Update README.md
Browse files
README.md
CHANGED
|
@@ -33,7 +33,31 @@ Prompt format is Alpaca. I used the same system prompt as the original Samantha.
|
|
| 33 |
### Response:
|
| 34 |
"""
|
| 35 |
```
|
| 36 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 37 |
|
| 38 |
2 epoch finetuning from llama-3-8b took 1 hour on a single A100 with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
|
| 39 |
|
|
|
|
| 33 |
### Response:
|
| 34 |
"""
|
| 35 |
```
|
| 36 |
+
|
| 37 |
+
# Training
|
| 38 |
+
|
| 39 |
+
[gptchain](https://github.com/RuslanPeresy/gptchain) framework has been used for training.
|
| 40 |
+
|
| 41 |
+
## Training hyperparameters
|
| 42 |
+
|
| 43 |
+
- learning_rate: 2e-4
|
| 44 |
+
- seed: 3407
|
| 45 |
+
- gradient_accumulation_steps: 4
|
| 46 |
+
- per_device_train_batch_size: 2
|
| 47 |
+
- optimizer: adamw_8bit
|
| 48 |
+
- lr_scheduler_type: linear
|
| 49 |
+
- warmup_steps: 5
|
| 50 |
+
- num_epochs: 2
|
| 51 |
+
- weight_decay: 0.01
|
| 52 |
+
|
| 53 |
+
## Training results
|
| 54 |
+
|
| 55 |
+
|Training Loss | Epoch | Step |
|
| 56 |
+
|--------------|-------|------|
|
| 57 |
+
|2.0778 |0.0 |1 |
|
| 58 |
+
|0.6255 |0.18 |120 |
|
| 59 |
+
|0.6208 |0.94 |620 |
|
| 60 |
+
|0.6244 |2.0 |1306 |
|
| 61 |
|
| 62 |
2 epoch finetuning from llama-3-8b took 1 hour on a single A100 with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
|
| 63 |
|