langchain/templates/python-lint/python_lint/agent_executor.py

217 lines
5.8 KiB
Python
Raw Normal View History

import os
import re
import subprocess # nosec
import tempfile
from langchain.agents import AgentType, initialize_agent
from langchain.agents.tools import Tool
from langchain.prompts import ChatPromptTemplate
from langchain.pydantic_v1 import BaseModel, Field, ValidationError, validator
from langchain_community.chat_models import ChatOpenAI
from langchain_core.language_models import BaseLLM
from langchain_core.runnables import ConfigurableField, Runnable
def strip_python_markdown_tags(text: str) -> str:
pat = re.compile(r"```python\n(.*)```", re.DOTALL)
code = pat.match(text)
if code:
return code.group(1)
else:
return text
def format_black(filepath: str):
"""Format a file with black."""
subprocess.run( # nosec
f"black {filepath}",
stderr=subprocess.STDOUT,
text=True,
shell=True,
timeout=3,
check=False,
)
def format_ruff(filepath: str):
"""Run ruff format on a file."""
subprocess.run( # nosec
f"ruff check --fix {filepath}",
shell=True,
text=True,
timeout=3,
universal_newlines=True,
check=False,
)
subprocess.run( # nosec
f"ruff format {filepath}",
stderr=subprocess.STDOUT,
shell=True,
timeout=3,
text=True,
check=False,
)
def check_ruff(filepath: str):
"""Run ruff check on a file."""
subprocess.check_output( # nosec
f"ruff check {filepath}",
stderr=subprocess.STDOUT,
shell=True,
timeout=3,
text=True,
)
def check_mypy(filepath: str, strict: bool = True, follow_imports: str = "skip"):
"""Run mypy on a file."""
cmd = (
f"mypy {'--strict' if strict else ''} "
f"--follow-imports={follow_imports} {filepath}"
)
subprocess.check_output( # nosec
cmd,
stderr=subprocess.STDOUT,
shell=True,
text=True,
timeout=3,
)
class PythonCode(BaseModel):
code: str = Field(
description="Python code conforming to "
"ruff, black, and *strict* mypy standards.",
)
@validator("code")
@classmethod
def check_code(cls, v: str) -> str:
v = strip_python_markdown_tags(v).strip()
try:
with tempfile.NamedTemporaryFile(mode="w", delete=False) as temp_file:
temp_file.write(v)
temp_file_path = temp_file.name
try:
# format with black and ruff
format_black(temp_file_path)
format_ruff(temp_file_path)
except subprocess.CalledProcessError:
pass
# update `v` with formatted code
with open(temp_file_path, "r") as temp_file:
v = temp_file.read()
# check
complaints = dict(ruff=None, mypy=None)
try:
check_ruff(temp_file_path)
except subprocess.CalledProcessError as e:
complaints["ruff"] = e.output
try:
check_mypy(temp_file_path)
except subprocess.CalledProcessError as e:
complaints["mypy"] = e.output
# raise ValueError if ruff or mypy had complaints
if any(complaints.values()):
code_str = f"```{temp_file_path}\n{v}```"
error_messages = [
f"```{key}\n{value}```"
for key, value in complaints.items()
if value
]
raise ValueError("\n\n".join([code_str] + error_messages))
finally:
os.remove(temp_file_path)
return v
def check_code(code: str) -> str:
try:
code_obj = PythonCode(code=code)
return (
f"# LGTM\n"
f"# use the `submit` tool to submit this code:\n\n"
f"```python\n{code_obj.code}\n```"
)
except ValidationError as e:
return e.errors()[0]["msg"]
prompt = ChatPromptTemplate.from_messages(
[
(
"system",
"You are a world class Python coder who uses "
"black, ruff, and *strict* mypy for all of your code. "
"Provide complete, end-to-end Python code "
"to meet the user's description/requirements. "
"Always `check` your code. When you're done, "
"you must ALWAYS use the `submit` tool.",
),
(
"human",
": {input}",
),
],
)
check_code_tool = Tool.from_function(
check_code,
name="check-code",
description="Always check your code before submitting it!",
)
submit_code_tool = Tool.from_function(
strip_python_markdown_tags,
name="submit-code",
description="THIS TOOL is the most important. "
"use it to submit your code to the user who requested it... "
"but be sure to `check` it first!",
return_direct=True,
)
tools = [check_code_tool, submit_code_tool]
def get_agent_executor(
llm: BaseLLM,
agent_type: AgentType = AgentType.OPENAI_FUNCTIONS,
) -> Runnable:
_agent_executor = initialize_agent(
tools,
llm,
agent=agent_type,
verbose=True,
handle_parsing_errors=True,
prompt=prompt,
)
return _agent_executor | (lambda output: output["output"])
class Instruction(BaseModel):
__root__: str
agent_executor = (
get_agent_executor(ChatOpenAI(model_name="gpt-4-1106-preview", temperature=0.0))
.configurable_alternatives(
ConfigurableField("model_name"),
default_key="gpt4turbo",
gpt4=get_agent_executor(ChatOpenAI(model_name="gpt-4", temperature=0.0)),
gpt35t=get_agent_executor(
ChatOpenAI(model_name="gpt-3.5-turbo", temperature=0.0),
),
)
.with_types(input_type=Instruction, output_type=str)
)