modify from rag to only vs

.evn

@@ -0,0 +1 @@
+EMBEDDINGS="BAAI/bge-m3"

app.py

@@ -2,107 +2,64 @@ import os
 import gradio as gr
 from gradio.components import Textbox, Button, Slider, Checkbox
 from AinaTheme import theme
-from urllib.error import HTTPError
 
-from rag import RAG
-from utils import setup
+from vectorstore import VectorStore
 
 MAX_NEW_TOKENS = 700
-SHOW_MODEL_PARAMETERS_IN_UI = os.environ.get("SHOW_MODEL_PARAMETERS_IN_UI", default="True") == "True"
 
-setup()
-
-
-rag = RAG(embeddings_model=os.getenv("EMBEDDINGS"))
 
+vectorStore = VectorStore(embeddings_model=os.getenv("EMBEDDINGS", "BAAI/bge-m3"))
 
 def eadop_rag(prompt, num_chunks):
-    model_parameters = {"NUM_CHUNKS": num_chunks}
-    try:
-        return rag.get_context(prompt, model_parameters)
-    except HTTPError as err:
-        if err.code == 400:
-            gr.Warning(
-                "The inference endpoint is only available Monday through Friday, from 08:00 to 20:00 CET."
-            )
-        return None, None, None
-    except:
+    prompt = prompt.strip()
+    if prompt == "":
         gr.Warning(
-            "Inference endpoint is not available right now. Please try again later."
+            "Prompt can't be empty!"
         )
-        return None, None, None
-
-
+        raise ValueError("Prompt can't be empty!")
+    return vectorStore.get_context(prompt, num_chunks)
+    
 def clear():
     return (
-        None, 
-        None,
         None,
         None,
         gr.Slider(value=2.0),
     )
 
-
 def gradio_app():
     with gr.Blocks(theme=theme) as demo:
         with gr.Row(equal_height=True):
-            with gr.Column(variant="panel"):
-                input_ = Textbox(
-                    lines=11,
-                    label="Input",
-                    placeholder="Quina és la finalitat del Servei Meteorològic de Catalunya?",
-                )
-                with gr.Row(variant="panel"):
-                    clear_btn = Button(
-                        "Clear",
-                    )
-                    submit_btn = Button("Submit", variant="primary", interactive=False)
+            output = Textbox(
+                lines=10, 
+                label="Context", 
+                interactive=False, 
+                show_copy_button=True
+            )
 
-                with gr.Row(variant="panel"):
-                    with gr.Accordion("Model parameters", open=False, visible=SHOW_MODEL_PARAMETERS_IN_UI):
-                        num_chunks = Slider(
+        with gr.Row(equal_height=True):
+            input_ = Textbox(
+                label="Input",
+                placeholder="Quina és la finalitat del Servei Meteorològic de Catalunya?",
+            )
+
+        with gr.Row(equal_height=True):
+            num_chunks = Slider(
                             minimum=1,
                             maximum=6,
                             step=1,
                             value=2,
                             label="Number of chunks"
                         )
-
-            with gr.Column(variant="panel"):
-                output = Textbox(
-                    lines=10, 
-                    label="Context", 
-                    interactive=False, 
-                    show_copy_button=True
-                )
-                with gr.Accordion("Sources and context:", open=False):
-                    source_context = gr.Markdown(
-                        label="Sources",
-                        show_label=False,
-                    )
-                    with gr.Accordion("See full context evaluation:", open=False):
-                        context_evaluation = gr.Markdown(
-                            label="Full context",
-                            show_label=False,
-                        )
-                
-
- 
-
-        input_.change(
-            fn=None,
-            inputs=[input_],
-            api_name=False,
-            js="""(i, m) => {
-            document.getElementById('inputlength').textContent = i.length + '  '
-            document.getElementById('inputlength').style.color =  (i.length > m) ? "#ef4444" : "";
-        }""",
-        )
+        with gr.Row(equal_height=True):
+            clear_btn = Button("Clear")
+        with gr.Row(equal_height=True):
+            submit_btn = Button("Submit", variant="primary")
+                        
 
         clear_btn.click(
             fn=clear, 
             inputs=[], 
-            outputs=[input_, output, source_context, context_evaluation, num_chunks],
+            outputs=[input_, output, num_chunks],
               queue=False, 
               api_name=False
         )
@@ -110,7 +67,7 @@ def gradio_app():
         submit_btn.click(
             fn=eadop_rag, 
             inputs=[input_, num_chunks], 
-            outputs=[output, source_context, context_evaluation],
+            outputs=[output],
             api_name="get-eadop-rag"
         )
 

handler.py

@@ -1,14 +0,0 @@
-import json
-
-class ContentHandler():
-    content_type = "application/json"
-    accepts = "application/json"
-
-    def transform_input(self, prompt: str, model_kwargs: dict) -> bytes:
-        input_str = json.dumps({'inputs': prompt, 'parameters': model_kwargs})
-        return input_str.encode('utf-8')
-      
-    def transform_output(self, output: bytes) -> str:
-        response_json = json.loads(output.read().decode("utf-8"))        
-        return response_json[0]["generated_text"]
-    

input_reader.py

@@ -1,22 +0,0 @@
-from typing import List
-
-from llama_index.core.constants import DEFAULT_CHUNK_OVERLAP, DEFAULT_CHUNK_SIZE
-from llama_index.core.readers import SimpleDirectoryReader
-from llama_index.core.schema import Document
-from llama_index.core import Settings
-
-
-class InputReader:
-    def __init__(self, input_dir: str) -> None:
-        self.reader = SimpleDirectoryReader(input_dir=input_dir)
-
-    def parse_documents(
-        self,
-        show_progress: bool = True,
-        chunk_size: int = DEFAULT_CHUNK_SIZE,
-        chunk_overlap: int = DEFAULT_CHUNK_OVERLAP,
-    ) -> List[Document]:
-        Settings.chunk_size = chunk_size
-        Settings.chunk_overlap = chunk_overlap
-        documents = self.reader.load_data(show_progress=show_progress)
-        return documents

requirements.txt

@@ -1,4 +1,4 @@
-gradio==4.29.0
+gradio==4.44.1
 huggingface-hub==0.23.4
 openai==1.35.13
 python-dotenv==1.0.0

utils.py

@@ -3,12 +3,7 @@ import warnings
 
 from dotenv import load_dotenv
 
-
-from rag import RAG
-
 USER_INPUT = 100
-
-
 def setup():
     load_dotenv()
     warnings.filterwarnings("ignore")
@@ -16,18 +11,3 @@ def setup():
     logging.addLevelName(USER_INPUT, "USER_INPUT")
     logging.basicConfig(format="[%(levelname)s]: %(message)s", level=logging.INFO)
 
-
-def interactive(model: RAG):
-    logging.info("Write `exit` when you want to stop the model.")
-    print()
-
-    query = ""
-    while query.lower() != "exit":
-        logging.log(USER_INPUT, "Write the query or `exit`:")
-        query = input()
-
-        if query.lower() == "exit":
-            break
-
-        response = model.get_response(query)
-        print(response, end="\n\n")

rag.py → vectorstore.py

@@ -2,39 +2,24 @@ import logging
 import os
 import requests
 
-
-
 from langchain_community.vectorstores import FAISS
 from langchain_community.embeddings import HuggingFaceEmbeddings
 
-
-class RAG:
-    NO_ANSWER_MESSAGE: str = "Ho sento, no he pogut respondre la teva pregunta."
-
-    #vectorstore = "index-intfloat_multilingual-e5-small-500-100-CA-ES" # mixed
-    #vectorstore = "vectorestore" # CA only
+class VectorStore:
     vectorstore = "index-BAAI_bge-m3-1500-200-recursive_splitter-CA_ES_UE"
 
     def __init__(self, embeddings_model):
-
-        
         # load vectore store
         embeddings = HuggingFaceEmbeddings(model_name=embeddings_model, model_kwargs={'device': 'cpu'})
-        self.vectore_store = FAISS.load_local(self.vectorstore, embeddings, allow_dangerous_deserialization=True)#, allow_dangerous_deserialization=True)
-
+        self.vectore_store = FAISS.load_local(self.vectorstore, embeddings, allow_dangerous_deserialization=True)
         logging.info("RAG loaded!")
     
     def get_context(self, instruction, number_of_contexts=2):
-
         documentos = self.vectore_store.similarity_search_with_score(instruction, k=number_of_contexts)
-
-        return documentos
+        return self.beautiful_context(documentos)
         
-      
     def beautiful_context(self, docs):
-
         text_context = ""
-
         full_context = ""
         source_context = []
         for doc in docs:
@@ -44,12 +29,4 @@ class RAG:
             full_context += doc[0].page_content + "\n"
             source_context.append(doc[0].metadata["url"])
 
-        return text_context, full_context, source_context
-
-    def get_context(self, prompt: str, model_parameters: dict) -> str:
-        try:
-            docs = self.get_context(prompt, model_parameters["NUM_CHUNKS"])
-            return self.beautiful_context(docs)
-        except Exception as err:
-            print(err)
-            return None, None, None
+        return full_context