
+
"weight_decay": 0.01,
|
| 32 |
+
"lr_scheduler_type": "cosine"
|
| 33 |
+
```
|