kaggle_benchmarks 0.2.0__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- kaggle_benchmarks/__init__.py +44 -0
- kaggle_benchmarks/_config.py +129 -0
- kaggle_benchmarks/actors/__init__.py +16 -0
- kaggle_benchmarks/actors/base.py +75 -0
- kaggle_benchmarks/actors/llms.py +465 -0
- kaggle_benchmarks/assertions.py +399 -0
- kaggle_benchmarks/chats.py +146 -0
- kaggle_benchmarks/clients.py +81 -0
- kaggle_benchmarks/content_types/__init__.py +15 -0
- kaggle_benchmarks/content_types/images.py +135 -0
- kaggle_benchmarks/contexts.py +114 -0
- kaggle_benchmarks/envs/__init__.py +23 -0
- kaggle_benchmarks/envs/docker.py +109 -0
- kaggle_benchmarks/envs/environment.py +42 -0
- kaggle_benchmarks/envs/local.py +53 -0
- kaggle_benchmarks/envs/mixins.py +48 -0
- kaggle_benchmarks/events.py +51 -0
- kaggle_benchmarks/kaggle/__init__.py +27 -0
- kaggle_benchmarks/kaggle/benchmark_types_pb2.py +66 -0
- kaggle_benchmarks/kaggle/benchmark_types_pb2.pyi +1214 -0
- kaggle_benchmarks/kaggle/client.py +147 -0
- kaggle_benchmarks/kaggle/model_proxy.py +71 -0
- kaggle_benchmarks/kaggle/models.py +62 -0
- kaggle_benchmarks/kaggle/serialization.py +579 -0
- kaggle_benchmarks/messages.py +148 -0
- kaggle_benchmarks/orchestration/__init__.py +20 -0
- kaggle_benchmarks/orchestration/scheduling.py +105 -0
- kaggle_benchmarks/orchestration/task_queue.py +77 -0
- kaggle_benchmarks/prompting.py +173 -0
- kaggle_benchmarks/results.py +121 -0
- kaggle_benchmarks/runs.py +226 -0
- kaggle_benchmarks/tasks.py +343 -0
- kaggle_benchmarks/tools/__init__.py +15 -0
- kaggle_benchmarks/tools/python.py +199 -0
- kaggle_benchmarks/tools/search.py +138 -0
- kaggle_benchmarks/tools/web.py +126 -0
- kaggle_benchmarks/ui/__init__.py +15 -0
- kaggle_benchmarks/ui/console.py +42 -0
- kaggle_benchmarks/ui/ipython_magics.py +206 -0
- kaggle_benchmarks/ui/panel.py +389 -0
- kaggle_benchmarks/ui/setup_panel.py +92 -0
- kaggle_benchmarks/utils.py +188 -0
- kaggle_benchmarks-0.2.0.dist-info/METADATA +142 -0
- kaggle_benchmarks-0.2.0.dist-info/RECORD +46 -0
- kaggle_benchmarks-0.2.0.dist-info/WHEEL +4 -0
- kaggle_benchmarks-0.2.0.dist-info/licenses/LICENSE +202 -0
|
@@ -0,0 +1,44 @@
|
|
|
1
|
+
# Copyright 2025 Kaggle Inc.
|
|
2
|
+
#
|
|
3
|
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
4
|
+
# you may not use this file except in compliance with the License.
|
|
5
|
+
# You may obtain a copy of the License at
|
|
6
|
+
#
|
|
7
|
+
# http://www.apache.org/licenses/LICENSE-2.0
|
|
8
|
+
#
|
|
9
|
+
# Unless required by applicable law or agreed to in writing, software
|
|
10
|
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
11
|
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
12
|
+
# See the License for the specific language governing permissions and
|
|
13
|
+
# limitations under the License.
|
|
14
|
+
|
|
15
|
+
from kaggle_benchmarks import (
|
|
16
|
+
actors,
|
|
17
|
+
assertions,
|
|
18
|
+
chats,
|
|
19
|
+
clients,
|
|
20
|
+
content_types,
|
|
21
|
+
envs,
|
|
22
|
+
kaggle,
|
|
23
|
+
orchestration,
|
|
24
|
+
prompting,
|
|
25
|
+
tasks,
|
|
26
|
+
tools,
|
|
27
|
+
ui,
|
|
28
|
+
utils,
|
|
29
|
+
)
|
|
30
|
+
from kaggle_benchmarks._config import ExecutionMode, config
|
|
31
|
+
from kaggle_benchmarks.actors import Actor, LLMChat, system, user
|
|
32
|
+
from kaggle_benchmarks.runs import Run, Runs
|
|
33
|
+
from kaggle_benchmarks.tasks import benchmark, task
|
|
34
|
+
|
|
35
|
+
llm = kaggle.load_default_model()
|
|
36
|
+
judge_llm = kaggle.load_judge_model()
|
|
37
|
+
llms = kaggle.load_available_models()
|
|
38
|
+
|
|
39
|
+
|
|
40
|
+
client: clients.Client = clients.resolve_client()
|
|
41
|
+
config.apply()
|
|
42
|
+
|
|
43
|
+
|
|
44
|
+
__version__ = "0.2.0"
|
|
@@ -0,0 +1,129 @@
|
|
|
1
|
+
# Copyright 2025 Kaggle Inc.
|
|
2
|
+
#
|
|
3
|
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
4
|
+
# you may not use this file except in compliance with the License.
|
|
5
|
+
# You may obtain a copy of the License at
|
|
6
|
+
#
|
|
7
|
+
# http://www.apache.org/licenses/LICENSE-2.0
|
|
8
|
+
#
|
|
9
|
+
# Unless required by applicable law or agreed to in writing, software
|
|
10
|
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
11
|
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
12
|
+
# See the License for the specific language governing permissions and
|
|
13
|
+
# limitations under the License.
|
|
14
|
+
|
|
15
|
+
import dataclasses
|
|
16
|
+
import enum
|
|
17
|
+
import logging
|
|
18
|
+
import os
|
|
19
|
+
from pathlib import Path
|
|
20
|
+
from typing import Any, Self
|
|
21
|
+
|
|
22
|
+
import dotenv
|
|
23
|
+
import panel as pn
|
|
24
|
+
|
|
25
|
+
base_dir = Path(__file__).parent.parent.parent
|
|
26
|
+
dotenv.load_dotenv(override=True)
|
|
27
|
+
|
|
28
|
+
|
|
29
|
+
def string_to_bool(s: str) -> bool:
|
|
30
|
+
"""Converts a string to a boolean, handling various truthy values."""
|
|
31
|
+
return s.lower() in ("true", "1", "t", "y", "yes")
|
|
32
|
+
|
|
33
|
+
|
|
34
|
+
class ExecutionMode(enum.Enum):
|
|
35
|
+
NOTEBOOK = 0
|
|
36
|
+
INTERACTIVE = 0
|
|
37
|
+
RUN = 1
|
|
38
|
+
TRIAL = 1
|
|
39
|
+
TESTING = 2
|
|
40
|
+
DEV = 3
|
|
41
|
+
DOC = 4
|
|
42
|
+
|
|
43
|
+
|
|
44
|
+
@dataclasses.dataclass()
|
|
45
|
+
class Config:
|
|
46
|
+
execution_mode: ExecutionMode = dataclasses.field(
|
|
47
|
+
default_factory=lambda: ExecutionMode[os.getenv("BENCHMARK_MODE", "NOTEBOOK")]
|
|
48
|
+
)
|
|
49
|
+
|
|
50
|
+
disable_caching: bool = dataclasses.field(
|
|
51
|
+
default_factory=lambda: string_to_bool(
|
|
52
|
+
os.environ.get("DISABLE_LOCAL_CACHE", "True")
|
|
53
|
+
)
|
|
54
|
+
)
|
|
55
|
+
|
|
56
|
+
cache_directory: Path = dataclasses.field(
|
|
57
|
+
default_factory=lambda: Path(
|
|
58
|
+
os.environ.get("CACHE_DIRECTORY", base_dir / ".cache")
|
|
59
|
+
)
|
|
60
|
+
)
|
|
61
|
+
|
|
62
|
+
interactive_mode: bool = dataclasses.field(
|
|
63
|
+
default_factory=lambda: string_to_bool(
|
|
64
|
+
os.environ.get("INTERACTIVE_UI", "False")
|
|
65
|
+
)
|
|
66
|
+
)
|
|
67
|
+
ui_handler: Any = None
|
|
68
|
+
|
|
69
|
+
_ui_theme: str = dataclasses.field(
|
|
70
|
+
default_factory=lambda: os.environ.get("UI_THEME", "default")
|
|
71
|
+
)
|
|
72
|
+
suppress_ui_warnings: bool = dataclasses.field(
|
|
73
|
+
default_factory=lambda: string_to_bool(
|
|
74
|
+
os.environ.get("SUPPRESS_UI_WARNINGS", "true")
|
|
75
|
+
)
|
|
76
|
+
)
|
|
77
|
+
continue_with_exceptions: bool = dataclasses.field(
|
|
78
|
+
default_factory=lambda: (
|
|
79
|
+
os.environ.get("KAGGLE_KERNEL_RUN_TYPE", "").lower() == "batch"
|
|
80
|
+
or string_to_bool(os.environ.get("CONTINUE_WITH_EXCEPTIONS", "False"))
|
|
81
|
+
)
|
|
82
|
+
)
|
|
83
|
+
render_subruns: bool = dataclasses.field(
|
|
84
|
+
default_factory=lambda: (
|
|
85
|
+
string_to_bool(os.environ.get("RENDER_SUBRUNS", "True"))
|
|
86
|
+
)
|
|
87
|
+
)
|
|
88
|
+
|
|
89
|
+
show_message_details: bool = False
|
|
90
|
+
|
|
91
|
+
def apply(self) -> Self:
|
|
92
|
+
pn.config.theme = self.ui_theme # type: ignore
|
|
93
|
+
|
|
94
|
+
if self.suppress_ui_warnings:
|
|
95
|
+
logger = logging.getLogger("bokeh")
|
|
96
|
+
logger.setLevel(logging.ERROR)
|
|
97
|
+
if self.interactive_mode:
|
|
98
|
+
from kaggle_benchmarks import events, ui
|
|
99
|
+
|
|
100
|
+
if self.ui_handler is None:
|
|
101
|
+
self.ui_handler = ui.panel.PanelUI()
|
|
102
|
+
events.manager.bind(self.ui_handler)
|
|
103
|
+
|
|
104
|
+
return self
|
|
105
|
+
|
|
106
|
+
@property
|
|
107
|
+
def ui_theme(self):
|
|
108
|
+
return self._ui_theme
|
|
109
|
+
|
|
110
|
+
@ui_theme.setter
|
|
111
|
+
def ui_theme(self, value):
|
|
112
|
+
self._ui_theme = value
|
|
113
|
+
self.apply()
|
|
114
|
+
|
|
115
|
+
def enable_dark_mode(self):
|
|
116
|
+
self.ui_theme = "dark"
|
|
117
|
+
|
|
118
|
+
def disable_dark_mode(self):
|
|
119
|
+
self.ui_theme = "default"
|
|
120
|
+
|
|
121
|
+
def enable_interactive_mode(self):
|
|
122
|
+
self.interactive_mode = True
|
|
123
|
+
self.apply()
|
|
124
|
+
|
|
125
|
+
def disable_tqdm(self):
|
|
126
|
+
return self.execution_mode != ExecutionMode.NOTEBOOK
|
|
127
|
+
|
|
128
|
+
|
|
129
|
+
config = Config()
|
|
@@ -0,0 +1,16 @@
|
|
|
1
|
+
# Copyright 2025 Kaggle Inc.
|
|
2
|
+
#
|
|
3
|
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
4
|
+
# you may not use this file except in compliance with the License.
|
|
5
|
+
# You may obtain a copy of the License at
|
|
6
|
+
#
|
|
7
|
+
# http://www.apache.org/licenses/LICENSE-2.0
|
|
8
|
+
#
|
|
9
|
+
# Unless required by applicable law or agreed to in writing, software
|
|
10
|
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
11
|
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
12
|
+
# See the License for the specific language governing permissions and
|
|
13
|
+
# limitations under the License.
|
|
14
|
+
|
|
15
|
+
from kaggle_benchmarks.actors.base import Actor, assertion, system, user
|
|
16
|
+
from kaggle_benchmarks.actors.llms import LLMChat
|
|
@@ -0,0 +1,75 @@
|
|
|
1
|
+
# Copyright 2025 Kaggle Inc.
|
|
2
|
+
#
|
|
3
|
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
4
|
+
# you may not use this file except in compliance with the License.
|
|
5
|
+
# You may obtain a copy of the License at
|
|
6
|
+
#
|
|
7
|
+
# http://www.apache.org/licenses/LICENSE-2.0
|
|
8
|
+
#
|
|
9
|
+
# Unless required by applicable law or agreed to in writing, software
|
|
10
|
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
11
|
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
12
|
+
# See the License for the specific language governing permissions and
|
|
13
|
+
# limitations under the License.
|
|
14
|
+
|
|
15
|
+
from typing import Iterable, Literal, TypeVar
|
|
16
|
+
|
|
17
|
+
from kaggle_benchmarks import utils
|
|
18
|
+
from kaggle_benchmarks.messages import Message
|
|
19
|
+
|
|
20
|
+
T = TypeVar("T")
|
|
21
|
+
|
|
22
|
+
|
|
23
|
+
class Actor:
|
|
24
|
+
def __init__(
|
|
25
|
+
self,
|
|
26
|
+
name: str | None = None,
|
|
27
|
+
role: Literal["system", "user", "assistant", "developer", "tool"] = "assistant",
|
|
28
|
+
avatar: str = "",
|
|
29
|
+
id: str | None = None,
|
|
30
|
+
):
|
|
31
|
+
self.name = type(self).__name__ if name is None else name
|
|
32
|
+
self.id = id or name
|
|
33
|
+
self.role = role
|
|
34
|
+
self.avatar = avatar
|
|
35
|
+
|
|
36
|
+
def send(
|
|
37
|
+
self, message: T | Message[T], is_visible_to_llm: bool = True
|
|
38
|
+
) -> Message[T]:
|
|
39
|
+
return self._send(message, is_visible_to_llm)
|
|
40
|
+
|
|
41
|
+
def _send(self, message: T | Message[T], is_visible_to_llm: bool) -> Message[T]:
|
|
42
|
+
from kaggle_benchmarks import chats
|
|
43
|
+
|
|
44
|
+
if isinstance(message, Message):
|
|
45
|
+
message.is_visible_to_llm = is_visible_to_llm
|
|
46
|
+
else:
|
|
47
|
+
message = Message(
|
|
48
|
+
sender=self, content=message, is_visible_to_llm=is_visible_to_llm
|
|
49
|
+
)
|
|
50
|
+
|
|
51
|
+
chats.send(message)
|
|
52
|
+
return message
|
|
53
|
+
|
|
54
|
+
def stream(self, content: Iterable[str]) -> Message[str]:
|
|
55
|
+
msg = Message(content="", sender=self, _status=utils.Status.RUNNING)
|
|
56
|
+
self.send(msg)
|
|
57
|
+
msg.stream(content)
|
|
58
|
+
msg.status = utils.Status.SUCCESS
|
|
59
|
+
return msg
|
|
60
|
+
|
|
61
|
+
def as_dict(self) -> dict[str, str]:
|
|
62
|
+
"""Returns a dictionary representation of the agent."""
|
|
63
|
+
return dict(id=self.id, name=self.name, role=self.role, avatar=self.avatar)
|
|
64
|
+
|
|
65
|
+
def __repr__(self) -> str:
|
|
66
|
+
cls = type(self)
|
|
67
|
+
return f"{cls.__name__}(name={self.name!r}, avatar={self.avatar!r})"
|
|
68
|
+
|
|
69
|
+
def __str__(self) -> str:
|
|
70
|
+
return f"{self.avatar} {self.name}"
|
|
71
|
+
|
|
72
|
+
|
|
73
|
+
system = Actor(name="System", role="system", avatar="⚙️")
|
|
74
|
+
assertion = Actor(name="Assertion", role="system", avatar="🚨️")
|
|
75
|
+
user = Actor(name="User", role="user", avatar="👤")
|