Create agent.py
Browse files
agent.py
ADDED
@@ -0,0 +1,204 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import time
|
3 |
+
import random
|
4 |
+
from dotenv import load_dotenv
|
5 |
+
from typing import List, Dict, Any, TypedDict, Annotated
|
6 |
+
import operator
|
7 |
+
|
8 |
+
from langchain_core.tools import tool
|
9 |
+
from langchain_community.tools.tavily_search import TavilySearchResults
|
10 |
+
from langchain_community.document_loaders import WikipediaLoader
|
11 |
+
from langchain_community.vectorstores import Chroma
|
12 |
+
from langchain.tools.retriever import create_retriever_tool
|
13 |
+
from langchain_text_splitters import RecursiveCharacterTextSplitter
|
14 |
+
from langchain_core.messages import SystemMessage, HumanMessage, AIMessage
|
15 |
+
from langchain_community.embeddings import SentenceTransformerEmbeddings
|
16 |
+
|
17 |
+
from langgraph.graph import StateGraph, START, END
|
18 |
+
from langgraph.checkpoint.memory import MemorySaver
|
19 |
+
|
20 |
+
# ---- Tool Definitions ----
|
21 |
+
@tool
|
22 |
+
def multiply(a: int, b: int) -> int:
|
23 |
+
"""Multiply two integers and return the product."""
|
24 |
+
return a * b
|
25 |
+
|
26 |
+
@tool
|
27 |
+
def add(a: int, b: int) -> int:
|
28 |
+
"""Add two integers and return the sum."""
|
29 |
+
return a + b
|
30 |
+
|
31 |
+
@tool
|
32 |
+
def subtract(a: int, b: int) -> int:
|
33 |
+
"""Subtract the second integer from the first and return the difference."""
|
34 |
+
return a - b
|
35 |
+
|
36 |
+
@tool
|
37 |
+
def divide(a: int, b: int) -> float:
|
38 |
+
"""Divide the first integer by the second and return the quotient."""
|
39 |
+
if b == 0:
|
40 |
+
raise ValueError("Cannot divide by zero.")
|
41 |
+
return a / b
|
42 |
+
|
43 |
+
@tool
|
44 |
+
def modulus(a: int, b: int) -> int:
|
45 |
+
"""Return the remainder of the division of the first integer by the second."""
|
46 |
+
return a % b
|
47 |
+
|
48 |
+
@tool
|
49 |
+
def optimized_web_search(query: str) -> str:
|
50 |
+
"""Perform an optimized web search using TavilySearchResults and return concatenated document snippets."""
|
51 |
+
try:
|
52 |
+
time.sleep(random.uniform(1, 2))
|
53 |
+
docs = TavilySearchResults(max_results=2).invoke(query=query)
|
54 |
+
return "\n\n---\n\n".join(
|
55 |
+
f"<Doc url='{d.get('url','')}'>{d.get('content','')[:500]}</Doc>"
|
56 |
+
for d in docs
|
57 |
+
)
|
58 |
+
except Exception as e:
|
59 |
+
return f"Web search failed: {e}"
|
60 |
+
|
61 |
+
@tool
|
62 |
+
def optimized_wiki_search(query: str) -> str:
|
63 |
+
"""Perform an optimized Wikipedia search and return concatenated document snippets."""
|
64 |
+
try:
|
65 |
+
time.sleep(random.uniform(0.5, 1))
|
66 |
+
docs = WikipediaLoader(query=query, load_max_docs=1).load()
|
67 |
+
return "\n\n---\n\n".join(
|
68 |
+
f"<Doc src='{d.metadata['source']}'>{d.page_content[:800]}</Doc>"
|
69 |
+
for d in docs
|
70 |
+
)
|
71 |
+
except Exception as e:
|
72 |
+
return f"Wikipedia search failed: {e}"
|
73 |
+
|
74 |
+
# ---- LLM Integrations ----
|
75 |
+
load_dotenv()
|
76 |
+
|
77 |
+
from langchain_groq import ChatGroq
|
78 |
+
from langchain_nvidia_ai_endpoints import ChatNVIDIA
|
79 |
+
from google import genai
|
80 |
+
|
81 |
+
import requests
|
82 |
+
|
83 |
+
def baidu_ernie_generate(prompt, api_key=None):
|
84 |
+
url = "https://api.baidu.com/ernie/v1/generate"
|
85 |
+
headers = {"Authorization": f"Bearer {api_key}"}
|
86 |
+
data = {"model": "ernie-4.5", "prompt": prompt}
|
87 |
+
try:
|
88 |
+
resp = requests.post(url, headers=headers, json=data, timeout=30)
|
89 |
+
return resp.json().get("result", "")
|
90 |
+
except Exception as e:
|
91 |
+
return f"ERNIE API error: {e}"
|
92 |
+
|
93 |
+
def deepseek_generate(prompt, api_key=None):
|
94 |
+
url = "https://api.deepseek.com/v1/chat/completions"
|
95 |
+
headers = {"Authorization": f"Bearer {api_key}"}
|
96 |
+
data = {"model": "deepseek-chat", "messages": [{"role": "user", "content": prompt}]}
|
97 |
+
try:
|
98 |
+
resp = requests.post(url, headers=headers, json=data, timeout=30)
|
99 |
+
choices = resp.json().get("choices", [{}])
|
100 |
+
if choices and "message" in choices[0]:
|
101 |
+
return choices[0]["message"].get("content", "")
|
102 |
+
return ""
|
103 |
+
except Exception as e:
|
104 |
+
return f"DeepSeek API error: {e}"
|
105 |
+
|
106 |
+
class EnhancedAgentState(TypedDict):
|
107 |
+
messages: Annotated[List[HumanMessage|AIMessage], operator.add]
|
108 |
+
query: str
|
109 |
+
agent_type: str
|
110 |
+
final_answer: str
|
111 |
+
perf: Dict[str,Any]
|
112 |
+
agno_resp: str
|
113 |
+
|
114 |
+
class HybridLangGraphMultiLLMSystem:
|
115 |
+
def __init__(self):
|
116 |
+
self.tools = [
|
117 |
+
multiply, add, subtract, divide, modulus,
|
118 |
+
optimized_web_search, optimized_wiki_search
|
119 |
+
]
|
120 |
+
self.graph = self._build_graph()
|
121 |
+
|
122 |
+
def _build_graph(self):
|
123 |
+
groq_llm = ChatGroq(model="llama3-70b-8192", temperature=0, api_key=os.getenv("GROQ_API_KEY"))
|
124 |
+
nvidia_llm = ChatNVIDIA(model="meta/llama3-70b-instruct", temperature=0, api_key=os.getenv("NVIDIA_API_KEY"))
|
125 |
+
|
126 |
+
def router(st: EnhancedAgentState) -> EnhancedAgentState:
|
127 |
+
q = st["query"].lower()
|
128 |
+
if "groq" in q: t = "groq"
|
129 |
+
elif "nvidia" in q: t = "nvidia"
|
130 |
+
elif "gemini" in q or "google" in q: t = "gemini"
|
131 |
+
elif "deepseek" in q: t = "deepseek"
|
132 |
+
elif "ernie" in q or "baidu" in q: t = "baidu"
|
133 |
+
else: t = "groq" # default
|
134 |
+
return {**st, "agent_type": t}
|
135 |
+
|
136 |
+
def groq_node(st: EnhancedAgentState) -> EnhancedAgentState:
|
137 |
+
t0 = time.time()
|
138 |
+
sys = SystemMessage(content="Answer as an expert.")
|
139 |
+
res = groq_llm.invoke([sys, HumanMessage(content=st["query"])])
|
140 |
+
return {**st, "final_answer": res.content, "perf": {"time": time.time() - t0, "prov": "Groq"}}
|
141 |
+
|
142 |
+
def nvidia_node(st: EnhancedAgentState) -> EnhancedAgentState:
|
143 |
+
t0 = time.time()
|
144 |
+
sys = SystemMessage(content="Answer as an expert.")
|
145 |
+
res = nvidia_llm.invoke([sys, HumanMessage(content=st["query"])])
|
146 |
+
return {**st, "final_answer": res.content, "perf": {"time": time.time() - t0, "prov": "NVIDIA"}}
|
147 |
+
|
148 |
+
def gemini_node(st: EnhancedAgentState) -> EnhancedAgentState:
|
149 |
+
t0 = time.time()
|
150 |
+
genai.configure(api_key=os.getenv("GEMINI_API_KEY"))
|
151 |
+
model = genai.GenerativeModel("gemini-1.5-pro-latest")
|
152 |
+
res = model.generate_content(st["query"])
|
153 |
+
return {**st, "final_answer": res.text, "perf": {"time": time.time() - t0, "prov": "Gemini"}}
|
154 |
+
|
155 |
+
def deepseek_node(st: EnhancedAgentState) -> EnhancedAgentState:
|
156 |
+
t0 = time.time()
|
157 |
+
resp = deepseek_generate(st["query"], api_key=os.getenv("DEEPSEEK_API_KEY"))
|
158 |
+
return {**st, "final_answer": resp, "perf": {"time": time.time() - t0, "prov": "DeepSeek"}}
|
159 |
+
|
160 |
+
def baidu_node(st: EnhancedAgentState) -> EnhancedAgentState:
|
161 |
+
t0 = time.time()
|
162 |
+
resp = baidu_ernie_generate(st["query"], api_key=os.getenv("BAIDU_API_KEY"))
|
163 |
+
return {**st, "final_answer": resp, "perf": {"time": time.time() - t0, "prov": "ERNIE"}}
|
164 |
+
|
165 |
+
def pick(st: EnhancedAgentState) -> str:
|
166 |
+
return st["agent_type"]
|
167 |
+
|
168 |
+
g = StateGraph(EnhancedAgentState)
|
169 |
+
g.add_node("router", router)
|
170 |
+
g.add_node("groq", groq_node)
|
171 |
+
g.add_node("nvidia", nvidia_node)
|
172 |
+
g.add_node("gemini", gemini_node)
|
173 |
+
g.add_node("deepseek", deepseek_node)
|
174 |
+
g.add_node("baidu", baidu_node)
|
175 |
+
g.set_entry_point("router")
|
176 |
+
g.add_conditional_edges("router", pick, {
|
177 |
+
"groq": "groq",
|
178 |
+
"nvidia": "nvidia",
|
179 |
+
"gemini": "gemini",
|
180 |
+
"deepseek": "deepseek",
|
181 |
+
"baidu": "baidu"
|
182 |
+
})
|
183 |
+
for n in ["groq", "nvidia", "gemini", "deepseek", "baidu"]:
|
184 |
+
g.add_edge(n, END)
|
185 |
+
return g.compile(checkpointer=MemorySaver())
|
186 |
+
|
187 |
+
def process_query(self, q: str) -> str:
|
188 |
+
state = {
|
189 |
+
"messages": [HumanMessage(content=q)],
|
190 |
+
"query": q,
|
191 |
+
"agent_type": "",
|
192 |
+
"final_answer": "",
|
193 |
+
"perf": {},
|
194 |
+
"agno_resp": ""
|
195 |
+
}
|
196 |
+
cfg = {"configurable": {"thread_id": f"hyb_{hash(q)}"}}
|
197 |
+
out = self.graph.invoke(state, cfg)
|
198 |
+
raw_answer = out["final_answer"]
|
199 |
+
parts = raw_answer.split('\n\n', 1)
|
200 |
+
answer_part = parts[1].strip() if len(parts) > 1 else raw_answer.strip()
|
201 |
+
return answer_part
|
202 |
+
|
203 |
+
def build_graph():
|
204 |
+
return HybridLangGraphMultiLLMSystem().graph
|