squarelike
commited on
Commit
β’
9f5005c
1
Parent(s):
6547be7
Update README.md
Browse files
README.md
CHANGED
@@ -8,7 +8,7 @@ language:
|
|
8 |
pipeline_tag: translation
|
9 |
---
|
10 |
|
11 |
-
# Gugugo-koen-7B-V1.1-
|
12 |
Detail repo: [https://github.com/jwj7140/Gugugo](https://github.com/jwj7140/Gugugo)
|
13 |
![Gugugo](./logo.png)
|
14 |
|
@@ -30,53 +30,4 @@ I trained with 1x A6000 GPUs for 90 hours.
|
|
30 |
```
|
31 |
### μμ΄: {sentence}</λ>
|
32 |
### νκ΅μ΄:
|
33 |
-
```
|
34 |
-
|
35 |
-
## **Implementation Code**
|
36 |
-
```python
|
37 |
-
from transformers import AutoModelForCausalLM, AutoTokenizer, StoppingCriteria, StoppingCriteriaList
|
38 |
-
import torch
|
39 |
-
repo = "squarelike/Gugugo-koen-7B-V1.1"
|
40 |
-
model = AutoModelForCausalLM.from_pretrained(
|
41 |
-
repo,
|
42 |
-
load_in_4bit=True
|
43 |
-
device_map='auto'
|
44 |
-
)
|
45 |
-
tokenizer = AutoTokenizer.from_pretrained(repo)
|
46 |
-
|
47 |
-
class StoppingCriteriaSub(StoppingCriteria):
|
48 |
-
def __init__(self, stops = [], encounters=1):
|
49 |
-
super().__init__()
|
50 |
-
self.stops = [stop for stop in stops]
|
51 |
-
|
52 |
-
def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor):
|
53 |
-
for stop in self.stops:
|
54 |
-
if torch.all((stop == input_ids[0][-len(stop):])).item():
|
55 |
-
return True
|
56 |
-
|
57 |
-
return False
|
58 |
-
|
59 |
-
stop_words_ids = torch.tensor([[829, 45107, 29958], [1533, 45107, 29958], [829, 45107, 29958], [21106, 45107, 29958]]).to("cuda")
|
60 |
-
stopping_criteria = StoppingCriteriaList([StoppingCriteriaSub(stops=stop_words_ids)])
|
61 |
-
|
62 |
-
def gen(lan="en", x=""):
|
63 |
-
if (lan == "ko"):
|
64 |
-
prompt = f"### νκ΅μ΄: {x}</λ>\n### μμ΄:"
|
65 |
-
else:
|
66 |
-
prompt = f"### μμ΄: {x}</λ>\n### νκ΅μ΄:"
|
67 |
-
gened = model.generate(
|
68 |
-
**tokenizer(
|
69 |
-
prompt,
|
70 |
-
return_tensors='pt',
|
71 |
-
return_token_type_ids=False
|
72 |
-
).to("cuda"),
|
73 |
-
max_new_tokens=2000,
|
74 |
-
temperature=0.1,
|
75 |
-
do_sample=True,
|
76 |
-
stopping_criteria=stopping_criteria
|
77 |
-
)
|
78 |
-
return tokenizer.decode(gened[0][1:]).replace(prompt+" ", "").replace("</λ>", "")
|
79 |
-
|
80 |
-
|
81 |
-
print(gen(lan="en", x="Hello, world!"))
|
82 |
```
|
|
|
8 |
pipeline_tag: translation
|
9 |
---
|
10 |
|
11 |
+
# Gugugo-koen-7B-V1.1-GGUF
|
12 |
Detail repo: [https://github.com/jwj7140/Gugugo](https://github.com/jwj7140/Gugugo)
|
13 |
![Gugugo](./logo.png)
|
14 |
|
|
|
30 |
```
|
31 |
### μμ΄: {sentence}</λ>
|
32 |
### νκ΅μ΄:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
```
|