Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,31 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
datasets:
|
| 3 |
+
- DILAB-HYU/KoQuality
|
| 4 |
+
language:
|
| 5 |
+
- ko
|
| 6 |
+
pipeline_tag: text-generation
|
| 7 |
+
tags:
|
| 8 |
+
- KoQuality
|
| 9 |
+
- llama
|
| 10 |
+
base_model: hyunseoki/ko-ref-llama2-7b
|
| 11 |
+
---
|
| 12 |
+
|
| 13 |
+
This model is a instruct-tuned hyunseoki/ko-ref-llama2-7b model using koquality.
|
| 14 |
+
|
| 15 |
+
|
| 16 |
+
## Training hyperparameters
|
| 17 |
+
- learning_rate: 5e-5
|
| 18 |
+
- train_batch_size: 1
|
| 19 |
+
- seed: 42
|
| 20 |
+
- distributed_type: multi-GPU (A30 24G) + CPU Offloading (384GB)
|
| 21 |
+
- num_devices: 2
|
| 22 |
+
- gradient_accumulation_steps: 32
|
| 23 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 24 |
+
- lr_scheduler_type: linear
|
| 25 |
+
- num_epochs: 2.0
|
| 26 |
+
|
| 27 |
+
## Framework versions
|
| 28 |
+
- Transformers 4.34.1
|
| 29 |
+
- Pytorch 2.0.1+cu117
|
| 30 |
+
- Datasets 2.11.0
|
| 31 |
+
- deepspeed 0.9.5
|