SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Agents planning (#31702)

Browse Source 
* Allow planning for agents
This commit is contained in:
Aymeric Roucher
2024-07-22 10:49:57 +02:00
committed by GitHub
parent 0fdea8607d
commit b381880597
6 changed files with 770 additions and 273 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tests/agents/test_agents.py
View File

@@ -223,7 +223,7 @@ Action:
# check that add_base_tools will not interfere with existing tools
with pytest.raises(KeyError) as e:
agent = ReactJsonAgent(tools=toolset_3, llm_engine=fake_react_json_llm, add_base_tools=True)
assert "python_interpreter already exists in the toolbox" in str(e)
assert "already exists in the toolbox" in str(e)
# check that python_interpreter base tool does not get added to code agents
agent = ReactCodeAgent(tools=[], llm_engine=fake_react_code_llm, add_base_tools=True)

161
tests/agents/test_python_interpreter.py
View File

@@ -15,6 +15,7 @@
import unittest
import numpy as np
import pytest
from transformers import load_tool
@@ -241,8 +242,41 @@ for block in text_block:
code = """
digits, i = [1, 2, 3], 1
digits[i], digits[i + 1] = digits[i + 1], digits[i]"""
evaluate_python_code(code, {"range": range, "print": print, "int": int}, {})
code = """
def calculate_isbn_10_check_digit(number):
total = sum((10 - i) * int(digit) for i, digit in enumerate(number))
remainder = total % 11
check_digit = 11 - remainder
if check_digit == 10:
return 'X'
elif check_digit == 11:
return '0'
else:
return str(check_digit)
# Given 9-digit numbers
numbers = [
"478225952",
"643485613",
"739394228",
"291726859",
"875262394",
"542617795",
"031810713",
"957007669",
"871467426"
]
# Calculate check digits for each number
check_digits = [calculate_isbn_10_check_digit(number) for number in numbers]
print(check_digits)
"""
state = {}
evaluate_python_code(code, {"range": range, "print": print, "int": int}, state)
evaluate_python_code(
code, {"range": range, "print": print, "sum": sum, "enumerate": enumerate, "int": int, "str": str}, state
)
def test_listcomp(self):
code = "x = [i for i in range(3)]"
@@ -273,6 +307,17 @@ digits[i], digits[i + 1] = digits[i + 1], digits[i]"""
result = evaluate_python_code(code, {"range": range}, state={})
assert result == {0: 0, 1: 1, 2: 4}
code = "{num: name for num, name in {101: 'a', 102: 'b'}.items() if name not in ['a']}"
result = evaluate_python_code(code, {"print": print}, state={}, authorized_imports=["pandas"])
assert result == {102: "b"}
code = """
shifts = {'A': ('6:45', '8:00'), 'B': ('10:00', '11:45')}
shift_minutes = {worker: ('a', 'b') for worker, (start, end) in shifts.items()}
"""
result = evaluate_python_code(code, {}, state={})
assert result == {"A": ("a", "b"), "B": ("a", "b")}
def test_tuple_assignment(self):
code = "a, b = 0, 1\nb"
result = evaluate_python_code(code, BASE_PYTHON_TOOLS, state={})
@@ -341,7 +386,7 @@ if char.isalpha():
result = evaluate_python_code(code, BASE_PYTHON_TOOLS, state={})
assert result == "lose"
code = "import time\ntime.sleep(0.1)"
code = "import time, re\ntime.sleep(0.1)"
result = evaluate_python_code(code, BASE_PYTHON_TOOLS, state={})
assert result is None
@@ -369,6 +414,23 @@ if char.isalpha():
result = evaluate_python_code(code, BASE_PYTHON_TOOLS, state={})
assert result == "LATIN CAPITAL LETTER A"
# Test submodules are handled properly, thus not raising error
code = "import numpy.random as rd\nrng = rd.default_rng(12345)\nrng.random()"
result = evaluate_python_code(code, BASE_PYTHON_TOOLS, state={}, authorized_imports=["numpy"])
code = "from numpy.random import default_rng as d_rng\nrng = d_rng(12345)\nrng.random()"
result = evaluate_python_code(code, BASE_PYTHON_TOOLS, state={}, authorized_imports=["numpy"])
def test_additional_imports(self):
code = "import numpy as np"
evaluate_python_code(code, authorized_imports=["numpy"], state={})
code = "import numpy.random as rd"
evaluate_python_code(code, authorized_imports=["numpy.random"], state={})
evaluate_python_code(code, authorized_imports=["numpy"], state={})
with pytest.raises(InterpreterError):
evaluate_python_code(code, authorized_imports=["random"], state={})
def test_multiple_comparators(self):
code = "0 <= -1 < 4 and 0 <= -5 < 4"
result = evaluate_python_code(code, BASE_PYTHON_TOOLS, state={})
@@ -400,7 +462,7 @@ def function():
print("2")
function()"""
state = {}
evaluate_python_code(code, {"print": print}, state)
evaluate_python_code(code, {"print": print}, state=state)
assert state["print_outputs"] == "1\n2\n"
def test_tuple_target_in_iterator(self):
@@ -612,7 +674,7 @@ assert lock.locked == False
"""
state = {}
tools = {}
evaluate_python_code(code, tools, state)
evaluate_python_code(code, tools, state=state)
def test_default_arg_in_function(self):
code = """
@@ -672,3 +734,94 @@ returns_none(1)
state = {}
result = evaluate_python_code(code, {"print": print, "range": range, "ord": ord, "chr": chr}, state=state)
assert result is None
def test_nested_for_loop(self):
code = """
all_res = []
for i in range(10):
subres = []
for j in range(i):
subres.append(j)
all_res.append(subres)
out = [i for sublist in all_res for i in sublist]
out[:10]
"""
state = {}
result = evaluate_python_code(code, {"print": print, "range": range}, state=state)
assert result == [0, 0, 1, 0, 1, 2, 0, 1, 2, 3]
def test_pandas(self):
code = """
import pandas as pd
df = pd.DataFrame.from_dict({'SetCount': ['5', '4', '5'], 'Quantity': [1, 0, -1]})
df['SetCount'] = pd.to_numeric(df['SetCount'], errors='coerce')
parts_with_5_set_count = df[df['SetCount'] == 5.0]
parts_with_5_set_count[['Quantity', 'SetCount']].values[1]
"""
state = {}
result = evaluate_python_code(code, {}, state=state, authorized_imports=["pandas"])
assert np.array_equal(result, [-1, 5])
code = """
import pandas as pd
df = pd.DataFrame.from_dict({"AtomicNumber": [111, 104, 105], "ok": [0, 1, 2]})
print("HH0")
# Filter the DataFrame to get only the rows with outdated atomic numbers
filtered_df = df.loc[df['AtomicNumber'].isin([104])]
"""
result = evaluate_python_code(code, {"print": print}, state={}, authorized_imports=["pandas"])
assert np.array_equal(result.values[0], [104, 1])
code = """import pandas as pd
data = pd.DataFrame.from_dict([
{"Pclass": 1, "Survived": 1},
{"Pclass": 2, "Survived": 0},
{"Pclass": 2, "Survived": 1}
])
survival_rate_by_class = data.groupby('Pclass')['Survived'].mean()
"""
result = evaluate_python_code(code, {}, state={}, authorized_imports=["pandas"])
assert result.values[1] == 0.5
def test_starred(self):
code = """
from math import radians, sin, cos, sqrt, atan2
def haversine(lat1, lon1, lat2, lon2):
R = 6371000 # Radius of the Earth in meters
lat1, lon1, lat2, lon2 = map(radians, [lat1, lon1, lat2, lon2])
dlat = lat2 - lat1
dlon = lon2 - lon1
a = sin(dlat / 2) ** 2 + cos(lat1) * cos(lat2) * sin(dlon / 2) ** 2
c = 2 * atan2(sqrt(a), sqrt(1 - a))
distance = R * c
return distance
coords_geneva = (46.1978, 6.1342)
coords_barcelona = (41.3869, 2.1660)
distance_geneva_barcelona = haversine(*coords_geneva, *coords_barcelona)
"""
result = evaluate_python_code(code, {"print": print, "map": map}, state={}, authorized_imports=["math"])
assert round(result, 1) == 622395.4
def test_for(self):
code = """
shifts = {
"Worker A": ("6:45 pm", "8:00 pm"),
"Worker B": ("10:00 am", "11:45 am")
}
shift_intervals = {}
for worker, (start, end) in shifts.items():
shift_intervals[worker] = end
shift_intervals
"""
result = evaluate_python_code(code, {"print": print, "map": map}, state={})
assert result == {"Worker A": "8:00 pm", "Worker B": "11:45 am"}