import os import time import numpy as np from contextlib import asynccontextmanager from fastapi import FastAPI, HTTPException, Header from fastapi.concurrency import run_in_threadpool from app.models import models from app.schemas import ( AnalysisRequest, EmbeddingRequest, SarcasmRequest, AbsaRequest ) API_KEY = os.getenv("API_KEY", "") def verify_api_key(key: str | None): if API_KEY and key != API_KEY: raise HTTPException( status_code=401, detail="Invalid API key" ) def classify(pipe, text): result = pipe(text)[0] return { "label": result["label"], "confidence": round(float(result["score"]), 6) } def detect_absa(request: AbsaRequest): absa_model = models.absa() start = time.perf_counter() # Format input untuk ABSA: [sentence] [SEP] [aspect] text_input = f"{request.text} [SEP] {request.aspect}" # Prediksi sentimen pred = absa_model(text_input) return { "absa":pred, "metadata": { "pipeline": ["absa"], "inference_time_ms": round( (time.perf_counter() - start) * 1000, 2 ) }, } def detect_sarcasm_text(request: SarcasmRequest): # # PHASE 3: Sarcasm Detection sarcasm_model = models.sarcasm() text = request.text start = time.perf_counter() sarcasm = classify( sarcasm_model, text ) return { # "entities": entities, # "sentiment": sentiment, "sarcasm": { **sarcasm, # "is_sarcastic": ( # "sarcas" in sarcasm["label"].lower() # or sarcasm["label"] == "LABEL_1" or sarcasm["label"] == "1" # ) }, "metadata": { "pipeline": [ # "dynamic_ner", # "sentiment", "sarcasm" ], "inference_time_ms": round( (time.perf_counter() - start) * 1000, 2 ) } } def analyze_text(request: AnalysisRequest): text = request.text start = time.perf_counter() # PHASE 1: Dynamic NER ner_model = models.ner() entities = ner_model.predict_entities( text, request.entity_labels, threshold=request.entity_threshold ) # PHASE 2: Sentiment Analysis sentiment_model = models.sentiment() sentiment = classify( sentiment_model, text ) # # PHASE 3: Sarcasm Detection # sarcasm_model = models.sarcasm() # sarcasm = classify( # sarcasm_model, # text # ) return { "entities": entities, "sentiment": sentiment, # "sarcasm": { # **sarcasm, # "is_sarcastic": ( # "sarcas" in sarcasm["label"].lower() # or sarcasm["label"] == "LABEL_1" # ) # }, "metadata": { "pipeline": [ "dynamic_ner", "sentiment", # "sarcasm" ], "inference_time_ms": round( (time.perf_counter() - start) * 1000, 2 ) } } @asynccontextmanager async def lifespan(app: FastAPI): yield app = FastAPI( title="NLP Inference Service", version="1.0.0", lifespan=lifespan ) @app.get("/health") async def health(): return { "status": "healthy", "loaded_models": list(models._models.keys()) } @app.post("/v1/analyze") async def analyze( request: AnalysisRequest, x_api_key: str | None = Header(default=None) ): verify_api_key(x_api_key) try: return await run_in_threadpool( analyze_text, request ) except HTTPException: raise except Exception as exc: raise HTTPException( status_code=500, detail=str(exc) ) @app.post("/v1/detect_sarcasm") async def detect_sarcasm( request: SarcasmRequest, x_api_key: str | None = Header(default=None) ): verify_api_key(x_api_key) try: return await run_in_threadpool( detect_sarcasm_text, request ) except HTTPException: raise except Exception as exc: raise HTTPException( status_code=500, detail=str(exc) ) @app.post("/v1/absa") async def absa( request: AbsaRequest, x_api_key: str | None = Header(default=None) ): verify_api_key(x_api_key) try: return await run_in_threadpool( detect_absa, request ) except HTTPException: raise except Exception as exc: raise HTTPException( status_code=500, detail=str(exc) ) from app.embedding import generate_embeddings @app.post("/v1/embeddings") async def embeddings( request: EmbeddingRequest, x_api_key: str | None = Header(default=None) ): verify_api_key(x_api_key) try: return await run_in_threadpool( generate_embeddings, request ) except HTTPException: raise except Exception as exc: raise HTTPException( status_code=500, detail=str(exc) )