Update app.py
Browse files
app.py
CHANGED
|
@@ -16,7 +16,79 @@ MAX_MAX_NEW_TOKENS = 2048
|
|
| 16 |
DEFAULT_MAX_NEW_TOKENS = 1024
|
| 17 |
total_count=0
|
| 18 |
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
|
| 21 |
DESCRIPTION="""CODE"""
|
| 22 |
|
|
@@ -43,6 +115,7 @@ def gen(
|
|
| 43 |
print(total_count)
|
| 44 |
os.system("nvidia-smi")
|
| 45 |
conversation = []
|
|
|
|
| 46 |
if system_prompt:
|
| 47 |
conversation.append({"role": "system", "content": system_prompt})
|
| 48 |
for user, assistant in chat_history:
|
|
|
|
| 16 |
DEFAULT_MAX_NEW_TOKENS = 1024
|
| 17 |
total_count=0
|
| 18 |
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
|
| 19 |
+
import gradio as gr
|
| 20 |
+
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
|
| 21 |
+
|
| 22 |
+
dict_map = {
|
| 23 |
+
"òa": "oà",
|
| 24 |
+
"Òa": "Oà",
|
| 25 |
+
"ÒA": "OÀ",
|
| 26 |
+
"óa": "oá",
|
| 27 |
+
"Óa": "Oá",
|
| 28 |
+
"ÓA": "OÁ",
|
| 29 |
+
"ỏa": "oả",
|
| 30 |
+
"Ỏa": "Oả",
|
| 31 |
+
"ỎA": "OẢ",
|
| 32 |
+
"õa": "oã",
|
| 33 |
+
"Õa": "Oã",
|
| 34 |
+
"ÕA": "OÃ",
|
| 35 |
+
"ọa": "oạ",
|
| 36 |
+
"Ọa": "Oạ",
|
| 37 |
+
"ỌA": "OẠ",
|
| 38 |
+
"òe": "oè",
|
| 39 |
+
"Òe": "Oè",
|
| 40 |
+
"ÒE": "OÈ",
|
| 41 |
+
"óe": "oé",
|
| 42 |
+
"Óe": "Oé",
|
| 43 |
+
"ÓE": "OÉ",
|
| 44 |
+
"ỏe": "oẻ",
|
| 45 |
+
"Ỏe": "Oẻ",
|
| 46 |
+
"ỎE": "OẺ",
|
| 47 |
+
"õe": "oẽ",
|
| 48 |
+
"Õe": "Oẽ",
|
| 49 |
+
"ÕE": "OẼ",
|
| 50 |
+
"ọe": "oẹ",
|
| 51 |
+
"Ọe": "Oẹ",
|
| 52 |
+
"ỌE": "OẸ",
|
| 53 |
+
"ùy": "uỳ",
|
| 54 |
+
"Ùy": "Uỳ",
|
| 55 |
+
"ÙY": "UỲ",
|
| 56 |
+
"úy": "uý",
|
| 57 |
+
"Úy": "Uý",
|
| 58 |
+
"ÚY": "UÝ",
|
| 59 |
+
"ủy": "uỷ",
|
| 60 |
+
"Ủy": "Uỷ",
|
| 61 |
+
"ỦY": "UỶ",
|
| 62 |
+
"ũy": "uỹ",
|
| 63 |
+
"Ũy": "Uỹ",
|
| 64 |
+
"ŨY": "UỸ",
|
| 65 |
+
"ụy": "uỵ",
|
| 66 |
+
"Ụy": "Uỵ",
|
| 67 |
+
"ỤY": "UỴ",
|
| 68 |
+
}
|
| 69 |
|
| 70 |
+
tokenizer_vi2en = AutoTokenizer.from_pretrained("vinai/vinai-translate-vi2en-v2", src_lang="vi_VN")
|
| 71 |
+
model_vi2en = AutoModelForSeq2SeqLM.from_pretrained("vinai/vinai-translate-vi2en-v2")
|
| 72 |
+
|
| 73 |
+
def translate_vi2en(vi_text: str) -> str:
|
| 74 |
+
for i, j in dict_map.items():
|
| 75 |
+
vi_text = vi_text.replace(i, j)
|
| 76 |
+
input_ids = tokenizer_vi2en(vi_text, return_tensors="pt").input_ids
|
| 77 |
+
output_ids = model_vi2en.generate(
|
| 78 |
+
input_ids,
|
| 79 |
+
decoder_start_token_id=tokenizer_vi2en.lang_code_to_id["en_XX"],
|
| 80 |
+
num_return_sequences=1,
|
| 81 |
+
# # With sampling
|
| 82 |
+
# do_sample=True,
|
| 83 |
+
# top_k=100,
|
| 84 |
+
# top_p=0.8,
|
| 85 |
+
# With beam search
|
| 86 |
+
num_beams=5,
|
| 87 |
+
early_stopping=True
|
| 88 |
+
)
|
| 89 |
+
en_text = tokenizer_vi2en.batch_decode(output_ids, skip_special_tokens=True)
|
| 90 |
+
en_text = " ".join(en_text)
|
| 91 |
+
return en_text
|
| 92 |
|
| 93 |
DESCRIPTION="""CODE"""
|
| 94 |
|
|
|
|
| 115 |
print(total_count)
|
| 116 |
os.system("nvidia-smi")
|
| 117 |
conversation = []
|
| 118 |
+
message = translate_vi2en(message)
|
| 119 |
if system_prompt:
|
| 120 |
conversation.append({"role": "system", "content": system_prompt})
|
| 121 |
for user, assistant in chat_history:
|