Spaces:
Sleeping
Sleeping
Upload 2 files
Browse files- app.py +122 -0
- requirements.txt +4 -0
app.py
ADDED
@@ -0,0 +1,122 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from fastapi import FastAPI, Depends, HTTPException, Request
|
2 |
+
from fastapi.security import APIKeyQuery
|
3 |
+
from pydantic import BaseModel
|
4 |
+
from typing import List, Union, Dict
|
5 |
+
from functools import lru_cache
|
6 |
+
import jwt
|
7 |
+
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
|
8 |
+
import torch
|
9 |
+
from flores200_codes import flores_codes
|
10 |
+
import gradio as gr
|
11 |
+
|
12 |
+
CUSTOM_PATH = "/gradio"
|
13 |
+
|
14 |
+
app = FastAPI()
|
15 |
+
|
16 |
+
|
17 |
+
# This should be a secure secret key in a real application
|
18 |
+
SECRET_KEY = "your_secret_key_here"
|
19 |
+
|
20 |
+
# Define the security scheme
|
21 |
+
api_key_query = APIKeyQuery(name="jwtToken", auto_error=False)
|
22 |
+
|
23 |
+
|
24 |
+
class TranslationRequest(BaseModel):
|
25 |
+
strings: List[Union[str, Dict[str, str]]]
|
26 |
+
|
27 |
+
|
28 |
+
class TranslationResponse(BaseModel):
|
29 |
+
data: Dict[str, List[str]]
|
30 |
+
|
31 |
+
|
32 |
+
@lru_cache()
|
33 |
+
def load_model():
|
34 |
+
model_name_dict = {
|
35 |
+
"nllb-distilled-600M": "facebook/nllb-200-distilled-600M",
|
36 |
+
}
|
37 |
+
|
38 |
+
call_name = "nllb-distilled-600M"
|
39 |
+
real_name = model_name_dict[call_name]
|
40 |
+
print(f"\tLoading model: {call_name}")
|
41 |
+
|
42 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
43 |
+
model = AutoModelForSeq2SeqLM.from_pretrained(real_name).to(device)
|
44 |
+
tokenizer = AutoTokenizer.from_pretrained(real_name)
|
45 |
+
|
46 |
+
return model, tokenizer
|
47 |
+
|
48 |
+
|
49 |
+
model, tokenizer = load_model()
|
50 |
+
|
51 |
+
|
52 |
+
def translate_text(text: List[str], source_lang: str, target_lang: str) -> List[str]:
|
53 |
+
source = flores_codes[source_lang]
|
54 |
+
target = flores_codes[target_lang]
|
55 |
+
|
56 |
+
translator = pipeline(
|
57 |
+
"translation",
|
58 |
+
model=model,
|
59 |
+
tokenizer=tokenizer,
|
60 |
+
src_lang=source,
|
61 |
+
tgt_lang=target,
|
62 |
+
)
|
63 |
+
output = translator(text, max_length=400)
|
64 |
+
|
65 |
+
return [item["translation_text"] for item in output]
|
66 |
+
|
67 |
+
|
68 |
+
async def verify_token(token: str = Depends(api_key_query)):
|
69 |
+
if not token:
|
70 |
+
raise HTTPException(status_code=401, detail={"message": "Token is missing"})
|
71 |
+
try:
|
72 |
+
jwt.decode(token, SECRET_KEY, algorithms=["HS256"])
|
73 |
+
except:
|
74 |
+
raise HTTPException(status_code=401, detail={"message": "Token is invalid"})
|
75 |
+
return token
|
76 |
+
|
77 |
+
|
78 |
+
@app.post("/translate/", response_model=TranslationResponse)
|
79 |
+
async def translate(
|
80 |
+
request: Request,
|
81 |
+
source: str,
|
82 |
+
target: str,
|
83 |
+
project_id: str,
|
84 |
+
token: str = Depends(verify_token),
|
85 |
+
):
|
86 |
+
if not all([source, target, project_id]):
|
87 |
+
raise HTTPException(
|
88 |
+
status_code=400, detail={"message": "Missing required parameters"}
|
89 |
+
)
|
90 |
+
|
91 |
+
data = await request.json()
|
92 |
+
strings = data.get("strings", [])
|
93 |
+
|
94 |
+
if not strings:
|
95 |
+
raise HTTPException(
|
96 |
+
status_code=400, detail={"message": "No strings provided for translation"}
|
97 |
+
)
|
98 |
+
|
99 |
+
try:
|
100 |
+
if isinstance(strings[0], dict): # Extended request
|
101 |
+
translations = translate_text([s["text"] for s in strings], source, target)
|
102 |
+
else: # Simple request
|
103 |
+
translations = translate_text(strings, source, target)
|
104 |
+
|
105 |
+
return TranslationResponse(data={"translations": translations})
|
106 |
+
except Exception as e:
|
107 |
+
raise HTTPException(status_code=500, detail={"message": str(e)})
|
108 |
+
|
109 |
+
|
110 |
+
@app.get("/logo.png")
|
111 |
+
async def logo():
|
112 |
+
# TODO: Implement logic to serve the logo
|
113 |
+
return "Logo placeholder"
|
114 |
+
|
115 |
+
|
116 |
+
io = gr.Interface(lambda x: "Hello, " + x + "!", "textbox", "textbox")
|
117 |
+
app = gr.mount_gradio_app(app, io, path=CUSTOM_PATH)
|
118 |
+
|
119 |
+
if __name__ == "__main__":
|
120 |
+
import uvicorn
|
121 |
+
|
122 |
+
uvicorn.run(app, host="0.0.0.0", port=8000)
|
requirements.txt
ADDED
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
1 |
+
flask
|
2 |
+
jwt
|
3 |
+
transformers
|
4 |
+
torch
|