Update README.md
Browse files
README.md
CHANGED
|
@@ -9,4 +9,43 @@ language:
|
|
| 9 |
pipeline_tag: text-generation
|
| 10 |
---
|
| 11 |
|
|
|
|
| 12 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
pipeline_tag: text-generation
|
| 10 |
---
|
| 11 |
|
| 12 |
+
# ver_4.1_sft
|
| 13 |
|
| 14 |
+
This model is a fine-tuned version of [yanolja/EEVE-Korean-10.8B-v1.0](https://huggingface.co/yanolja/EEVE-Korean-10.8B-v1.0) on the Custom dataset.
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
## Training procedure
|
| 18 |
+
|
| 19 |
+
### Training hyperparameters
|
| 20 |
+
|
| 21 |
+
The following hyperparameters were used during training:
|
| 22 |
+
- learning_rate: 5e-05
|
| 23 |
+
- train_batch_size: 1
|
| 24 |
+
- eval_batch_size: 1
|
| 25 |
+
- seed: 42
|
| 26 |
+
- distributed_type: multi-GPU
|
| 27 |
+
- num_devices: 8
|
| 28 |
+
- total_train_batch_size: 8
|
| 29 |
+
- total_eval_batch_size: 8
|
| 30 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 31 |
+
- lr_scheduler_type: cosine
|
| 32 |
+
- lr_scheduler_warmup_ratio: 0.1
|
| 33 |
+
- training_steps: 0
|
| 34 |
+
|
| 35 |
+
### Training results
|
| 36 |
+
|
| 37 |
+
| Training Loss | Epoch | Step |
|
| 38 |
+
|:-------------:|:-----:|:----:|
|
| 39 |
+
| 0.7739 | 0.2 | 1124 |
|
| 40 |
+
| 0.7214 | 0.4 | 2248 |
|
| 41 |
+
| 0.6832 | 0.6 | 3372 |
|
| 42 |
+
| 0.6935 | 0.8 | 4496 |
|
| 43 |
+
|
| 44 |
+
|
| 45 |
+
### Framework versions
|
| 46 |
+
|
| 47 |
+
- PEFT 0.10.0
|
| 48 |
+
- Transformers 4.39.3
|
| 49 |
+
- Pytorch 2.0.1
|
| 50 |
+
- Datasets 2.18.0
|
| 51 |
+
- Tokenizers 0.15.2
|