choisir par scene

rag.py

@@ -9,6 +9,7 @@ from uuid import uuid4
 from prompt import *
 import random
 from itext2kg.models import KnowledgeGraph
+from langchain.text_splitter import RecursiveCharacterTextSplitter
 
 
 import faiss
@@ -57,6 +58,10 @@ with open("chunks_ia_signature.pkl", "rb") as file:
     chunks = pickle.load(file)
     print("Chunks loaded")
 
+with open("scenes.pkl", "rb") as file:
+    scenes = pickle.load(file)
+    print("Scenes loaded")
+
 class sphinx_output(BaseModel):
     question: str = Field(description="The question to ask the user to test if they read the entire book")
     answers: list[str] = Field(description="The possible answers to the question to test if the user read the entire book")
@@ -70,12 +75,30 @@ class verification_score(BaseModel):
     score: float = Field(description="The score of the user's response from 0 to 10 to the question")
 
 
-
 llm = ChatOpenAI(model="gpt-4o", max_tokens=300, temperature=0.5)
 
-
-def get_random_chunk() -> str:
-    return chunks[random.randint(0, len(chunks) - 1)]
+def split_texts(text : str) -> list[str]:
+    splitter = RecursiveCharacterTextSplitter(
+        chunk_size=1000,
+        chunk_overlap=200,
+        length_function=len,
+        is_separator_regex=False,
+    )
+    return splitter.split_text(text)
+
+#########################################################################
+### PAR ICI , CHOISIR UNE SCENE SPECIFIQUE DANS L'ARGUMENT DE LA FONCTION
+def get_random_chunk(scene_specific = 5) :
+    if scene_specific:
+        chunks_scene = split_texts(scenes[scene_specific-1])
+        print(f"Scene {scene_specific} has {len(chunks_scene)} chunks")
+        print([chunk[0:50] for chunk in chunks_scene])
+        print('---')
+        chunk_chosen = chunks_scene[random.randint(0, len(chunks_scene) - 1)]
+        print(f"Chosen chunk: {chunk_chosen}")
+        return chunk_chosen, scene_specific
+    
+    return chunks[random.randint(0, len(chunks) - 1)],scene_specific
 
 
 def get_vectorstore() -> FAISS:
@@ -98,7 +121,9 @@ def generate_sphinx_response() -> sphinx_output:
     writer = "Laurent Tripied"
     book_name =  "Limites de l'imaginaire ou limites planétaires"
     summary = summary_text
-    excerpt = get_random_chunk()
+    excerpt , scene_number = get_random_chunk()
+    if scene_number:
+        summary = "scene " + str(scene_number)
     prompt = PromptTemplate.from_template(template_sphinx)
     structured_llm = llm.with_structured_output(sphinx_output)
     # Create an LLM chain with the prompt and the LLM
@@ -106,6 +131,8 @@ def generate_sphinx_response() -> sphinx_output:
 
     return llm_chain.invoke({"writer":writer,"book_name":book_name,"summary":summary,"excerpt":excerpt})
 
+#############################################################
+### PAR ICI , CHOISIR LE DEGRE DE SEVERITE DE LA VERIFICATION
 def verify_response(response:str,answers:list[str],question:str) -> bool:
     prompt = PromptTemplate.from_template(template_verify)
     structured_llm = llm.with_structured_output(verification_score)
@@ -120,7 +147,7 @@ def retrieve_context_from_vectorestore(query:str) -> str:
     return retriever.invoke(query)
 
         
-def generate_stream(query:str,messages = [], model = "gpt-4o-mini", max_tokens = 300, temperature = 0.5,index_name="",stream=True,vector_store=None):
+def generate_stream(query:str,messages = [], model = "gpt-4o-mini", max_tokens = 300, temperature = 1,index_name="",stream=True,vector_store=None):
     try:
         print("init chat")
         print("init template")

scenes.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:17fc4636b752c5b8f1434d0c97c95ea3b12605b083689e6d79daacd060f6c110
+size 142917