add question generation

server/.env.example

@@ -1 +1,3 @@
 DATABASE_URI="sqlite:///database.db"
+LANGUAGE_MODEL_API="http://localhost:8080/v1"
+MODEL_NAME="SmolLM3-Q4_K_M.gguf"

server/README.md

@@ -3,6 +3,7 @@ FastAPI server to run Manolia
 
 # Run
 ```
+cd src/app
 fastapi dev main.py
 ```
 

server/requirements.txt

@@ -1,3 +1,7 @@
 fastapi==0.128.6
 sqlmodel==0.0.32
 python-dotenv==1.2.1
+openai==2.21.0
+spacy==3.8.11
+# python -m spacy download en_core_web_sm
+# python -m spacy download fr_core_news_sm

server/src/app/api/v1/knowledges.py

@@ -4,6 +4,8 @@ from src.app.models.knowledge import Knowledge
 from src.app.crud.crud_knowledges import create_knowledge, read_knowledges, read_knowledge, update_knowledge, delete_knowledge
 from src.app.crud.crud_questions import read_questions as read_questions_crud
 
+from src.app.services.language_generation import completion
+
 #Added in __ini__
 router = APIRouter(tags=["knowledges"])
 
@@ -38,9 +40,9 @@ def delete(id: int):
 #TODO: find pattern 
 @router.post("/knowledges/{id}/questions")
 def create_questions(id: int):
-    #SLM Generation
+    question = completion(read_knowledge(id))
     #create_question()
-    return True
+    return question
 
 @router.get("/knowledges/{id}/questions")
 def read_questions(id: int):

server/src/app/database.py

@@ -1,9 +1,9 @@
-from dotenv import load_dotenv
 import os
-#import secrets
+from dotenv import load_dotenv
 from sqlmodel import Session, SQLModel, create_engine
 
 load_dotenv()
+
 database_uri=os.environ.get("DATABASE_URI")
 
 connect_args = {"check_same_thread": False}

server/src/app/services/language_generation.py

@@ -0,0 +1,47 @@
+import os
+import spacy
+
+from openai import OpenAI
+from pydantic import BaseModel
+from src.app.models.knowledge import Knowledge
+
+language_model_api=os.environ.get("LANGUAGE_MODEL_API")
+model_name=os.environ.get("LANGUAGE_MODEL_NAME")
+
+client = OpenAI(
+    base_url=language_model_api,
+    api_key = "sk-no-key-required"
+)
+
+nlp = spacy.load("fr_core_news_sm")
+
+def completion(knowledge: Knowledge):
+
+    context = "Texte : ```" + knowledge.content + "```"
+    instruction = "A partir du texte génère 3 questions :"
+    prompt = context + "\n" + instruction
+
+    #SLM processing
+    response = client.responses.create(
+        model=model_name,
+        input=[
+            {"role": "system", "content": "Question Generation"},
+            {"role": "user", "content": prompt}],
+        )
+    text_response = response.output[0].content[0].text
+
+    #Sentence segmentation
+    doc = nlp(text_response)
+    sents = list()
+    for sentence in doc.sents:
+        sents.append(sentence.text)
+
+    #Interrogation sentence detection
+    questions = list()
+    for sent in sents:
+        index_mark = sent.rfind("?")
+        if(index_mark > 0):
+            questions.append(sent[0:index_mark+1])
+
+    return {"questions": questions, "sentence":sents}
+