304 lines
11 KiB
Python
304 lines
11 KiB
Python
import glob
|
|
import math
|
|
import os
|
|
import subprocess
|
|
import sys
|
|
from abc import ABC
|
|
from pathlib import Path
|
|
from typing import Any, Dict, List
|
|
|
|
import openai
|
|
import pytest
|
|
|
|
from agbenchmark.agent_api_interface import run_api_agent
|
|
from agbenchmark.start_benchmark import OPTIONAL_CATEGORIES
|
|
from agbenchmark.utils.data_types import ChallengeData, Ground
|
|
from agbenchmark.utils.prompts import (
|
|
END_PROMPT,
|
|
FEW_SHOT_EXAMPLES,
|
|
PROMPT_MAP,
|
|
SCORING_MAP,
|
|
)
|
|
from agbenchmark.utils.utils import agent_eligibible_for_optional_categories
|
|
|
|
|
|
class Challenge(ABC):
|
|
"""The parent class to all specific challenges classes.
|
|
Defines helper methods for running a challenge"""
|
|
|
|
_data_cache: Dict[str, ChallengeData] = {}
|
|
CHALLENGE_LOCATION: str = ""
|
|
ARTIFACTS_LOCATION: str = "" # this is for suites
|
|
scores: dict[str, Any] = {} # this is for suites
|
|
|
|
@property
|
|
def data(self) -> ChallengeData:
|
|
if self.CHALLENGE_LOCATION not in self._data_cache:
|
|
self._data_cache[self.CHALLENGE_LOCATION] = ChallengeData.deserialize(
|
|
self.CHALLENGE_LOCATION
|
|
)
|
|
return self._data_cache[self.CHALLENGE_LOCATION]
|
|
|
|
@property
|
|
def task(self) -> str:
|
|
return self.data.task
|
|
|
|
@property
|
|
def dependencies(self) -> list:
|
|
return self.data.dependencies
|
|
|
|
async def setup_challenge(self, config: Dict[str, Any], cutoff: int) -> None:
|
|
from agbenchmark.agent_interface import copy_artifacts_into_workspace, run_agent
|
|
|
|
artifact_paths = [
|
|
self.ARTIFACTS_LOCATION,
|
|
str(Path(self.CHALLENGE_LOCATION).parent),
|
|
]
|
|
|
|
for path in artifact_paths:
|
|
copy_artifacts_into_workspace(config["workspace"], "artifacts_in", path)
|
|
|
|
if not self.task:
|
|
return
|
|
|
|
print(
|
|
f"\033[1;35m============Starting {self.data.name} challenge============\033[0m"
|
|
)
|
|
print(f"\033[1;30mTask: {self.task}\033[0m")
|
|
|
|
if "--mock" in sys.argv:
|
|
print("Running mock agent")
|
|
for path in artifact_paths:
|
|
copy_artifacts_into_workspace(
|
|
config["workspace"], "artifacts_out", path
|
|
)
|
|
elif config.get("api_mode"):
|
|
await run_api_agent(self.data, config, self.ARTIFACTS_LOCATION, cutoff)
|
|
else:
|
|
run_agent(self.task, cutoff)
|
|
|
|
# hidden files are added after the agent runs. Hidden files can be python test files.
|
|
# We copy them in the workspace to make it easy to import the code produced by the agent
|
|
|
|
for path in artifact_paths:
|
|
copy_artifacts_into_workspace(config["workspace"], "custom_python", path)
|
|
|
|
def test_method(self, config: Dict[str, Any]) -> None:
|
|
raise NotImplementedError
|
|
|
|
@staticmethod
|
|
def open_file(workspace: str, filename: str) -> str:
|
|
script_dir = workspace
|
|
workspace_dir = os.path.join(script_dir, filename)
|
|
with open(workspace_dir, "r") as f:
|
|
return f.read()
|
|
|
|
def get_artifacts_out(
|
|
self, workspace: str | dict[str, str], ground: Ground
|
|
) -> List[str]:
|
|
if isinstance(workspace, dict):
|
|
workspace = workspace["output"]
|
|
|
|
script_dir = workspace
|
|
files_contents = []
|
|
|
|
for file_pattern in ground.files:
|
|
# Check if it is a file extension
|
|
if file_pattern.startswith("."):
|
|
# Find all files with the given extension in the workspace
|
|
matching_files = glob.glob(os.path.join(script_dir, "*" + file_pattern))
|
|
else:
|
|
# Otherwise, it is a specific file
|
|
matching_files = [os.path.join(script_dir, file_pattern)]
|
|
|
|
for file_path in matching_files:
|
|
if ground.eval.type == "python":
|
|
result = subprocess.run(
|
|
[sys.executable, file_path],
|
|
cwd=os.path.abspath(workspace),
|
|
capture_output=True,
|
|
text=True,
|
|
)
|
|
if "error" in result.stderr or result.returncode != 0:
|
|
print(result.stderr)
|
|
assert False, result.stderr
|
|
files_contents.append(f"Output: {result.stdout}\n")
|
|
else:
|
|
with open(file_path, "r") as f:
|
|
files_contents.append(f.read())
|
|
|
|
return files_contents
|
|
|
|
@staticmethod
|
|
def write_to_file(workspace: str, filename: str, content: str) -> None:
|
|
script_dir = workspace
|
|
print("Writing file at", script_dir)
|
|
workspace_dir = os.path.join(script_dir, filename)
|
|
|
|
# Open the file in write mode.
|
|
with open(workspace_dir, "w") as f:
|
|
# Write the content to the file.
|
|
f.write(content)
|
|
|
|
def get_filenames_in_workspace(self, workspace: str) -> List[str]:
|
|
return [
|
|
filename
|
|
for filename in os.listdir(workspace)
|
|
if os.path.isfile(os.path.join(workspace, filename))
|
|
]
|
|
|
|
def scoring(self, config: Dict[str, Any], content: str, ground: Ground) -> float:
|
|
print("\033[1;34mScoring content:\033[0m", content)
|
|
if ground.should_contain:
|
|
for should_contain_word in ground.should_contain:
|
|
print_content = (
|
|
f"\033[1;34mWord that should exist\033[0m - {should_contain_word}:"
|
|
)
|
|
if should_contain_word not in content:
|
|
print(print_content, "False")
|
|
return 0.0
|
|
else:
|
|
print(print_content, "True")
|
|
|
|
if ground.should_not_contain:
|
|
for should_not_contain_word in ground.should_not_contain:
|
|
print_content = f"\033[1;34mWord that should not exist\033[0m - {should_not_contain_word}:"
|
|
if should_not_contain_word in content:
|
|
print(print_content, "False")
|
|
return 0.0
|
|
else:
|
|
print(print_content, "True")
|
|
|
|
return 1.0
|
|
|
|
def llm_eval(self, config: Dict[str, Any], content: str, ground: Ground) -> float:
|
|
openai.api_key = os.getenv("OPENAI_API_KEY")
|
|
if "--mock" in sys.argv:
|
|
return 1.0
|
|
|
|
# the validation for this is done in the Eval BaseModel
|
|
scoring = SCORING_MAP[ground.eval.scoring] # type: ignore
|
|
prompt = PROMPT_MAP[ground.eval.template].format(task=self.data.task, scoring=scoring, answer=ground.answer, response=content) # type: ignore
|
|
|
|
if ground.eval.examples:
|
|
prompt += FEW_SHOT_EXAMPLES.format(examples=ground.eval.examples)
|
|
|
|
prompt += END_PROMPT
|
|
|
|
answer = openai.ChatCompletion.create(
|
|
model="gpt-4",
|
|
messages=[
|
|
{"role": "system", "content": prompt},
|
|
],
|
|
)
|
|
|
|
return float(answer["choices"][0]["message"]["content"]) # type: ignore
|
|
|
|
def get_scores(self, config: Dict[str, Any]) -> dict[str, Any]:
|
|
scores = []
|
|
scores_dict: Any = {}
|
|
percentage = None
|
|
|
|
try:
|
|
if self.data.task == "" and "--mock" in sys.argv:
|
|
scores = [1.0]
|
|
elif isinstance(self.data.ground, Ground):
|
|
files_contents = self.get_artifacts_out(
|
|
config["workspace"], self.data.ground
|
|
)
|
|
|
|
for file_content in files_contents:
|
|
score = self.scoring(config, file_content, self.data.ground)
|
|
print("\033[1;32mYour score is:\033[0m", score)
|
|
scores.append(score)
|
|
|
|
if self.data.ground.eval.type == "llm":
|
|
llm_eval = self.llm_eval(
|
|
config, "\n".join(files_contents), self.data.ground
|
|
)
|
|
if self.data.ground.eval.scoring == "percentage":
|
|
scores.append(math.ceil(llm_eval / 100))
|
|
elif self.data.ground.eval.scoring == "scale":
|
|
scores.append(math.ceil(llm_eval / 10))
|
|
print("\033[1;32mYour score is:\033[0m", llm_eval)
|
|
|
|
scores.append(llm_eval)
|
|
elif isinstance(self.data.ground, dict):
|
|
# if it's a dict then we know its a combined suite
|
|
for ground_key in self.data.ground:
|
|
ground = self.data.ground[ground_key]
|
|
files_contents = self.get_artifacts_out(config["workspace"], ground)
|
|
|
|
for file_content in files_contents:
|
|
score = self.scoring(config, file_content, ground)
|
|
scores_dict.setdefault(ground_key, []).append(score)
|
|
print(
|
|
f"\033[1;35mScore for {ground_key}:\033[0m",
|
|
scores_dict[ground_key],
|
|
)
|
|
|
|
if ground.eval.type == "llm":
|
|
llm_eval = self.llm_eval(
|
|
config, "\n".join(files_contents), ground
|
|
)
|
|
|
|
if ground.eval.scoring == "percentage":
|
|
scores_dict[ground_key].append(math.ceil(llm_eval / 100))
|
|
elif ground.eval.scoring == "scale":
|
|
scores_dict[ground_key].append(math.ceil(llm_eval / 10))
|
|
scores_dict[ground_key].append(llm_eval)
|
|
|
|
# Count the number of times the value 1.0 appears in the dictionary
|
|
num_ones = sum(
|
|
1
|
|
for scores in scores_dict.values()
|
|
for score in scores
|
|
if score == 1.0
|
|
)
|
|
|
|
# Calculate the percentage
|
|
percentage = round((num_ones / len(scores_dict)) * 100, 2)
|
|
|
|
# Print the result in green
|
|
print(f"\033[1;92mPercentage of 1.0 scores:\033[0m {percentage}%")
|
|
|
|
# TODO: in an ideal world it only returns 1.0 if all of the tests pass but then the dependencies break.
|
|
# So for now we return 1.0 if there's any that pass
|
|
if percentage > 0:
|
|
scores.append(1.0)
|
|
if percentage != 100:
|
|
print(
|
|
"\033[1;93mWARNING:\033[0m Your agent did not pass all the tests in the suite."
|
|
)
|
|
except Exception as e:
|
|
print("Error getting scores", e)
|
|
|
|
scores_data = {
|
|
"values": scores,
|
|
"scores_obj": scores_dict,
|
|
"percentage": percentage,
|
|
}
|
|
|
|
self.scores[self.__class__.__name__] = scores_data
|
|
|
|
return scores_data
|
|
|
|
def get_dummy_scores(self, test_name: str, scores: dict[str, Any]) -> int | None:
|
|
return 1 # remove this once this works
|
|
if 1 in scores.get("scores_obj", {}).get(test_name, []):
|
|
return 1
|
|
|
|
return None
|
|
|
|
def skip_optional_categories(self, config: Dict[str, Any]) -> None:
|
|
challenge_category = self.data.category
|
|
categories = [
|
|
category
|
|
for category in OPTIONAL_CATEGORIES
|
|
if category in challenge_category
|
|
]
|
|
if not agent_eligibible_for_optional_categories(
|
|
categories, config.get("category", [])
|
|
):
|
|
pytest.skip("Agent is not eligible for this category")
|