Merge pull request #166 from FrancescoCaracciolo/master

Big code and performance improvements

Author: qwersyk

data/io.github.qwersyk.Newelle.gschema.xml

@@ -127,6 +127,9 @@
         </key>
         <key name="current-profile" type="s">
           <default>"Assistant"</default>
+        </key>
+        <key name="user-name" type="s">
+          <default>"User"</default>
         </key>
 	      <key name="startup-mode" type="s">
           <default>"normal"</default>

io.github.qwersyk.Newelle.json

@@ -1,7 +1,7 @@
 {
   "app-id": "io.github.qwersyk.Newelle",
   "runtime": "org.gnome.Platform",
-  "runtime-version": "47",
+  "runtime-version": "48",
   "sdk": "org.gnome.Sdk",
   "command": "newelle",
   "finish-args": [

src/constants.py

@@ -6,6 +6,9 @@
 from .handlers.memory import MemoripyHandler, UserSummaryHandler, SummaryMemoripyHanlder
 from .handlers.rag import LlamaIndexHanlder
 
+DIR_NAME = "Newelle"
+SCHEMA_ID = 'io.github.qwersyk.Newelle'
+
 AVAILABLE_LLMS = {
     "newelle": {
         "key": "newelle",

src/controller.py

@@ -55,6 +55,8 @@ def get_models(self):
                 models = client.models.list()
                 result = tuple()
                 for model in models:
+                    if "embedding" in model.display_name.lower() or "legacy" in model.display_name.lower():
+                        continue
                     result += ((model.display_name, model.name,),)
                 self.models = result
                 self.set_setting("models", json.dumps(result))

src/handlers/llm/gemini_handler.py

@@ -12,6 +12,7 @@ class LLMHandler(Handler):
     schema_key = "llm-settings"
 
     def __init__(self, settings, path):
+        super().__init__(settings, path)
         self.settings = settings
         self.path = path
 

src/handlers/llm/llm.py

@@ -10,6 +10,8 @@ def __init__(self, settings, path):
         super().__init__(settings, path)
         self.memoripy = None
         self.user_summary = None
+        self.llm = None 
+        self.embedding = None
 
     def is_installed(self) -> bool:
         memoripy, user_summary = self.initialize_handlers() 

src/handlers/memory/summary_memoripy_handler.py

@@ -1,9 +1,9 @@
 from typing import Any, List
-import threading
-from ...handlers.llm import LLMHandler
-from ...handlers.embeddings.embedding import EmbeddingHandler
-from ...handlers import ExtraSettings
-from .rag_handler import RAGHandler
+import threading 
+from ...handlers.llm import LLMHandler 
+from ...handlers.embeddings.embedding import EmbeddingHandler 
+from ...handlers import ExtraSettings 
+from .rag_handler import RAGHandler, RAGIndex 
 from ...utility.pip import find_module, install_module
 import os
 
@@ -141,6 +141,7 @@ def get_paths(self):
             documents_path = self.documents_path
             data_path = self.data_path
         return documents_path, data_path
+    
     def create_index(self, button=None):  
         if not self.is_installed():
             return
@@ -170,16 +171,11 @@ def create_index(self, button=None):
             print(e)
             self.indexing = False
             self.indexing_status = 1
-   
-    def query_document(self, prompt: str, documents: list[str], chunk_size: int|None = None) -> list[str]: 
-        from llama_index.core.settings import Settings
-        from llama_index.core import VectorStoreIndex, SimpleDirectoryReader, Document
-        from llama_index.core.retrievers import VectorIndexRetriever
+  
+    @staticmethod 
+    def parse_document_list(documents: list[str]):
+        from llama_index.core import SimpleDirectoryReader, Document
         import requests
-        self.llm.load_model(None)
-        self.embedding.load_model()
-        Settings.embed_model = self.get_embedding_adapter(self.embedding)
-        chunk_size = int(self.get_setting("chunk_size")) if chunk_size is None else chunk_size
         document_list = []
         urls = []
         for document in documents:
@@ -192,20 +188,28 @@ def query_document(self, prompt: str, documents: list[str], chunk_size: int|None
             elif document.startswith("url:"):
                 url = document.lstrip("url:")
                 urls.append(url)
+        t = []
+        for url in urls:
+            def request(url):
+                r = requests.get(url)
+                document_list.append(Document(text=r.text))
+            th = threading.Thread(target=request, args=(url, ))
+            th.start()
+        [t.join() for t in t]
+        return document_list
+    
+    def build_index(self, documents: list[str], chunk_size: int | None = None) -> RAGIndex: 
+        from llama_index.core.settings import Settings
+        from llama_index.core import VectorStoreIndex, SimpleDirectoryReader, Document
+        from llama_index.core.retrievers import VectorIndexRetriever
+        import requests
+        self.llm.load_model(None)
+        self.embedding.load_model()
+        Settings.embed_model = self.get_embedding_adapter(self.embedding)
+        chunk_size = int(self.get_setting("chunk_size")) if chunk_size is None else chunk_size
+        document_list = self.parse_document_list(documents)
         index = VectorStoreIndex.from_documents(document_list)
-        retriever = VectorIndexRetriever(
-            index=index,
-            similarity_top_k=int(self.get_setting("return_documents")),
-        )
-        r = []
-        nodes = retriever.retrieve(prompt)
-        for node in nodes:
-            if node.score < float(self.get_setting("similarity_threshold")):
-                continue
-            r.append("--")
-            r.append("- Source: " + node.metadata.get("file_name"))
-            r.append(node.node.get_content())
-        return r
+        return LlamaIndexIndex(index, int(self.get_setting("return_documents")), float(self.get_setting("similarity_threshold"))) 
 
     def get_embedding_adapter(self, embedding: EmbeddingHandler):
         from llama_index.core.embeddings import BaseEmbedding
@@ -271,3 +275,32 @@ def stream_complete(
         adapter = LLMAdapter()
         adapter.set_llm(self.llm) 
         return adapter
+
+
+class LlamaIndexIndex(RAGIndex):
+    def __init__(self, index, return_documents, similarity_threshold):
+        self.index = index
+        self.retriever = None
+        self.return_documents = return_documents
+        self.similarity_threshold = similarity_threshold
+
+    def query(self, query: str) -> list[str]:
+        from llama_index.core.retrievers import VectorIndexRetriever
+        if self.retriever is None:
+            retriever = VectorIndexRetriever(
+                index=self.index,
+                similarity_top_k=int(self.return_documents))
+            self.retriever = retriever 
+        r = []
+        nodes = self.retriever.retrieve(query)
+        for node in nodes:
+            if node.score < float(self.similarity_threshold):
+                continue
+            r.append("--")
+            r.append(node.node.get_content())
+        return r
+
+    def insert(self, documents: list[str]):
+        documents_list = LlamaIndexHanlder.parse_document_list(documents)
+        for document in documents_list:
+            self.index.insert(document)

src/handlers/rag/llamaindex_handler.py

@@ -5,6 +5,18 @@
 from abc import abstractmethod
 import os
 
+class RAGIndex:
+    def __init__(self):
+        pass
+
+    @abstractmethod 
+    def query(self, query:str) -> list[str]:
+        pass
+
+    @abstractmethod
+    def insert(self, documents: list[str]):
+        pass
+
 class RAGHandler(Handler):
     key = ""
     schema_key = "rag-settings"
@@ -42,7 +54,6 @@ def load(self):
     def get_context(self, prompt:str, history: list[dict[str, str]]) -> list[str]:
         return []
 
-    @abstractmethod
     def query_document(self, prompt: str, documents: list[str], chunk_size: int|None = None) -> list[str]:
         """
         Query the document
@@ -58,8 +69,13 @@ def query_document(self, prompt: str, documents: list[str], chunk_size: int|None
         Returns:
             The query result 
         """
-        pass
-    
+        index = self.build_index(documents, chunk_size)
+        return index.query(prompt)
+
+    @abstractmethod
+    def build_index(self, documents: list[str], chunk_size: int|None = None) -> RAGIndex:
+       pass
+
     @abstractmethod 
     def index_exists(self) -> bool:
         """
@@ -108,4 +124,3 @@ def index_button_pressed(self, button=None):
             self.indexing = True
             self.create_index()
 
-

src/handlers/rag/rag_handler.py

@@ -144,7 +144,7 @@ def thread_editing_action(self, *a):
         threadediting.present()
 
     def settings_action(self, *a):
-        settings = Settings(self)
+        settings = Settings(self, self.win.controller)
         settings.present()
         settings.connect("close-request", self.close_settings)
         self.settingswindow = settings