Merge pull request #5 from krflorian/feature/intent_classification

Feature/intent classification

Dockerfile

@@ -1,11 +1,10 @@
 # 
 FROM python:3.11
-#RUN pip install poetry==1.4.2
 
 # 
 WORKDIR /app
 
-#COPY pyproject.toml poetry.lock app.py ./
+# COPY pyproject.toml poetry.lock app.py ./
 COPY app.py requirements.txt ./
 RUN pip install --no-cache-dir --upgrade -r requirements.txt
 
@@ -14,10 +13,5 @@ RUN touch README.md
 ARG HF_HOME="app/data/.cache"
 ENV HF_HOME="app/data/.cache"
 
-# 
-#RUN poetry config virtualenvs.create false
-#RUN poetry install
-#RUN pip install . 
-
 # 
 CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "8000"]

app.py

@@ -4,13 +4,11 @@
 from fastapi import FastAPI
 from pydantic import BaseModel, Field
 from sentence_transformers import SentenceTransformer, CrossEncoder
-from transformers import pipeline
 
 from src.objects import Card, Document
 from src.vector_db import VectorDB
 from src.util import load_config
 from src.hallucination import validate_answer
-from src.nli import classify_intent
 
 config: dict = load_config(Path("configs/config.yaml"))
 vector_model: SentenceTransformer = SentenceTransformer(
@@ -19,13 +17,6 @@
 hallucination_model: CrossEncoder = CrossEncoder(
     config.get("halucination_model_name", "vectara/hallucination_evaluation_model")
 )
-nli_classifier_model = pipeline(
-    "zero-shot-classification",
-    model=config.get(
-        "nli_classifier", "MoritzLaurer/deberta-v3-large-zeroshot-v1.1-all-33"
-    ),
-)
-
 
 db: dict[str, VectorDB] = {
     "card": VectorDB.load(config.get("cards_db_file", None)),
@@ -150,13 +141,5 @@ async def validate_rag_chunks(
     ]
 
 
-@app.post("/nli/")
-async def classify_user_intent(
-    request: NLIClassificationRequest,
-) -> NLIClassificationResponse:
-    intent, score = classify_intent(request.text, classifier=nli_classifier_model)
-    return NLIClassificationResponse(intent=intent, score=score)
-
-
 if __name__ == "__main__":
     uvicorn.run(app, host="0.0.0.0", port=8000, log_level="debug", proxy_headers=True)