Spaces:
Sleeping
Sleeping
Upload app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,125 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
VoiceAura Translation API
|
| 3 |
+
Models:
|
| 4 |
+
1. SLPG/English_to_Urdu_Unsupervised_MT (en β ur)
|
| 5 |
+
2. SLPG/Punjabi_Shahmukhi_to_Gurmukhi (pa-s β pa-g)
|
| 6 |
+
3. SLPG/Punjabi_Gurmukhi_to_Shahmukhi (pa-g β pa-s)
|
| 7 |
+
"""
|
| 8 |
+
|
| 9 |
+
from fastapi import FastAPI
|
| 10 |
+
from fastapi.middleware.cors import CORSMiddleware
|
| 11 |
+
from pydantic import BaseModel
|
| 12 |
+
import os, requests, argparse, torch
|
| 13 |
+
|
| 14 |
+
# β
PyTorch 2.6 fix
|
| 15 |
+
torch.serialization.add_safe_globals([argparse.Namespace])
|
| 16 |
+
|
| 17 |
+
app = FastAPI()
|
| 18 |
+
app.add_middleware(
|
| 19 |
+
CORSMiddleware,
|
| 20 |
+
allow_origins=["*"],
|
| 21 |
+
allow_methods=["*"],
|
| 22 |
+
allow_headers=["*"],
|
| 23 |
+
)
|
| 24 |
+
|
| 25 |
+
# ββ Model URLs βββββββββββββββββββββββββββββββββββββββββββ
|
| 26 |
+
MODELS_CONFIG = {
|
| 27 |
+
"en-ur": {
|
| 28 |
+
"files": {
|
| 29 |
+
"checkpoint_8_96000.pt": "https://huggingface.co/SLPG/English_to_Urdu_Unsupervised_MT/resolve/main/checkpoint_8_96000.pt",
|
| 30 |
+
"dict.en.txt": "https://huggingface.co/SLPG/English_to_Urdu_Unsupervised_MT/resolve/main/dict.en.txt",
|
| 31 |
+
"dict.ur.txt": "https://huggingface.co/SLPG/English_to_Urdu_Unsupervised_MT/resolve/main/dict.ur.txt",
|
| 32 |
+
},
|
| 33 |
+
"dir": "models/en_ur",
|
| 34 |
+
"checkpoint": "checkpoint_8_96000.pt",
|
| 35 |
+
"instance": None,
|
| 36 |
+
},
|
| 37 |
+
"pa-s-pa-g": {
|
| 38 |
+
"files": {
|
| 39 |
+
"checkpoint_5_78000.pt": "https://huggingface.co/SLPG/Punjabi_Shahmukhi_to_Gurmukhi_Transliteration/resolve/main/checkpoint_5_78000.pt",
|
| 40 |
+
"dict.pa.txt": "https://huggingface.co/SLPG/Punjabi_Shahmukhi_to_Gurmukhi_Transliteration/resolve/main/dict.pa.txt",
|
| 41 |
+
"dict.pk.txt": "https://huggingface.co/SLPG/Punjabi_Shahmukhi_to_Gurmukhi_Transliteration/resolve/main/dict.pk.txt",
|
| 42 |
+
},
|
| 43 |
+
"dir": "models/pa_s_pa_g",
|
| 44 |
+
"checkpoint": "checkpoint_5_78000.pt",
|
| 45 |
+
"instance": None,
|
| 46 |
+
},
|
| 47 |
+
}
|
| 48 |
+
|
| 49 |
+
# ββ Helpers ββββββββββββββββββββββββββββββββββββββββββββββ
|
| 50 |
+
def download_file(url: str, path: str):
|
| 51 |
+
if os.path.exists(path):
|
| 52 |
+
print(f"[β] Exists: {path}")
|
| 53 |
+
return
|
| 54 |
+
print(f"[β] Downloading: {path} ...")
|
| 55 |
+
os.makedirs(os.path.dirname(path), exist_ok=True)
|
| 56 |
+
with requests.get(url, stream=True) as r:
|
| 57 |
+
r.raise_for_status()
|
| 58 |
+
with open(path, "wb") as f:
|
| 59 |
+
for chunk in r.iter_content(chunk_size=8192):
|
| 60 |
+
f.write(chunk)
|
| 61 |
+
print(f"[β] Done: {path}")
|
| 62 |
+
|
| 63 |
+
def patched_torch_load(*args, **kwargs):
|
| 64 |
+
kwargs["weights_only"] = False
|
| 65 |
+
return _original_torch_load(*args, **kwargs)
|
| 66 |
+
|
| 67 |
+
_original_torch_load = torch.load
|
| 68 |
+
|
| 69 |
+
def load_model(pair: str):
|
| 70 |
+
cfg = MODELS_CONFIG[pair]
|
| 71 |
+
if cfg["instance"] is not None:
|
| 72 |
+
return cfg["instance"]
|
| 73 |
+
|
| 74 |
+
# Download files
|
| 75 |
+
for fname, url in cfg["files"].items():
|
| 76 |
+
download_file(url, os.path.join(cfg["dir"], fname))
|
| 77 |
+
|
| 78 |
+
# Patch torch.load for fairseq
|
| 79 |
+
torch.load = patched_torch_load
|
| 80 |
+
from fairseq.models.transformer import TransformerModel
|
| 81 |
+
model = TransformerModel.from_pretrained(
|
| 82 |
+
cfg["dir"],
|
| 83 |
+
checkpoint_file=cfg["checkpoint"],
|
| 84 |
+
data_name_or_path=cfg["dir"],
|
| 85 |
+
)
|
| 86 |
+
torch.load = _original_torch_load
|
| 87 |
+
model.eval()
|
| 88 |
+
cfg["instance"] = model
|
| 89 |
+
print(f"[β] Model ready: {pair}")
|
| 90 |
+
return model
|
| 91 |
+
|
| 92 |
+
# ββ Startup β load all models ββββββββββββββββββββββββββββ
|
| 93 |
+
@app.on_event("startup")
|
| 94 |
+
async def startup():
|
| 95 |
+
for pair in MODELS_CONFIG:
|
| 96 |
+
load_model(pair)
|
| 97 |
+
|
| 98 |
+
# ββ Endpoints ββββββββββββββββββββββββββββββββββββββββββββ
|
| 99 |
+
class Req(BaseModel):
|
| 100 |
+
text: str
|
| 101 |
+
from_lang: str = "en"
|
| 102 |
+
to_lang: str = "ur"
|
| 103 |
+
|
| 104 |
+
@app.get("/")
|
| 105 |
+
def root():
|
| 106 |
+
loaded = {k: MODELS_CONFIG[k]["instance"] is not None for k in MODELS_CONFIG}
|
| 107 |
+
return {"status": "VoiceAura API β", "models": loaded}
|
| 108 |
+
|
| 109 |
+
@app.post("/translate")
|
| 110 |
+
def translate(req: Req):
|
| 111 |
+
if not req.text.strip():
|
| 112 |
+
return {"success": False, "translation": ""}
|
| 113 |
+
|
| 114 |
+
pair = f"{req.from_lang}-{req.to_lang}"
|
| 115 |
+
|
| 116 |
+
if pair not in MODELS_CONFIG:
|
| 117 |
+
return {"success": False, "translation": f"β οΈ Pair '{pair}' not supported."}
|
| 118 |
+
|
| 119 |
+
try:
|
| 120 |
+
model = load_model(pair)
|
| 121 |
+
result = model.translate(req.text.strip())
|
| 122 |
+
return {"success": True, "translation": result, "pair": pair}
|
| 123 |
+
except Exception as e:
|
| 124 |
+
print(f"Error [{pair}]: {e}")
|
| 125 |
+
return {"success": False, "translation": str(e)}
|