Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -11,10 +11,54 @@ from typing import List, Dict, Any
|
|
| 11 |
from io import BytesIO
|
| 12 |
import logging
|
| 13 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
# Configure logging
|
| 15 |
logging.basicConfig(level=logging.INFO)
|
| 16 |
logger = logging.getLogger(__name__)
|
| 17 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
class OCRDataset(Dataset):
|
| 19 |
def __init__(self, training_data: List[Dict], transform=None):
|
| 20 |
self.data = training_data
|
|
@@ -71,6 +115,7 @@ class SimpleCRNN(nn.Module):
|
|
| 71 |
output = self.classifier(rnn_out)
|
| 72 |
return output
|
| 73 |
|
|
|
|
| 74 |
def train_model(training_request: Dict[str, Any]) -> Dict[str, Any]:
|
| 75 |
try:
|
| 76 |
training_data = training_request['trainingData']
|
|
@@ -121,6 +166,43 @@ def train_model(training_request: Dict[str, Any]) -> Dict[str, Any]:
|
|
| 121 |
'error': str(e)
|
| 122 |
}
|
| 123 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 124 |
def train_api(training_request_json: str) -> str:
|
| 125 |
try:
|
| 126 |
training_request = json.loads(training_request_json)
|
|
@@ -132,7 +214,8 @@ def train_api(training_request_json: str) -> str:
|
|
| 132 |
def health_check() -> str:
|
| 133 |
return json.dumps({'status': 'healthy', 'gpu_available': torch.cuda.is_available()})
|
| 134 |
|
| 135 |
-
|
|
|
|
| 136 |
gr.Markdown("# Rysa OCR Training Service")
|
| 137 |
gr.Markdown("Multi-script OCR model training for Sanskrit and Indian scripts.")
|
| 138 |
|
|
@@ -153,5 +236,8 @@ with gr.Blocks(title="Rysa OCR Training Service") as app:
|
|
| 153 |
|
| 154 |
health_button.click(health_check, outputs=health_output)
|
| 155 |
|
|
|
|
|
|
|
|
|
|
| 156 |
if __name__ == "__main__":
|
| 157 |
-
|
|
|
|
| 11 |
from io import BytesIO
|
| 12 |
import logging
|
| 13 |
|
| 14 |
+
# Add FastAPI imports
|
| 15 |
+
from fastapi import FastAPI
|
| 16 |
+
from pydantic import BaseModel
|
| 17 |
+
import uvicorn
|
| 18 |
+
|
| 19 |
# Configure logging
|
| 20 |
logging.basicConfig(level=logging.INFO)
|
| 21 |
logger = logging.getLogger(__name__)
|
| 22 |
|
| 23 |
+
# Create FastAPI app
|
| 24 |
+
fastapi_app = FastAPI()
|
| 25 |
+
|
| 26 |
+
# Pydantic models for API
|
| 27 |
+
class TrainingData(BaseModel):
|
| 28 |
+
imageUrl: str
|
| 29 |
+
groundTruthText: str
|
| 30 |
+
sourceScript: str
|
| 31 |
+
imageType: str
|
| 32 |
+
difficultyLevel: str
|
| 33 |
+
|
| 34 |
+
class TrainingConfig(BaseModel):
|
| 35 |
+
epochs: int
|
| 36 |
+
batchSize: int
|
| 37 |
+
learningRate: float
|
| 38 |
+
validationSplit: float
|
| 39 |
+
saveInterval: int
|
| 40 |
+
modelName: str
|
| 41 |
+
sampleLimit: int
|
| 42 |
+
|
| 43 |
+
class TrainingRequest(BaseModel):
|
| 44 |
+
trainingData: List[TrainingData]
|
| 45 |
+
config: TrainingConfig
|
| 46 |
+
sessionId: str
|
| 47 |
+
|
| 48 |
+
class TrainingMetrics(BaseModel):
|
| 49 |
+
finalLoss: float
|
| 50 |
+
finalAccuracy: float
|
| 51 |
+
epochs: int
|
| 52 |
+
sampleCount: int
|
| 53 |
+
|
| 54 |
+
class TrainingResponse(BaseModel):
|
| 55 |
+
success: bool
|
| 56 |
+
jobId: str = None
|
| 57 |
+
modelUrl: str = None
|
| 58 |
+
metrics: TrainingMetrics = None
|
| 59 |
+
error: str = None
|
| 60 |
+
|
| 61 |
+
# Your existing classes (keep these)
|
| 62 |
class OCRDataset(Dataset):
|
| 63 |
def __init__(self, training_data: List[Dict], transform=None):
|
| 64 |
self.data = training_data
|
|
|
|
| 115 |
output = self.classifier(rnn_out)
|
| 116 |
return output
|
| 117 |
|
| 118 |
+
# Your existing training function (keep this)
|
| 119 |
def train_model(training_request: Dict[str, Any]) -> Dict[str, Any]:
|
| 120 |
try:
|
| 121 |
training_data = training_request['trainingData']
|
|
|
|
| 166 |
'error': str(e)
|
| 167 |
}
|
| 168 |
|
| 169 |
+
# ADD THESE REST API ENDPOINTS (This is what was missing!)
|
| 170 |
+
@fastapi_app.get("/api/health")
|
| 171 |
+
async def health_endpoint():
|
| 172 |
+
return {"status": "healthy", "gpu_available": torch.cuda.is_available()}
|
| 173 |
+
|
| 174 |
+
@fastapi_app.post("/api/train")
|
| 175 |
+
async def train_endpoint(request: TrainingRequest):
|
| 176 |
+
try:
|
| 177 |
+
# Convert Pydantic model to dict for your existing function
|
| 178 |
+
training_request_dict = {
|
| 179 |
+
'trainingData': [item.dict() for item in request.trainingData],
|
| 180 |
+
'config': request.config.dict(),
|
| 181 |
+
'sessionId': request.sessionId
|
| 182 |
+
}
|
| 183 |
+
|
| 184 |
+
result = train_model(training_request_dict)
|
| 185 |
+
|
| 186 |
+
if result['success']:
|
| 187 |
+
return TrainingResponse(
|
| 188 |
+
success=True,
|
| 189 |
+
jobId=request.sessionId,
|
| 190 |
+
modelUrl=result['modelUrl'],
|
| 191 |
+
metrics=TrainingMetrics(**result['metrics'])
|
| 192 |
+
)
|
| 193 |
+
else:
|
| 194 |
+
return TrainingResponse(
|
| 195 |
+
success=False,
|
| 196 |
+
error=result['error']
|
| 197 |
+
)
|
| 198 |
+
except Exception as e:
|
| 199 |
+
logger.error(f"API training failed: {e}")
|
| 200 |
+
return TrainingResponse(
|
| 201 |
+
success=False,
|
| 202 |
+
error=str(e)
|
| 203 |
+
)
|
| 204 |
+
|
| 205 |
+
# Your existing Gradio functions (keep these)
|
| 206 |
def train_api(training_request_json: str) -> str:
|
| 207 |
try:
|
| 208 |
training_request = json.loads(training_request_json)
|
|
|
|
| 214 |
def health_check() -> str:
|
| 215 |
return json.dumps({'status': 'healthy', 'gpu_available': torch.cuda.is_available()})
|
| 216 |
|
| 217 |
+
# Your existing Gradio interface (keep this)
|
| 218 |
+
with gr.Blocks(title="Rysa OCR Training Service") as gradio_app:
|
| 219 |
gr.Markdown("# Rysa OCR Training Service")
|
| 220 |
gr.Markdown("Multi-script OCR model training for Sanskrit and Indian scripts.")
|
| 221 |
|
|
|
|
| 236 |
|
| 237 |
health_button.click(health_check, outputs=health_output)
|
| 238 |
|
| 239 |
+
# Mount Gradio app to FastAPI (This combines both!)
|
| 240 |
+
app = gr.mount_gradio_app(fastapi_app, gradio_app, path="/")
|
| 241 |
+
|
| 242 |
if __name__ == "__main__":
|
| 243 |
+
uvicorn.run(app, host="0.0.0.0", port=7860)
|