Spaces:
Runtime error
Runtime error
Update agent.py
Browse files
agent.py
CHANGED
@@ -1,47 +1,38 @@
|
|
|
|
|
|
1 |
import os
|
2 |
-
import re
|
3 |
import time
|
4 |
import functools
|
5 |
-
from typing import Dict, Any, List
|
6 |
-
|
7 |
import pandas as pd
|
|
|
|
|
8 |
|
9 |
-
# LangGraph
|
10 |
from langgraph.graph import StateGraph, START, END, MessagesState
|
11 |
-
from langgraph.prebuilt import ToolNode
|
12 |
-
|
13 |
-
# LangChain Core
|
14 |
from langchain_core.messages import SystemMessage, HumanMessage
|
15 |
from langchain_core.tools import tool
|
16 |
-
|
17 |
-
# Google Gemini
|
18 |
from langchain_google_genai import ChatGoogleGenerativeAI
|
19 |
-
|
20 |
-
# Tools
|
21 |
from langchain_community.tools.tavily_search import TavilySearchResults
|
22 |
from langchain_community.utilities.wikipedia import WikipediaAPIWrapper
|
23 |
|
24 |
-
# Python REPL Tool
|
25 |
try:
|
26 |
from langchain_experimental.tools.python.tool import PythonAstREPLTool
|
27 |
except ImportError:
|
28 |
from langchain.tools.python.tool import PythonAstREPLTool
|
29 |
|
30 |
# ---------------------------------------------------------------------
|
31 |
-
#
|
32 |
# ---------------------------------------------------------------------
|
33 |
-
|
34 |
if os.getenv("LANGCHAIN_API_KEY"):
|
35 |
os.environ["LANGCHAIN_TRACING_V2"] = "true"
|
36 |
os.environ["LANGCHAIN_ENDPOINT"] = "https://api.smith.langchain.com"
|
37 |
os.environ.setdefault("LANGCHAIN_PROJECT", "gaia-agent")
|
38 |
-
print("
|
39 |
|
40 |
# ---------------------------------------------------------------------
|
41 |
-
#
|
42 |
# ---------------------------------------------------------------------
|
43 |
def error_guard(fn):
|
44 |
-
"""Fängt Tool-Fehler ab & gibt String zurück (bricht Agent nicht ab)."""
|
45 |
@functools.wraps(fn)
|
46 |
def wrapper(*args, **kw):
|
47 |
try:
|
@@ -50,77 +41,47 @@ def error_guard(fn):
|
|
50 |
return f"ERROR: {e}"
|
51 |
return wrapper
|
52 |
|
53 |
-
|
54 |
-
def with_backoff(fn, tries: int = 4, delay: int = 4):
|
55 |
-
"""Synchrones Retry-Wrapper für LLM-Aufrufe."""
|
56 |
-
for t in range(tries):
|
57 |
-
try:
|
58 |
-
return fn()
|
59 |
-
except Exception as e:
|
60 |
-
if ("429" in str(e) or "RateLimit" in str(e)) and t < tries - 1:
|
61 |
-
time.sleep(delay)
|
62 |
-
delay *= 2
|
63 |
-
continue
|
64 |
-
raise
|
65 |
-
|
66 |
# ---------------------------------------------------------------------
|
67 |
-
#
|
68 |
# ---------------------------------------------------------------------
|
69 |
@tool
|
70 |
@error_guard
|
71 |
def parse_csv(file_path: str, query: str = "") -> str:
|
72 |
-
"""Load a CSV file and (optional) run a pandas query."""
|
73 |
df = pd.read_csv(file_path)
|
74 |
if not query:
|
75 |
return f"Rows={len(df)}, Cols={list(df.columns)}"
|
76 |
-
|
77 |
-
return df.query(query).to_markdown(index=False)
|
78 |
-
except Exception as e:
|
79 |
-
return f"ERROR query: {e}"
|
80 |
-
|
81 |
|
82 |
@tool
|
83 |
@error_guard
|
84 |
def parse_excel(file_path: str, sheet: str | int | None = None, query: str = "") -> str:
|
85 |
-
"""Load an Excel sheet (name or index) and (optional) run a pandas query."""
|
86 |
sheet_arg = int(sheet) if isinstance(sheet, str) and sheet.isdigit() else sheet or 0
|
87 |
df = pd.read_excel(file_path, sheet_name=sheet_arg)
|
88 |
if not query:
|
89 |
return f"Rows={len(df)}, Cols={list(df.columns)}"
|
90 |
-
|
91 |
-
return df.query(query).to_markdown(index=False)
|
92 |
-
except Exception as e:
|
93 |
-
return f"ERROR query: {e}"
|
94 |
|
95 |
-
# ---------------------------------------------------------------------
|
96 |
-
# 3) Externe Search-Tools (Tavily, Wikipedia)
|
97 |
-
# ---------------------------------------------------------------------
|
98 |
@tool
|
99 |
@error_guard
|
100 |
def web_search(query: str, max_results: int = 5) -> str:
|
101 |
-
"""Search the web via Tavily and return markdown list of results."""
|
102 |
api_key = os.getenv("TAVILY_API_KEY")
|
103 |
hits = TavilySearchResults(max_results=max_results, api_key=api_key).invoke(query)
|
104 |
if not hits:
|
105 |
return "No results."
|
106 |
return "\n".join(f"{h['title']} – {h['url']}" for h in hits)
|
107 |
|
108 |
-
|
109 |
@tool
|
110 |
@error_guard
|
111 |
def wiki_search(query: str, sentences: int = 3) -> str:
|
112 |
-
"""Quick Wikipedia summary."""
|
113 |
wrapper = WikipediaAPIWrapper(top_k_results=1, doc_content_chars_max=4000)
|
114 |
res = wrapper.run(query)
|
115 |
return "\n".join(res.split(". ")[:sentences]) if res else "No article found."
|
116 |
|
117 |
-
#
|
118 |
-
# 4) Python-REPL Tool (fertig aus LangChain)
|
119 |
-
# ---------------------------------------------------------------------
|
120 |
python_repl = PythonAstREPLTool()
|
121 |
|
122 |
# ---------------------------------------------------------------------
|
123 |
-
#
|
124 |
# ---------------------------------------------------------------------
|
125 |
gemini_llm = ChatGoogleGenerativeAI(
|
126 |
google_api_key=os.getenv("GOOGLE_API_KEY"),
|
@@ -129,81 +90,54 @@ gemini_llm = ChatGoogleGenerativeAI(
|
|
129 |
max_output_tokens=2048,
|
130 |
)
|
131 |
|
132 |
-
# ---------------------------------------------------------------------
|
133 |
-
# 6) System-Prompt (ReAct, keine Prefixe im Final-Output!)
|
134 |
-
# ---------------------------------------------------------------------
|
135 |
SYSTEM_PROMPT = SystemMessage(
|
136 |
content=(
|
137 |
-
"You are a helpful assistant with access to
|
138 |
-
"
|
139 |
-
"
|
140 |
-
"Tool: <tool_name>\n"
|
141 |
-
"Input: <input for the tool>\n\n"
|
142 |
-
"Wait for the tool result before continuing.\n"
|
143 |
-
"When you know the final answer, reply with the answer **only**.\n"
|
144 |
-
"Don't include any prefix, explanation or formatting around the answer.\n"
|
145 |
-
"Answer formatting:\n"
|
146 |
-
"- For numbers: no units unless requested\n"
|
147 |
-
"- For strings: no articles or abbreviations\n"
|
148 |
-
"- For lists: comma + space separated, correct order\n"
|
149 |
)
|
150 |
)
|
151 |
|
|
|
|
|
152 |
# ---------------------------------------------------------------------
|
153 |
-
#
|
154 |
# ---------------------------------------------------------------------
|
155 |
-
|
156 |
def planner(state: MessagesState):
|
157 |
-
|
158 |
-
if
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
|
163 |
-
# WICHTIG: Gib tool_calls weiter – sie lösen im ToolNode die Ausführung aus
|
164 |
-
return {
|
165 |
-
"messages": msgs + [resp],
|
166 |
-
"should_end": (
|
167 |
-
not getattr(resp, "tool_calls", None) # kein Tool gewünscht
|
168 |
-
and "\n" not in resp.content # einfache Heuristik
|
169 |
-
)
|
170 |
-
}
|
171 |
|
172 |
-
def
|
173 |
-
|
174 |
-
|
175 |
-
# Tool-Knoten
|
176 |
-
TOOLS = [web_search, wiki_search, parse_csv, parse_excel, python_repl]
|
177 |
|
|
|
|
|
|
|
178 |
graph = StateGraph(MessagesState)
|
179 |
-
|
180 |
graph.add_node("planner", planner)
|
181 |
graph.add_node("tools", ToolNode(TOOLS))
|
182 |
-
|
183 |
graph.add_edge(START, "planner")
|
184 |
-
graph.
|
185 |
-
|
186 |
-
|
187 |
-
"tools": "tools",
|
188 |
-
|
189 |
-
|
190 |
|
191 |
-
# compile → LangGraph-Executor
|
192 |
agent_executor = graph.compile()
|
193 |
|
194 |
# ---------------------------------------------------------------------
|
195 |
-
#
|
196 |
# ---------------------------------------------------------------------
|
197 |
class GaiaAgent:
|
198 |
-
"""LangChain·LangGraph-Agent für GAIA Level 1."""
|
199 |
-
|
200 |
def __init__(self):
|
201 |
print("✅ GaiaAgent initialised (LangGraph)")
|
202 |
|
203 |
def __call__(self, task_id: str, question: str) -> str:
|
204 |
-
""
|
205 |
-
|
206 |
-
|
207 |
-
# letze Message enthält Antwort
|
208 |
-
answer = final_state["messages"][-1].content
|
209 |
-
return answer.strip()
|
|
|
1 |
+
# agent.py
|
2 |
+
|
3 |
import os
|
|
|
4 |
import time
|
5 |
import functools
|
|
|
|
|
6 |
import pandas as pd
|
7 |
+
from typing import Dict, Any, List
|
8 |
+
import re
|
9 |
|
|
|
10 |
from langgraph.graph import StateGraph, START, END, MessagesState
|
11 |
+
from langgraph.prebuilt import ToolNode
|
|
|
|
|
12 |
from langchain_core.messages import SystemMessage, HumanMessage
|
13 |
from langchain_core.tools import tool
|
|
|
|
|
14 |
from langchain_google_genai import ChatGoogleGenerativeAI
|
|
|
|
|
15 |
from langchain_community.tools.tavily_search import TavilySearchResults
|
16 |
from langchain_community.utilities.wikipedia import WikipediaAPIWrapper
|
17 |
|
|
|
18 |
try:
|
19 |
from langchain_experimental.tools.python.tool import PythonAstREPLTool
|
20 |
except ImportError:
|
21 |
from langchain.tools.python.tool import PythonAstREPLTool
|
22 |
|
23 |
# ---------------------------------------------------------------------
|
24 |
+
# LangSmith optional
|
25 |
# ---------------------------------------------------------------------
|
|
|
26 |
if os.getenv("LANGCHAIN_API_KEY"):
|
27 |
os.environ["LANGCHAIN_TRACING_V2"] = "true"
|
28 |
os.environ["LANGCHAIN_ENDPOINT"] = "https://api.smith.langchain.com"
|
29 |
os.environ.setdefault("LANGCHAIN_PROJECT", "gaia-agent")
|
30 |
+
print("📱 LangSmith tracing enabled.")
|
31 |
|
32 |
# ---------------------------------------------------------------------
|
33 |
+
# Fehler-Wrapper
|
34 |
# ---------------------------------------------------------------------
|
35 |
def error_guard(fn):
|
|
|
36 |
@functools.wraps(fn)
|
37 |
def wrapper(*args, **kw):
|
38 |
try:
|
|
|
41 |
return f"ERROR: {e}"
|
42 |
return wrapper
|
43 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
# ---------------------------------------------------------------------
|
45 |
+
# Eigene Tools
|
46 |
# ---------------------------------------------------------------------
|
47 |
@tool
|
48 |
@error_guard
|
49 |
def parse_csv(file_path: str, query: str = "") -> str:
|
|
|
50 |
df = pd.read_csv(file_path)
|
51 |
if not query:
|
52 |
return f"Rows={len(df)}, Cols={list(df.columns)}"
|
53 |
+
return df.query(query).to_markdown(index=False)
|
|
|
|
|
|
|
|
|
54 |
|
55 |
@tool
|
56 |
@error_guard
|
57 |
def parse_excel(file_path: str, sheet: str | int | None = None, query: str = "") -> str:
|
|
|
58 |
sheet_arg = int(sheet) if isinstance(sheet, str) and sheet.isdigit() else sheet or 0
|
59 |
df = pd.read_excel(file_path, sheet_name=sheet_arg)
|
60 |
if not query:
|
61 |
return f"Rows={len(df)}, Cols={list(df.columns)}"
|
62 |
+
return df.query(query).to_markdown(index=False)
|
|
|
|
|
|
|
63 |
|
|
|
|
|
|
|
64 |
@tool
|
65 |
@error_guard
|
66 |
def web_search(query: str, max_results: int = 5) -> str:
|
|
|
67 |
api_key = os.getenv("TAVILY_API_KEY")
|
68 |
hits = TavilySearchResults(max_results=max_results, api_key=api_key).invoke(query)
|
69 |
if not hits:
|
70 |
return "No results."
|
71 |
return "\n".join(f"{h['title']} – {h['url']}" for h in hits)
|
72 |
|
|
|
73 |
@tool
|
74 |
@error_guard
|
75 |
def wiki_search(query: str, sentences: int = 3) -> str:
|
|
|
76 |
wrapper = WikipediaAPIWrapper(top_k_results=1, doc_content_chars_max=4000)
|
77 |
res = wrapper.run(query)
|
78 |
return "\n".join(res.split(". ")[:sentences]) if res else "No article found."
|
79 |
|
80 |
+
# Python Tool
|
|
|
|
|
81 |
python_repl = PythonAstREPLTool()
|
82 |
|
83 |
# ---------------------------------------------------------------------
|
84 |
+
# Gemini LLM
|
85 |
# ---------------------------------------------------------------------
|
86 |
gemini_llm = ChatGoogleGenerativeAI(
|
87 |
google_api_key=os.getenv("GOOGLE_API_KEY"),
|
|
|
90 |
max_output_tokens=2048,
|
91 |
)
|
92 |
|
|
|
|
|
|
|
93 |
SYSTEM_PROMPT = SystemMessage(
|
94 |
content=(
|
95 |
+
"You are a helpful assistant with access to tools.\n"
|
96 |
+
"Use tools when appropriate using tool calls.\n"
|
97 |
+
"If the answer is clear, return it directly without explanation."
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
98 |
)
|
99 |
)
|
100 |
|
101 |
+
TOOLS = [web_search, wiki_search, parse_csv, parse_excel, python_repl]
|
102 |
+
|
103 |
# ---------------------------------------------------------------------
|
104 |
+
# LangGraph Nodes
|
105 |
# ---------------------------------------------------------------------
|
|
|
106 |
def planner(state: MessagesState):
|
107 |
+
messages = state["messages"]
|
108 |
+
if not any(m.type == "system" for m in messages):
|
109 |
+
messages = [SYSTEM_PROMPT] + messages
|
110 |
+
resp = gemini_llm.invoke(messages)
|
111 |
+
return {"messages": messages + [resp]}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
112 |
|
113 |
+
def should_end(state: MessagesState) -> bool:
|
114 |
+
last = state["messages"][-1]
|
115 |
+
return not getattr(last, "tool_calls", None)
|
|
|
|
|
116 |
|
117 |
+
# ---------------------------------------------------------------------
|
118 |
+
# Build Graph
|
119 |
+
# ---------------------------------------------------------------------
|
120 |
graph = StateGraph(MessagesState)
|
|
|
121 |
graph.add_node("planner", planner)
|
122 |
graph.add_node("tools", ToolNode(TOOLS))
|
|
|
123 |
graph.add_edge(START, "planner")
|
124 |
+
graph.add_conditional_edges(
|
125 |
+
"planner",
|
126 |
+
lambda state: "END" if should_end(state) else "tools",
|
127 |
+
{"tools": "tools", "END": END},
|
128 |
+
)
|
129 |
+
graph.add_edge("tools", "planner")
|
130 |
|
|
|
131 |
agent_executor = graph.compile()
|
132 |
|
133 |
# ---------------------------------------------------------------------
|
134 |
+
# Öffentliche Klasse
|
135 |
# ---------------------------------------------------------------------
|
136 |
class GaiaAgent:
|
|
|
|
|
137 |
def __init__(self):
|
138 |
print("✅ GaiaAgent initialised (LangGraph)")
|
139 |
|
140 |
def __call__(self, task_id: str, question: str) -> str:
|
141 |
+
state = {"messages": [HumanMessage(content=question)]}
|
142 |
+
final = agent_executor.invoke(state)
|
143 |
+
return final["messages"][-1].content.strip()
|
|
|
|
|
|