Update app.py
Browse files
app.py
CHANGED
|
@@ -1,5 +1,5 @@
|
|
|
|
|
| 1 |
from fastapi import FastAPI, UploadFile, File, HTTPException
|
| 2 |
-
import uvicorn
|
| 3 |
import torchaudio
|
| 4 |
import torch.nn.functional as F
|
| 5 |
import torch
|
|
@@ -8,18 +8,39 @@ import onnxruntime as ort
|
|
| 8 |
from huggingface_hub import hf_hub_download
|
| 9 |
import os
|
| 10 |
|
| 11 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
|
| 13 |
-
|
| 14 |
-
# Yeh line model ko cache kar legi, baar baar download nahi karegi
|
| 15 |
-
model_path = hf_hub_download(repo_id="Hammad712/pakistani-lid-v3-sota", filename="pakistani_lid_v3.onnx")
|
| 16 |
|
| 17 |
-
|
| 18 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
|
| 20 |
labels = ("balochi", "english", "pashto", "sindhi", "urdu")
|
| 21 |
id2label = {i: label for i, label in enumerate(labels)}
|
| 22 |
|
|
|
|
|
|
|
|
|
|
| 23 |
def predict_audio(audio_path):
|
| 24 |
waveform, sr = torchaudio.load(audio_path)
|
| 25 |
if waveform.shape[0] > 1: waveform = waveform.mean(dim=0, keepdim=True)
|
|
@@ -53,21 +74,29 @@ def predict_audio(audio_path):
|
|
| 53 |
pred_id = np.argmax(probs, axis=1)[0]
|
| 54 |
return id2label[pred_id], float(probs[0][pred_id])
|
| 55 |
|
|
|
|
|
|
|
|
|
|
| 56 |
@app.post("/predict")
|
| 57 |
async def predict_language(file: UploadFile = File(...)):
|
|
|
|
|
|
|
| 58 |
if not file.filename.endswith(('.wav', '.mp3', '.m4a', '.ogg')):
|
|
|
|
| 59 |
raise HTTPException(status_code=400, detail="Invalid audio format. Please upload wav, mp3, m4a, or ogg.")
|
| 60 |
|
| 61 |
temp_audio_path = f"temp_{file.filename}"
|
| 62 |
try:
|
| 63 |
-
#
|
| 64 |
with open(temp_audio_path, "wb") as buffer:
|
| 65 |
buffer.write(await file.read())
|
| 66 |
|
| 67 |
-
#
|
|
|
|
| 68 |
lang, confidence = predict_audio(temp_audio_path)
|
|
|
|
| 69 |
|
| 70 |
-
#
|
| 71 |
os.remove(temp_audio_path)
|
| 72 |
|
| 73 |
return {
|
|
@@ -76,10 +105,7 @@ async def predict_language(file: UploadFile = File(...)):
|
|
| 76 |
"confidence": round(confidence * 100, 2)
|
| 77 |
}
|
| 78 |
except Exception as e:
|
|
|
|
| 79 |
if os.path.exists(temp_audio_path):
|
| 80 |
os.remove(temp_audio_path)
|
| 81 |
-
raise HTTPException(status_code=500, detail=
|
| 82 |
-
|
| 83 |
-
if __name__ == "__main__":
|
| 84 |
-
print("β¨ Server is LIVE at: http://localhost:8080")
|
| 85 |
-
uvicorn.run(app, host="0.0.0.0", port=8080)
|
|
|
|
| 1 |
+
import logging
|
| 2 |
from fastapi import FastAPI, UploadFile, File, HTTPException
|
|
|
|
| 3 |
import torchaudio
|
| 4 |
import torch.nn.functional as F
|
| 5 |
import torch
|
|
|
|
| 8 |
from huggingface_hub import hf_hub_download
|
| 9 |
import os
|
| 10 |
|
| 11 |
+
# ==========================================
|
| 12 |
+
# 1. Setup Production Logging
|
| 13 |
+
# ==========================================
|
| 14 |
+
logging.basicConfig(
|
| 15 |
+
level=logging.INFO,
|
| 16 |
+
format="%(asctime)s [%(levelname)s] %(message)s",
|
| 17 |
+
handlers=[logging.StreamHandler()]
|
| 18 |
+
)
|
| 19 |
+
logger = logging.getLogger("LID_Engine")
|
| 20 |
|
| 21 |
+
app = FastAPI(title="Pakistani LID AI Engine (Production)")
|
|
|
|
|
|
|
| 22 |
|
| 23 |
+
# ==========================================
|
| 24 |
+
# 2. Model Initialization (Runs once on startup)
|
| 25 |
+
# ==========================================
|
| 26 |
+
logger.info("Initializing Application...")
|
| 27 |
+
try:
|
| 28 |
+
logger.info("Checking/Downloading ONNX Model from Hugging Face...")
|
| 29 |
+
model_path = hf_hub_download(repo_id="Hammad712/pakistani-lid-v3-sota", filename="pakistani_lid_v3.onnx")
|
| 30 |
+
|
| 31 |
+
logger.info("Loading ONNX Session for CPU...")
|
| 32 |
+
session = ort.InferenceSession(model_path, providers=['CPUExecutionProvider'])
|
| 33 |
+
logger.info("β
ONNX Session successfully loaded and ready!")
|
| 34 |
+
except Exception as e:
|
| 35 |
+
logger.error(f"β Failed to load model during startup: {e}")
|
| 36 |
+
raise e
|
| 37 |
|
| 38 |
labels = ("balochi", "english", "pashto", "sindhi", "urdu")
|
| 39 |
id2label = {i: label for i, label in enumerate(labels)}
|
| 40 |
|
| 41 |
+
# ==========================================
|
| 42 |
+
# 3. Core Inference Logic
|
| 43 |
+
# ==========================================
|
| 44 |
def predict_audio(audio_path):
|
| 45 |
waveform, sr = torchaudio.load(audio_path)
|
| 46 |
if waveform.shape[0] > 1: waveform = waveform.mean(dim=0, keepdim=True)
|
|
|
|
| 74 |
pred_id = np.argmax(probs, axis=1)[0]
|
| 75 |
return id2label[pred_id], float(probs[0][pred_id])
|
| 76 |
|
| 77 |
+
# ==========================================
|
| 78 |
+
# 4. API Endpoints
|
| 79 |
+
# ==========================================
|
| 80 |
@app.post("/predict")
|
| 81 |
async def predict_language(file: UploadFile = File(...)):
|
| 82 |
+
logger.info(f"Received request for file: {file.filename}")
|
| 83 |
+
|
| 84 |
if not file.filename.endswith(('.wav', '.mp3', '.m4a', '.ogg')):
|
| 85 |
+
logger.warning(f"Rejected invalid file type: {file.filename}")
|
| 86 |
raise HTTPException(status_code=400, detail="Invalid audio format. Please upload wav, mp3, m4a, or ogg.")
|
| 87 |
|
| 88 |
temp_audio_path = f"temp_{file.filename}"
|
| 89 |
try:
|
| 90 |
+
# Save file
|
| 91 |
with open(temp_audio_path, "wb") as buffer:
|
| 92 |
buffer.write(await file.read())
|
| 93 |
|
| 94 |
+
# Predict
|
| 95 |
+
logger.info(f"Processing inference for {file.filename}...")
|
| 96 |
lang, confidence = predict_audio(temp_audio_path)
|
| 97 |
+
logger.info(f"β
Prediction successful: {lang.upper()} ({confidence:.2%})")
|
| 98 |
|
| 99 |
+
# Cleanup
|
| 100 |
os.remove(temp_audio_path)
|
| 101 |
|
| 102 |
return {
|
|
|
|
| 105 |
"confidence": round(confidence * 100, 2)
|
| 106 |
}
|
| 107 |
except Exception as e:
|
| 108 |
+
logger.error(f"β Error processing {file.filename}: {str(e)}", exc_info=True)
|
| 109 |
if os.path.exists(temp_audio_path):
|
| 110 |
os.remove(temp_audio_path)
|
| 111 |
+
raise HTTPException(status_code=500, detail="Internal Server Error")
|
|
|
|
|
|
|
|
|
|
|
|