22 lines
549 B
Markdown
22 lines
549 B
Markdown
|
---
|
||
|
|
license: apache-2.0
|
||
|
|
---
|
||
|
|
## Model
|
||
|
|
- base_model : yanolja/KoSOLAR-10.7B-v0.2
|
||
|
|
- training objective: freeze, instruction Tuning
|
||
|
|
## Dataset
|
||
|
|
공개 데이터 수집
|
||
|
|
- Deduplicating Training Data Makes Language Models Better 알고리즘 활용
|
||
|
|
- instruction version 1.4
|
||
|
|
|
||
|
|
## Code
|
||
|
|
```python
|
||
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||
|
|
import torch
|
||
|
|
|
||
|
|
model_name = "jjingyeom/freeze_KoSoLAR-10.7B-v0.2_1.4_dedup"
|
||
|
|
model = AutoModelForCausalLM.from_pretrained(
|
||
|
|
model_name,
|
||
|
|
)
|
||
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
||
|
|
```
|