Merge branch 'astream_experiment'

benchmarking/readme.md

@@ -1,2 +1,3 @@
 ### Benchmarking
-Standardized tasks and tests for Jarvis to evaluate performance.
+Standardized tasks and tests for Jarvis to evaluate performance.
+

core/agent.py

@@ -10,6 +10,14 @@
 
 
 
+class Agent1:
+    llm = ChatOpenAI(
+        model = Model.gpt_4o,
+        temperature=0,
+        max_tokens=512,
+    )
+
+
 class Agent:
     def __init__(self, model_type) -> None:
         #Langsmith Tracing
@@ -76,19 +84,14 @@ def run(self, user_prompt: str) -> tuple[str, int]:
         response and the total amount of tokens used.
         """
         first = True
-        for event in self.graph.stream({"messages": [("user", user_prompt)]}):
+        for event in self.graph.stream("tell me about orcas?"):
             for value in event.values():
                 messages = value["messages"][-1]
                 gathered = ""
                 # if messages.content and not isinstance(messages, HumanMessage):
                 #     print(messages.content, end="|", flush=True)
 
-                if isinstance(messages, AIMessageChunk):
-                    if first:
-                        gathered = messages
-                        first = False
-                    else:
-                        gathered += messages
+                gathered += messages
 
                 if isinstance(messages, BaseMessage):
                     if hasattr(messages, 'usage_metadata'):

core/graph.py → core/graphAgent.py

@@ -8,6 +8,9 @@
 import json
 from config import OPENAI_API_KEY
 from Agents.simpleagent import SimpleAgent
+from agent import Agent, Agent1 
+import asyncio
+from time import sleep
 
 
 class Graph:
@@ -57,29 +60,44 @@ def run_stream_only(self, user_prompt: str):
             yield chunk.content
 
     #for running the agent comment out for testing in terminal
-    def run(self, user_prompt: str) -> tuple[str, int]:
+    async def run(self, user_prompt: str, socketio) -> tuple[str, int]:
         """
         Run the agent with a user prompt and return a tuple containing the llm 
         response and the total amount of tokens used.
         """
-        first = True
-        for event in self.graph.stream({"messages": [("user", user_prompt)]}):
-            #print(event)
-            for value in event.values():
-                messages = value["messages"][-1]
-                gathered = ""
-                # if messages.content and not isinstance(messages, HumanMessage):
-                #     print(messages.content, end="|", flush=True)
+        try:
+            input = {"messages": [("human", user_prompt)]}
+            socketio.emit("start_message", " ")
+            async for chunk in self.graph.astream(input, stream_mode="values"):
+                if type(chunk["messages"][-1]) == HumanMessage:
+                    continue
+                event_message = chunk["messages"][-1].content
+                event_message = event_message.split(" ")
+                for word in event_message:
+                    sleep(0.05)
+                    socketio.emit("chunk", word+" ")
+                socketio.emit("chunk", "<br>")
+            socketio.emit("tokens", 0) # a way to emit ending of the message
+            return "success"
+        except Exception as e:
+            print(e)
+            return "error"
 
-                if isinstance(messages, AIMessageChunk):
-                    if first:
-                        gathered = messages
-                        first = False
-                    else:
-                        gathered += messages
+        # for event in self.graph.stream(input):
+            #print(event)
+            # for value in event.values():
+            #     messages = value["messages"][-1]
+            #     gathered = ""
+            #     # if messages.content and not isinstance(messages, HumanMessage):
+            #     #     print(messages.content, end="|", flush=True)
 
-                if isinstance(messages, BaseMessage):
-                    total_tokens = messages.usage_metadata.get('total_tokens', 0)
-
+            #     if isinstance(messages, AIMessageChunk):
+            #         if first:
+            #             gathered = messages
+            #             first = False
+            #         else:
+            #             gathered += messages
 
-                    return messages.content, total_tokens
+            #     if isinstance(messages, BaseMessage):
+            #         total_tokens = messages.usage_metadata.get('total_tokens', 0)
+            #         return messages.content, total_tokens

core/main.py

@@ -1,5 +1,5 @@
 from flask import Flask, request, url_for, jsonify
-from graph import Graph
+from graphAgent import Graph
 from models import Model
 from summarize_chat import summarize_chat
 from rag import embed_and_store
@@ -22,18 +22,11 @@
 #
 #
 #   HTTP(S) routes below
-#s
+#
 #
 @app.route("/")
 def hello_world():
     return app.send_static_file('index.html')
-
-@app.route('/send_message', methods=['POST', 'GET'])
-def llm_request():
-    if(request.method == 'POST'):
-        data = request.json
-        ai_message = jarvis.run(data['prompt'])
-        return {"message": ai_message}
 
 @app.route('/vectorize_chat', methods=['POST'])
 def summarize_store():
@@ -49,6 +42,8 @@ def summarize_store():
 
     return {"status": "success", "summary": summary}
 
+
+
 #
 #
 #   Socket.IO events below
@@ -68,24 +63,12 @@ def disconnect():
 # Custom event. Fired when the user sends a prompt.
 @socketio.on('user_prompt')
 def handle_prompt(data):
-    print("huh")
     try:
         conversation_id = data['conversation_id'] # grabs the conversation ID
         socketio.emit("start_message")
-        stream, tokens = jarvis.run(data['prompt']) # prompts Jarvis
-        #stream = jarvis.run_stream_only(data['prompt'])
-        chunk = ""
-        for char in stream:
-            if len(chunk) > 4:
-                socketio.emit("chunk", chunk)
-                chunk = char
-            else:
-                chunk += char
-        #asyncio.sleep(500)
-        socketio.emit("chunk", chunk)
-        socketio.emit("tokens", tokens)
-
-        return jsonify({"status": "success"})
+        asyncio.run(jarvis.run(data['prompt'], socketio), debug=True) # prompts Jarvis and hands off emitting to the graphAgent.
+
+        return jsonify({"status": response})
     except Exception as e:
         print(f'Something very bad happened: {e}')
         return jsonify({"status": "error"})

core/static/chat.js

@@ -47,13 +47,6 @@ endStreamedAIMessage = () => {
 
 }
 
-let startStreamedAIMessage = (uuid) => {
-    console.log("Message start")
-    addMessage(uuid); // Create an AI message when it begins streaming.
-    let ai_message = document.getElementById(uuid)
-    state.activeAIMessage = ai_message // Active element gets added to the state.
-}
-
 // Generates unique id on socket.on("start_message")
 let generateUUID = () => {
     return "10000000-1000-4000-8000-100000000000".replace(/[018]/g, c =>

core/static/index.css

@@ -146,11 +146,11 @@ body {
 }
 
 .chat{
-    width: 65%;
+    width: 100%;
 }
 
 .chatHistory {
-    width: 250px;
+    width: 400px;
     height: 80vh;
     margin: 6px;
     margin-top: 8px;
@@ -205,7 +205,7 @@ p{
 }
 
 .processesContainer{
-    width: 250px;
+    width: 400px;
     height: 80vh;
     margin-top: 8px;
     border-radius: 10px;

core/static/socketEvents.js

@@ -12,7 +12,13 @@ let uuid = 0
 
 // prints chunks that are streamed to the console and adds them to the chat
 socket.on("chunk", async (chunk)=>{
-    console.log(chunk);
+    if(!state.activeAIMessage){
+        console.log("STARTED MESSAGE")
+        uuid = generateUUID();
+        await addStreamedMessage(uuid, "");
+        ai_message = document.getElementById(uuid)
+        state.activeAIMessage = ai_message
+    }
     await addStreamedMessage(uuid, chunk);
 })
 
@@ -23,9 +29,7 @@ socket.on("tokens", async (tokens) => {
 })
 
 socket.on("start_message", async () => {
-    uuid = generateUUID();
-    console.log(uuid);
-    await addStreamedMessage(uuid, "");
+
 }) 
 
 // Remember to parse the streamed response

docker-compose.yml

@@ -1,5 +1,4 @@
 services:
-
   llm-service:
     build: ./core
     restart: unless-stopped