kennethreitz/langchain
1
0
Fork 0
mirror of https://github.com/kennethreitz/langchain.git synced 2026-06-05 23:00:18 +00:00
Code Issues Packages Projects Releases Wiki Activity

Callbacks Refactor [base] (#3256)

Browse Source 
Co-authored-by: Nuno Campos <nuno@boringbits.io>
Co-authored-by: Davis Chase <130488702+dev2049@users.noreply.github.com>
Co-authored-by: Zander Chase <130414180+vowelparrot@users.noreply.github.com>
Co-authored-by: Harrison Chase <hw.chase.17@gmail.com>
This commit is contained in:
Ankush Gola
2023-04-30 11:14:09 -07:00
committed by GitHub
parent 18ec22fe56
commit d3ec00b566
208 changed files with 6394 additions and 3353 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/unit_tests/callbacks/fake_callback_handler.py
+226 -122
View File
@@ -1,10 +1,9 @@
"""A fake callback handler for testing purposes."""
from typing import Any, Dict, List, Union
from typing import Any
from pydantic import BaseModel
from langchain.callbacks.base import AsyncCallbackHandler, BaseCallbackHandler
from langchain.schema import AgentAction, AgentFinish, LLMResult
class BaseFakeCallbackHandler(BaseModel):
@@ -17,12 +16,72 @@ class BaseFakeCallbackHandler(BaseModel):
ignore_llm_: bool = False
ignore_chain_: bool = False
ignore_agent_: bool = False
always_verbose_: bool = False
@property
def always_verbose(self) -> bool:
"""Whether to call verbose callbacks even if verbose is False."""
return self.always_verbose_
# add finer-grained counters for easier debugging of failing tests
chain_starts: int = 0
chain_ends: int = 0
llm_starts: int = 0
llm_ends: int = 0
llm_streams: int = 0
tool_starts: int = 0
tool_ends: int = 0
agent_actions: int = 0
agent_ends: int = 0
class BaseFakeCallbackHandlerMixin(BaseFakeCallbackHandler):
"""Base fake callback handler mixin for testing."""
def on_llm_start_common(self) -> None:
self.llm_starts += 1
self.starts += 1
def on_llm_end_common(self) -> None:
self.llm_ends += 1
self.ends += 1
def on_llm_error_common(self) -> None:
self.errors += 1
def on_llm_new_token_common(self) -> None:
self.llm_streams += 1
def on_chain_start_common(self) -> None:
self.chain_starts += 1
self.starts += 1
def on_chain_end_common(self) -> None:
self.chain_ends += 1
self.ends += 1
def on_chain_error_common(self) -> None:
self.errors += 1
def on_tool_start_common(self) -> None:
self.tool_starts += 1
self.starts += 1
def on_tool_end_common(self) -> None:
self.tool_ends += 1
self.ends += 1
def on_tool_error_common(self) -> None:
self.errors += 1
def on_agent_action_common(self) -> None:
self.agent_actions += 1
self.starts += 1
def on_agent_finish_common(self) -> None:
self.agent_ends += 1
self.ends += 1
def on_text_common(self) -> None:
self.text += 1
class FakeCallbackHandler(BaseCallbackHandler, BaseFakeCallbackHandlerMixin):
"""Fake callback handler for testing."""
@property
def ignore_llm(self) -> bool:
@@ -39,164 +98,209 @@ class BaseFakeCallbackHandler(BaseModel):
"""Whether to ignore agent callbacks."""
return self.ignore_agent_
# add finer-grained counters for easier debugging of failing tests
chain_starts: int = 0
chain_ends: int = 0
llm_starts: int = 0
llm_ends: int = 0
llm_streams: int = 0
tool_starts: int = 0
tool_ends: int = 0
agent_ends: int = 0
class FakeCallbackHandler(BaseFakeCallbackHandler, BaseCallbackHandler):
"""Fake callback handler for testing."""
def on_llm_start(
self, serialized: Dict[str, Any], prompts: List[str], **kwargs: Any
) -> None:
"""Run when LLM starts running."""
self.llm_starts += 1
self.starts += 1
self,
*args: Any,
**kwargs: Any,
) -> Any:
self.on_llm_start_common()
def on_llm_new_token(self, token: str, **kwargs: Any) -> None:
"""Run when LLM generates a new token."""
self.llm_streams += 1
def on_llm_new_token(
self,
*args: Any,
**kwargs: Any,
) -> Any:
self.on_llm_new_token_common()
def on_llm_end(self, response: LLMResult, **kwargs: Any) -> None:
"""Run when LLM ends running."""
self.llm_ends += 1
self.ends += 1
def on_llm_end(
self,
*args: Any,
**kwargs: Any,
) -> Any:
self.on_llm_end_common()
def on_llm_error(
self, error: Union[Exception, KeyboardInterrupt], **kwargs: Any
) -> None:
"""Run when LLM errors."""
self.errors += 1
self,
*args: Any,
**kwargs: Any,
) -> Any:
self.on_llm_error_common()
def on_chain_start(
self, serialized: Dict[str, Any], inputs: Dict[str, Any], **kwargs: Any
) -> None:
"""Run when chain starts running."""
self.chain_starts += 1
self.starts += 1
self,
*args: Any,
**kwargs: Any,
) -> Any:
self.on_chain_start_common()
def on_chain_end(self, outputs: Dict[str, Any], **kwargs: Any) -> None:
"""Run when chain ends running."""
self.chain_ends += 1
self.ends += 1
def on_chain_end(
self,
*args: Any,
**kwargs: Any,
) -> Any:
self.on_chain_end_common()
def on_chain_error(
self, error: Union[Exception, KeyboardInterrupt], **kwargs: Any
) -> None:
"""Run when chain errors."""
self.errors += 1
self,
*args: Any,
**kwargs: Any,
) -> Any:
self.on_chain_error_common()
def on_tool_start(
self, serialized: Dict[str, Any], input_str: str, **kwargs: Any
) -> None:
"""Run when tool starts running."""
self.tool_starts += 1
self.starts += 1
self,
*args: Any,
**kwargs: Any,
) -> Any:
self.on_tool_start_common()
def on_tool_end(self, output: str, **kwargs: Any) -> None:
"""Run when tool ends running."""
self.tool_ends += 1
self.ends += 1
def on_tool_end(
self,
*args: Any,
**kwargs: Any,
) -> Any:
self.on_tool_end_common()
def on_tool_error(
self, error: Union[Exception, KeyboardInterrupt], **kwargs: Any
) -> None:
"""Run when tool errors."""
self.errors += 1
self,
*args: Any,
**kwargs: Any,
) -> Any:
self.on_tool_error_common()
def on_text(self, text: str, **kwargs: Any) -> None:
"""Run when agent is ending."""
self.text += 1
def on_agent_action(
self,
*args: Any,
**kwargs: Any,
) -> Any:
self.on_agent_action_common()
def on_agent_finish(self, finish: AgentFinish, **kwargs: Any) -> None:
"""Run when agent ends running."""
self.agent_ends += 1
self.ends += 1
def on_agent_finish(
self,
*args: Any,
**kwargs: Any,
) -> Any:
self.on_agent_finish_common()
def on_agent_action(self, action: AgentAction, **kwargs: Any) -> Any:
"""Run on agent action."""
self.tool_starts += 1
self.starts += 1
def on_text(
self,
*args: Any,
**kwargs: Any,
) -> Any:
self.on_text_common()
def __deepcopy__(self, memo: dict) -> "FakeCallbackHandler":
return self
class FakeAsyncCallbackHandler(BaseFakeCallbackHandler, AsyncCallbackHandler):
class FakeAsyncCallbackHandler(AsyncCallbackHandler, BaseFakeCallbackHandlerMixin):
"""Fake async callback handler for testing."""
@property
def ignore_llm(self) -> bool:
"""Whether to ignore LLM callbacks."""
return self.ignore_llm_
@property
def ignore_chain(self) -> bool:
"""Whether to ignore chain callbacks."""
return self.ignore_chain_
@property
def ignore_agent(self) -> bool:
"""Whether to ignore agent callbacks."""
return self.ignore_agent_
async def on_llm_start(
self, serialized: Dict[str, Any], prompts: List[str], **kwargs: Any
self,
*args: Any,
**kwargs: Any,
) -> None:
"""Run when LLM starts running."""
self.llm_starts += 1
self.starts += 1
self.on_llm_start_common()
async def on_llm_new_token(self, token: str, **kwargs: Any) -> None:
"""Run when LLM generates a new token."""
self.llm_streams += 1
async def on_llm_new_token(
self,
*args: Any,
**kwargs: Any,
) -> None:
self.on_llm_new_token_common()
async def on_llm_end(self, response: LLMResult, **kwargs: Any) -> None:
"""Run when LLM ends running."""
self.llm_ends += 1
self.ends += 1
async def on_llm_end(
self,
*args: Any,
**kwargs: Any,
) -> None:
self.on_llm_end_common()
async def on_llm_error(
self, error: Union[Exception, KeyboardInterrupt], **kwargs: Any
self,
*args: Any,
**kwargs: Any,
) -> None:
"""Run when LLM errors."""
self.errors += 1
self.on_llm_error_common()
async def on_chain_start(
self, serialized: Dict[str, Any], inputs: Dict[str, Any], **kwargs: Any
self,
*args: Any,
**kwargs: Any,
) -> None:
"""Run when chain starts running."""
self.chain_starts += 1
self.starts += 1
self.on_chain_start_common()
async def on_chain_end(self, outputs: Dict[str, Any], **kwargs: Any) -> None:
"""Run when chain ends running."""
self.chain_ends += 1
self.ends += 1
async def on_chain_end(
self,
*args: Any,
**kwargs: Any,
) -> None:
self.on_chain_end_common()
async def on_chain_error(
self, error: Union[Exception, KeyboardInterrupt], **kwargs: Any
self,
*args: Any,
**kwargs: Any,
) -> None:
"""Run when chain errors."""
self.errors += 1
self.on_chain_error_common()
async def on_tool_start(
self, serialized: Dict[str, Any], input_str: str, **kwargs: Any
self,
*args: Any,
**kwargs: Any,
) -> None:
"""Run when tool starts running."""
self.tool_starts += 1
self.starts += 1
self.on_tool_start_common()
async def on_tool_end(self, output: str, **kwargs: Any) -> None:
"""Run when tool ends running."""
self.tool_ends += 1
self.ends += 1
async def on_tool_end(
self,
*args: Any,
**kwargs: Any,
) -> None:
self.on_tool_end_common()
async def on_tool_error(
self, error: Union[Exception, KeyboardInterrupt], **kwargs: Any
self,
*args: Any,
**kwargs: Any,
) -> None:
"""Run when tool errors."""
self.errors += 1
self.on_tool_error_common()
async def on_text(self, text: str, **kwargs: Any) -> None:
"""Run when agent is ending."""
self.text += 1
async def on_agent_action(
self,
*args: Any,
**kwargs: Any,
) -> None:
self.on_agent_action_common()
async def on_agent_finish(self, finish: AgentFinish, **kwargs: Any) -> None:
"""Run when agent ends running."""
self.agent_ends += 1
self.ends += 1
async def on_agent_finish(
self,
*args: Any,
**kwargs: Any,
) -> None:
self.on_agent_finish_common()
async def on_agent_action(self, action: AgentAction, **kwargs: Any) -> None:
"""Run on agent action."""
self.tool_starts += 1
self.starts += 1
async def on_text(
self,
*args: Any,
**kwargs: Any,
) -> None:
self.on_text_common()
def __deepcopy__(self, memo: dict) -> "FakeAsyncCallbackHandler":
return self
tests/unit_tests/callbacks/test_callback_manager.py
+162 -99
View File
@@ -1,15 +1,12 @@
"""Test CallbackManager."""
from typing import Tuple
from typing import List, Tuple
import pytest
from langchain.callbacks.base import (
AsyncCallbackManager,
BaseCallbackManager,
CallbackManager,
)
from langchain.callbacks.shared import SharedCallbackManager
from langchain.schema import AgentFinish, LLMResult
from langchain.callbacks.base import BaseCallbackHandler
from langchain.callbacks.manager import AsyncCallbackManager, CallbackManager
from langchain.callbacks.stdout import StdOutCallbackHandler
from langchain.schema import AgentAction, AgentFinish, LLMResult
from tests.unit_tests.callbacks.fake_callback_handler import (
BaseFakeCallbackHandler,
FakeAsyncCallbackHandler,
@@ -18,19 +15,26 @@ from tests.unit_tests.callbacks.fake_callback_handler import (
def _test_callback_manager(
manager: BaseCallbackManager, *handlers: BaseFakeCallbackHandler
manager: CallbackManager, *handlers: BaseFakeCallbackHandler
) -> None:
"""Test the CallbackManager."""
manager.on_llm_start({}, [])
manager.on_llm_end(LLMResult(generations=[]))
manager.on_llm_error(Exception())
manager.on_chain_start({"name": "foo"}, {})
manager.on_chain_end({})
manager.on_chain_error(Exception())
manager.on_tool_start({}, "")
manager.on_tool_end("")
manager.on_tool_error(Exception())
manager.on_agent_finish(AgentFinish(log="", return_values={}))
run_manager = manager.on_llm_start({}, [])
run_manager.on_llm_end(LLMResult(generations=[]))
run_manager.on_llm_error(Exception())
run_manager.on_llm_new_token("foo")
run_manager.on_text("foo")
run_manager_chain = manager.on_chain_start({"name": "foo"}, {})
run_manager_chain.on_chain_end({})
run_manager_chain.on_chain_error(Exception())
run_manager_chain.on_agent_action(AgentAction(tool_input="foo", log="", tool=""))
run_manager_chain.on_agent_finish(AgentFinish(log="", return_values={}))
run_manager_chain.on_text("foo")
run_manager_tool = manager.on_tool_start({}, "")
run_manager_tool.on_tool_end("")
run_manager_tool.on_tool_error(Exception())
run_manager_tool.on_text("foo")
_check_num_calls(handlers)
@@ -38,75 +42,62 @@ async def _test_callback_manager_async(
manager: AsyncCallbackManager, *handlers: BaseFakeCallbackHandler
) -> None:
"""Test the CallbackManager."""
await manager.on_llm_start({}, [])
await manager.on_llm_end(LLMResult(generations=[]))
await manager.on_llm_error(Exception())
await manager.on_chain_start({"name": "foo"}, {})
await manager.on_chain_end({})
await manager.on_chain_error(Exception())
await manager.on_tool_start({}, "")
await manager.on_tool_end("")
await manager.on_tool_error(Exception())
await manager.on_agent_finish(AgentFinish(log="", return_values={}))
run_manager = await manager.on_llm_start({}, [])
await run_manager.on_llm_end(LLMResult(generations=[]))
await run_manager.on_llm_error(Exception())
await run_manager.on_llm_new_token("foo")
await run_manager.on_text("foo")
run_manager_chain = await manager.on_chain_start({"name": "foo"}, {})
await run_manager_chain.on_chain_end({})
await run_manager_chain.on_chain_error(Exception())
await run_manager_chain.on_agent_action(
AgentAction(tool_input="foo", log="", tool="")
)
await run_manager_chain.on_agent_finish(AgentFinish(log="", return_values={}))
await run_manager_chain.on_text("foo")
run_manager_tool = await manager.on_tool_start({}, "")
await run_manager_tool.on_tool_end("")
await run_manager_tool.on_tool_error(Exception())
await run_manager_tool.on_text("foo")
_check_num_calls(handlers)
def _check_num_calls(handlers: Tuple[BaseFakeCallbackHandler, ...]) -> None:
for handler in handlers:
if handler.always_verbose:
assert handler.starts == 3
assert handler.ends == 4
assert handler.errors == 3
else:
assert handler.starts == 0
assert handler.ends == 0
assert handler.errors == 0
def _test_callback_manager_pass_in_verbose(
manager: BaseCallbackManager, *handlers: FakeCallbackHandler
) -> None:
"""Test the CallbackManager."""
manager.on_llm_start({}, [], verbose=True)
manager.on_llm_end(LLMResult(generations=[]), verbose=True)
manager.on_llm_error(Exception(), verbose=True)
manager.on_chain_start({"name": "foo"}, {}, verbose=True)
manager.on_chain_end({}, verbose=True)
manager.on_chain_error(Exception(), verbose=True)
manager.on_tool_start({}, "", verbose=True)
manager.on_tool_end("", verbose=True)
manager.on_tool_error(Exception(), verbose=True)
manager.on_agent_finish(AgentFinish(log="", return_values={}), verbose=True)
for handler in handlers:
assert handler.starts == 3
assert handler.starts == 4
assert handler.ends == 4
assert handler.errors == 3
assert handler.text == 3
assert handler.llm_starts == 1
assert handler.llm_ends == 1
assert handler.llm_streams == 1
assert handler.chain_starts == 1
assert handler.chain_ends == 1
assert handler.tool_starts == 1
assert handler.tool_ends == 1
def test_callback_manager() -> None:
"""Test the CallbackManager."""
handler1 = FakeCallbackHandler(always_verbose_=True)
handler2 = FakeCallbackHandler(always_verbose_=False)
handler1 = FakeCallbackHandler()
handler2 = FakeCallbackHandler()
manager = CallbackManager([handler1, handler2])
_test_callback_manager(manager, handler1, handler2)
def test_callback_manager_pass_in_verbose() -> None:
"""Test the CallbackManager."""
handler1 = FakeCallbackHandler()
handler2 = FakeCallbackHandler()
manager = CallbackManager([handler1, handler2])
_test_callback_manager_pass_in_verbose(manager, handler1, handler2)
def test_ignore_llm() -> None:
"""Test ignore llm param for callback handlers."""
handler1 = FakeCallbackHandler(ignore_llm_=True, always_verbose_=True)
handler2 = FakeCallbackHandler(always_verbose_=True)
handler1 = FakeCallbackHandler(ignore_llm_=True)
handler2 = FakeCallbackHandler()
manager = CallbackManager(handlers=[handler1, handler2])
manager.on_llm_start({}, [], verbose=True)
manager.on_llm_end(LLMResult(generations=[]), verbose=True)
manager.on_llm_error(Exception(), verbose=True)
run_manager = manager.on_llm_start({}, [])
run_manager.on_llm_end(LLMResult(generations=[]))
run_manager.on_llm_error(Exception())
assert handler1.starts == 0
assert handler1.ends == 0
assert handler1.errors == 0
@@ -117,12 +108,12 @@ def test_ignore_llm() -> None:
def test_ignore_chain() -> None:
"""Test ignore chain param for callback handlers."""
handler1 = FakeCallbackHandler(ignore_chain_=True, always_verbose_=True)
handler2 = FakeCallbackHandler(always_verbose_=True)
handler1 = FakeCallbackHandler(ignore_chain_=True)
handler2 = FakeCallbackHandler()
manager = CallbackManager(handlers=[handler1, handler2])
manager.on_chain_start({"name": "foo"}, {}, verbose=True)
manager.on_chain_end({}, verbose=True)
manager.on_chain_error(Exception(), verbose=True)
run_manager = manager.on_chain_start({"name": "foo"}, {})
run_manager.on_chain_end({})
run_manager.on_chain_error(Exception())
assert handler1.starts == 0
assert handler1.ends == 0
assert handler1.errors == 0
@@ -133,39 +124,24 @@ def test_ignore_chain() -> None:
def test_ignore_agent() -> None:
"""Test ignore agent param for callback handlers."""
handler1 = FakeCallbackHandler(ignore_agent_=True, always_verbose_=True)
handler2 = FakeCallbackHandler(always_verbose_=True)
handler1 = FakeCallbackHandler(ignore_agent_=True)
handler2 = FakeCallbackHandler()
manager = CallbackManager(handlers=[handler1, handler2])
manager.on_tool_start({}, "", verbose=True)
manager.on_tool_end("", verbose=True)
manager.on_tool_error(Exception(), verbose=True)
manager.on_agent_finish(AgentFinish({}, ""), verbose=True)
run_manager = manager.on_tool_start({}, "")
run_manager.on_tool_end("")
run_manager.on_tool_error(Exception())
assert handler1.starts == 0
assert handler1.ends == 0
assert handler1.errors == 0
assert handler2.starts == 1
assert handler2.ends == 2
assert handler2.ends == 1
assert handler2.errors == 1
def test_shared_callback_manager() -> None:
"""Test the SharedCallbackManager."""
manager1 = SharedCallbackManager()
manager2 = SharedCallbackManager()
assert manager1 is manager2
handler1 = FakeCallbackHandler(always_verbose_=True)
handler2 = FakeCallbackHandler()
manager1.add_handler(handler1)
manager2.add_handler(handler2)
_test_callback_manager(manager1, handler1, handler2)
@pytest.mark.asyncio
async def test_async_callback_manager() -> None:
"""Test the AsyncCallbackManager."""
handler1 = FakeAsyncCallbackHandler(always_verbose_=True)
handler1 = FakeAsyncCallbackHandler()
handler2 = FakeAsyncCallbackHandler()
manager = AsyncCallbackManager([handler1, handler2])
await _test_callback_manager_async(manager, handler1, handler2)
@@ -174,8 +150,95 @@ async def test_async_callback_manager() -> None:
@pytest.mark.asyncio
async def test_async_callback_manager_sync_handler() -> None:
"""Test the AsyncCallbackManager."""
handler1 = FakeCallbackHandler(always_verbose_=True)
handler1 = FakeCallbackHandler()
handler2 = FakeAsyncCallbackHandler()
handler3 = FakeAsyncCallbackHandler(always_verbose_=True)
handler3 = FakeAsyncCallbackHandler()
manager = AsyncCallbackManager([handler1, handler2, handler3])
await _test_callback_manager_async(manager, handler1, handler2, handler3)
def test_callback_manager_inheritance() -> None:
handler1, handler2, handler3, handler4 = (
FakeCallbackHandler(),
FakeCallbackHandler(),
FakeCallbackHandler(),
FakeCallbackHandler(),
)
callback_manager1 = CallbackManager([handler1, handler2])
assert callback_manager1.handlers == [handler1, handler2]
assert callback_manager1.inheritable_handlers == []
callback_manager2 = CallbackManager([])
assert callback_manager2.handlers == []
assert callback_manager2.inheritable_handlers == []
callback_manager2.set_handlers([handler1, handler2])
assert callback_manager2.handlers == [handler1, handler2]
assert callback_manager2.inheritable_handlers == [handler1, handler2]
callback_manager2.set_handlers([handler3, handler4], inherit=False)
assert callback_manager2.handlers == [handler3, handler4]
assert callback_manager2.inheritable_handlers == []
callback_manager2.add_handler(handler1)
assert callback_manager2.handlers == [handler3, handler4, handler1]
assert callback_manager2.inheritable_handlers == [handler1]
callback_manager2.add_handler(handler2, inherit=False)
assert callback_manager2.handlers == [handler3, handler4, handler1, handler2]
assert callback_manager2.inheritable_handlers == [handler1]
run_manager = callback_manager2.on_chain_start({"name": "foo"}, {})
child_manager = run_manager.get_child()
assert child_manager.handlers == [handler1]
assert child_manager.inheritable_handlers == [handler1]
run_manager_tool = child_manager.on_tool_start({}, "")
assert run_manager_tool.handlers == [handler1]
assert run_manager_tool.inheritable_handlers == [handler1]
child_manager2 = run_manager_tool.get_child()
assert child_manager2.handlers == [handler1]
assert child_manager2.inheritable_handlers == [handler1]
def test_callback_manager_configure() -> None:
"""Test callback manager configuration."""
handler1, handler2, handler3, handler4 = (
FakeCallbackHandler(),
FakeCallbackHandler(),
FakeCallbackHandler(),
FakeCallbackHandler(),
)
inheritable_callbacks: List[BaseCallbackHandler] = [handler1, handler2]
local_callbacks: List[BaseCallbackHandler] = [handler3, handler4]
configured_manager = CallbackManager.configure(
inheritable_callbacks=inheritable_callbacks,
local_callbacks=local_callbacks,
verbose=True,
)
assert len(configured_manager.handlers) == 5
assert len(configured_manager.inheritable_handlers) == 2
assert configured_manager.inheritable_handlers == inheritable_callbacks
assert configured_manager.handlers[:4] == inheritable_callbacks + local_callbacks
assert isinstance(configured_manager.handlers[4], StdOutCallbackHandler)
assert isinstance(configured_manager, CallbackManager)
async_local_callbacks = AsyncCallbackManager([handler3, handler4])
async_configured_manager = AsyncCallbackManager.configure(
inheritable_callbacks=inheritable_callbacks,
local_callbacks=async_local_callbacks,
verbose=False,
)
assert len(async_configured_manager.handlers) == 4
assert len(async_configured_manager.inheritable_handlers) == 2
assert async_configured_manager.inheritable_handlers == inheritable_callbacks
assert async_configured_manager.handlers == inheritable_callbacks + [
handler3,
handler4,
]
assert isinstance(async_configured_manager, AsyncCallbackManager)
tests/unit_tests/callbacks/tracers/test_tracer.py
+175 -222
View File
@@ -1,9 +1,9 @@
"""Test Tracer classes."""
from __future__ import annotations
import threading
from datetime import datetime
from typing import List, Optional, Union
from typing import List, Union
from uuid import uuid4
import pytest
from freezegun import freeze_time
@@ -12,9 +12,7 @@ from langchain.callbacks.tracers.base import (
BaseTracer,
ChainRun,
LLMRun,
SharedTracer,
ToolRun,
Tracer,
TracerException,
TracerSession,
)
@@ -24,88 +22,6 @@ from langchain.schema import LLMResult
TEST_SESSION_ID = 2023
@freeze_time("2023-01-01")
def _get_compare_run() -> Union[LLMRun, ChainRun, ToolRun]:
return ChainRun(
id=None,
error=None,
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=1,
serialized={},
inputs={},
outputs={},
session_id=TEST_SESSION_ID,
child_runs=[
ToolRun(
id=None,
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=2,
serialized={},
tool_input="test",
output="test",
action="{}",
session_id=TEST_SESSION_ID,
error=None,
child_runs=[
LLMRun(
id=None,
error=None,
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=3,
serialized={},
prompts=[],
response=LLMResult(generations=[[]]),
session_id=TEST_SESSION_ID,
)
],
),
LLMRun(
id=None,
error=None,
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=4,
serialized={},
prompts=[],
response=LLMResult(generations=[[]]),
session_id=TEST_SESSION_ID,
),
],
)
def _perform_nested_run(tracer: BaseTracer) -> None:
"""Perform a nested run."""
tracer.on_chain_start(serialized={}, inputs={})
tracer.on_tool_start(serialized={}, input_str="test")
tracer.on_llm_start(serialized={}, prompts=[])
tracer.on_llm_end(response=LLMResult(generations=[[]]))
tracer.on_tool_end("test")
tracer.on_llm_start(serialized={}, prompts=[])
tracer.on_llm_end(response=LLMResult(generations=[[]]))
tracer.on_chain_end(outputs={})
def _add_child_run(
parent_run: Union[ChainRun, ToolRun],
child_run: Union[LLMRun, ChainRun, ToolRun],
) -> None:
"""Add child run to a chain run or tool run."""
parent_run.child_runs.append(child_run)
def _generate_id() -> Optional[Union[int, str]]:
"""Generate an id for a run."""
return None
def load_session(session_name: str) -> TracerSession:
"""Load a tracing session."""
return TracerSession(id=1, name=session_name, start_time=datetime.utcnow())
@@ -121,7 +37,7 @@ def load_default_session() -> TracerSession:
return TracerSession(id=1, name="default", start_time=datetime.utcnow())
class FakeTracer(Tracer):
class FakeTracer(BaseTracer):
"""Fake tracer that records LangChain execution."""
def __init__(self) -> None:
@@ -133,58 +49,6 @@ class FakeTracer(Tracer):
"""Persist a run."""
self.runs.append(run)
def _add_child_run(
self,
parent_run: Union[ChainRun, ToolRun],
child_run: Union[LLMRun, ChainRun, ToolRun],
) -> None:
"""Add child run to a chain run or tool run."""
_add_child_run(parent_run, child_run)
def _generate_id(self) -> Optional[Union[int, str]]:
"""Generate an id for a run."""
return _generate_id()
def _persist_session(self, session: TracerSessionCreate) -> TracerSession:
"""Persist a tracing session."""
return _persist_session(session)
def load_session(self, session_name: str) -> TracerSession:
"""Load a tracing session."""
return load_session(session_name)
def load_default_session(self) -> TracerSession:
"""Load a tracing session."""
return load_default_session()
class FakeSharedTracer(SharedTracer):
"""Fake shared tracer that records LangChain execution."""
runs: List[Union[LLMRun, ChainRun, ToolRun]] = []
def _persist_run(self, run: Union[LLMRun, ChainRun, ToolRun]) -> None:
"""Persist a run."""
with self._lock:
self.runs.append(run)
def remove_runs(self) -> None:
"""Remove all runs."""
with self._lock:
self.runs = []
def _add_child_run(
self,
parent_run: Union[ChainRun, ToolRun],
child_run: Union[LLMRun, ChainRun, ToolRun],
) -> None:
"""Add child run to a chain run or tool run."""
_add_child_run(parent_run, child_run)
def _generate_id(self) -> Optional[Union[int, str]]:
"""Generate an id for a run."""
return _generate_id()
def _persist_session(self, session: TracerSessionCreate) -> TracerSession:
"""Persist a tracing session."""
return _persist_session(session)
@@ -201,12 +65,15 @@ class FakeSharedTracer(SharedTracer):
@freeze_time("2023-01-01")
def test_tracer_llm_run() -> None:
"""Test tracer on an LLM run."""
uuid = uuid4()
compare_run = LLMRun(
id=None,
uuid=str(uuid),
parent_uuid=None,
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=1,
child_execution_order=1,
serialized={},
prompts=[],
response=LLMResult(generations=[[]]),
@@ -216,20 +83,11 @@ def test_tracer_llm_run() -> None:
tracer = FakeTracer()
tracer.new_session()
tracer.on_llm_start(serialized={}, prompts=[])
tracer.on_llm_end(response=LLMResult(generations=[[]]))
tracer.on_llm_start(serialized={}, prompts=[], run_id=uuid)
tracer.on_llm_end(response=LLMResult(generations=[[]]), run_id=uuid)
assert tracer.runs == [compare_run]
@freeze_time("2023-01-01")
def test_tracer_llm_run_errors_no_session() -> None:
"""Test tracer on an LLM run without a session."""
tracer = FakeTracer()
with pytest.raises(TracerException):
tracer.on_llm_start(serialized={}, prompts=[])
@freeze_time("2023-01-01")
def test_tracer_llm_run_errors_no_start() -> None:
"""Test tracer on an LLM run without a start."""
@@ -237,18 +95,21 @@ def test_tracer_llm_run_errors_no_start() -> None:
tracer.new_session()
with pytest.raises(TracerException):
tracer.on_llm_end(response=LLMResult(generations=[[]]))
tracer.on_llm_end(response=LLMResult(generations=[[]]), run_id=uuid4())
@freeze_time("2023-01-01")
def test_tracer_multiple_llm_runs() -> None:
"""Test the tracer with multiple runs."""
uuid = uuid4()
compare_run = LLMRun(
id=None,
uuid=str(uuid),
parent_uuid=None,
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=1,
child_execution_order=1,
serialized={},
prompts=[],
response=LLMResult(generations=[[]]),
@@ -260,8 +121,8 @@ def test_tracer_multiple_llm_runs() -> None:
tracer.new_session()
num_runs = 10
for _ in range(num_runs):
tracer.on_llm_start(serialized={}, prompts=[])
tracer.on_llm_end(response=LLMResult(generations=[[]]))
tracer.on_llm_start(serialized={}, prompts=[], run_id=uuid)
tracer.on_llm_end(response=LLMResult(generations=[[]]), run_id=uuid)
assert tracer.runs == [compare_run] * num_runs
@@ -269,12 +130,15 @@ def test_tracer_multiple_llm_runs() -> None:
@freeze_time("2023-01-01")
def test_tracer_chain_run() -> None:
"""Test tracer on a Chain run."""
uuid = uuid4()
compare_run = ChainRun(
id=None,
uuid=str(uuid),
parent_uuid=None,
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=1,
child_execution_order=1,
serialized={},
inputs={},
outputs={},
@@ -284,20 +148,23 @@ def test_tracer_chain_run() -> None:
tracer = FakeTracer()
tracer.new_session()
tracer.on_chain_start(serialized={}, inputs={})
tracer.on_chain_end(outputs={})
tracer.on_chain_start(serialized={}, inputs={}, run_id=uuid)
tracer.on_chain_end(outputs={}, run_id=uuid)
assert tracer.runs == [compare_run]
@freeze_time("2023-01-01")
def test_tracer_tool_run() -> None:
"""Test tracer on a Tool run."""
uuid = uuid4()
compare_run = ToolRun(
id=None,
uuid=str(uuid),
parent_uuid=None,
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=1,
child_execution_order=1,
serialized={},
tool_input="test",
output="test",
@@ -308,8 +175,8 @@ def test_tracer_tool_run() -> None:
tracer = FakeTracer()
tracer.new_session()
tracer.on_tool_start(serialized={}, input_str="test")
tracer.on_tool_end("test")
tracer.on_tool_start(serialized={}, input_str="test", run_id=uuid)
tracer.on_tool_end("test", run_id=uuid)
assert tracer.runs == [compare_run]
@@ -318,21 +185,109 @@ def test_tracer_nested_run() -> None:
"""Test tracer on a nested run."""
tracer = FakeTracer()
tracer.new_session()
_perform_nested_run(tracer)
assert tracer.runs == [_get_compare_run()]
chain_uuid = uuid4()
tool_uuid = uuid4()
llm_uuid1 = uuid4()
llm_uuid2 = uuid4()
for _ in range(10):
tracer.on_chain_start(serialized={}, inputs={}, run_id=chain_uuid)
tracer.on_tool_start(
serialized={}, input_str="test", run_id=tool_uuid, parent_run_id=chain_uuid
)
tracer.on_llm_start(
serialized={}, prompts=[], run_id=llm_uuid1, parent_run_id=tool_uuid
)
tracer.on_llm_end(response=LLMResult(generations=[[]]), run_id=llm_uuid1)
tracer.on_tool_end("test", run_id=tool_uuid)
tracer.on_llm_start(
serialized={}, prompts=[], run_id=llm_uuid2, parent_run_id=chain_uuid
)
tracer.on_llm_end(response=LLMResult(generations=[[]]), run_id=llm_uuid2)
tracer.on_chain_end(outputs={}, run_id=chain_uuid)
compare_run = ChainRun(
uuid=str(chain_uuid),
error=None,
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=1,
child_execution_order=4,
serialized={},
inputs={},
outputs={},
session_id=TEST_SESSION_ID,
child_chain_runs=[],
child_tool_runs=[
ToolRun(
uuid=str(tool_uuid),
parent_uuid=str(chain_uuid),
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=2,
child_execution_order=3,
serialized={},
tool_input="test",
output="test",
action="{}",
session_id=TEST_SESSION_ID,
error=None,
child_chain_runs=[],
child_tool_runs=[],
child_llm_runs=[
LLMRun(
uuid=str(llm_uuid1),
parent_uuid=str(tool_uuid),
error=None,
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=3,
child_execution_order=3,
serialized={},
prompts=[],
response=LLMResult(generations=[[]]),
session_id=TEST_SESSION_ID,
)
],
),
],
child_llm_runs=[
LLMRun(
uuid=str(llm_uuid2),
parent_uuid=str(chain_uuid),
error=None,
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=4,
child_execution_order=4,
serialized={},
prompts=[],
response=LLMResult(generations=[[]]),
session_id=TEST_SESSION_ID,
),
],
)
assert tracer.runs == [compare_run] * 10
@freeze_time("2023-01-01")
def test_tracer_llm_run_on_error() -> None:
"""Test tracer on an LLM run with an error."""
exception = Exception("test")
uuid = uuid4()
compare_run = LLMRun(
id=None,
uuid=str(uuid),
parent_uuid=None,
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=1,
child_execution_order=1,
serialized={},
prompts=[],
response=None,
@@ -342,8 +297,8 @@ def test_tracer_llm_run_on_error() -> None:
tracer = FakeTracer()
tracer.new_session()
tracer.on_llm_start(serialized={}, prompts=[])
tracer.on_llm_error(exception)
tracer.on_llm_start(serialized={}, prompts=[], run_id=uuid)
tracer.on_llm_error(exception, run_id=uuid)
assert tracer.runs == [compare_run]
@@ -351,13 +306,16 @@ def test_tracer_llm_run_on_error() -> None:
def test_tracer_chain_run_on_error() -> None:
"""Test tracer on a Chain run with an error."""
exception = Exception("test")
uuid = uuid4()
compare_run = ChainRun(
id=None,
uuid=str(uuid),
parent_uuid=None,
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=1,
child_execution_order=1,
serialized={},
inputs={},
outputs=None,
@@ -367,8 +325,8 @@ def test_tracer_chain_run_on_error() -> None:
tracer = FakeTracer()
tracer.new_session()
tracer.on_chain_start(serialized={}, inputs={})
tracer.on_chain_error(exception)
tracer.on_chain_start(serialized={}, inputs={}, run_id=uuid)
tracer.on_chain_error(exception, run_id=uuid)
assert tracer.runs == [compare_run]
@@ -376,13 +334,16 @@ def test_tracer_chain_run_on_error() -> None:
def test_tracer_tool_run_on_error() -> None:
"""Test tracer on a Tool run with an error."""
exception = Exception("test")
uuid = uuid4()
compare_run = ToolRun(
id=None,
uuid=str(uuid),
parent_uuid=None,
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=1,
child_execution_order=1,
serialized={},
tool_input="test",
output=None,
@@ -393,8 +354,8 @@ def test_tracer_tool_run_on_error() -> None:
tracer = FakeTracer()
tracer.new_session()
tracer.on_tool_start(serialized={}, input_str="test")
tracer.on_tool_error(exception)
tracer.on_tool_start(serialized={}, input_str="test", run_id=uuid)
tracer.on_tool_error(exception, run_id=uuid)
assert tracer.runs == [compare_run]
@@ -405,37 +366,53 @@ def test_tracer_nested_runs_on_error() -> None:
tracer = FakeTracer()
tracer.new_session()
chain_uuid = uuid4()
tool_uuid = uuid4()
llm_uuid1 = uuid4()
llm_uuid2 = uuid4()
llm_uuid3 = uuid4()
for _ in range(3):
tracer.on_chain_start(serialized={}, inputs={})
tracer.on_llm_start(serialized={}, prompts=[])
tracer.on_llm_end(response=LLMResult(generations=[[]]))
tracer.on_llm_start(serialized={}, prompts=[])
tracer.on_llm_end(response=LLMResult(generations=[[]]))
tracer.on_tool_start(serialized={}, input_str="test")
tracer.on_llm_start(serialized={}, prompts=[])
tracer.on_llm_error(exception)
tracer.on_tool_error(exception)
tracer.on_chain_error(exception)
tracer.on_chain_start(serialized={}, inputs={}, run_id=chain_uuid)
tracer.on_llm_start(
serialized={}, prompts=[], run_id=llm_uuid1, parent_run_id=chain_uuid
)
tracer.on_llm_end(response=LLMResult(generations=[[]]), run_id=llm_uuid1)
tracer.on_llm_start(
serialized={}, prompts=[], run_id=llm_uuid2, parent_run_id=chain_uuid
)
tracer.on_llm_end(response=LLMResult(generations=[[]]), run_id=llm_uuid2)
tracer.on_tool_start(
serialized={}, input_str="test", run_id=tool_uuid, parent_run_id=chain_uuid
)
tracer.on_llm_start(
serialized={}, prompts=[], run_id=llm_uuid3, parent_run_id=tool_uuid
)
tracer.on_llm_error(exception, run_id=llm_uuid3)
tracer.on_tool_error(exception, run_id=tool_uuid)
tracer.on_chain_error(exception, run_id=chain_uuid)
compare_run = ChainRun(
id=None,
uuid=str(chain_uuid),
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=1,
child_execution_order=5,
serialized={},
session_id=TEST_SESSION_ID,
error=repr(exception),
inputs={},
outputs=None,
child_runs=[
child_llm_runs=[
LLMRun(
id=None,
uuid=str(llm_uuid1),
parent_uuid=str(chain_uuid),
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=2,
child_execution_order=2,
serialized={},
session_id=TEST_SESSION_ID,
error=None,
@@ -443,36 +420,45 @@ def test_tracer_nested_runs_on_error() -> None:
response=LLMResult(generations=[[]], llm_output=None),
),
LLMRun(
id=None,
uuid=str(llm_uuid2),
parent_uuid=str(chain_uuid),
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=3,
child_execution_order=3,
serialized={},
session_id=TEST_SESSION_ID,
error=None,
prompts=[],
response=LLMResult(generations=[[]], llm_output=None),
),
],
child_chain_runs=[],
child_tool_runs=[
ToolRun(
id=None,
uuid=str(tool_uuid),
parent_uuid=str(chain_uuid),
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=4,
child_execution_order=5,
serialized={},
session_id=TEST_SESSION_ID,
error=repr(exception),
tool_input="test",
output=None,
action="{}",
child_runs=[
child_llm_runs=[
LLMRun(
id=None,
uuid=str(llm_uuid3),
parent_uuid=str(tool_uuid),
start_time=datetime.utcnow(),
end_time=datetime.utcnow(),
extra={},
execution_order=5,
child_execution_order=5,
serialized={},
session_id=TEST_SESSION_ID,
error=repr(exception),
@@ -480,43 +466,10 @@ def test_tracer_nested_runs_on_error() -> None:
response=None,
)
],
child_llm_runs=[],
child_chain_runs=[],
child_tool_runs=[],
),
],
child_llm_runs=[],
child_chain_runs=[],
child_tool_runs=[],
)
assert tracer.runs == [compare_run] * 3
@freeze_time("2023-01-01")
def test_shared_tracer_nested_run() -> None:
"""Test shared tracer on a nested run."""
tracer = FakeSharedTracer()
tracer.new_session()
tracer.remove_runs()
_perform_nested_run(tracer)
assert tracer.runs == [_get_compare_run()]
@freeze_time("2023-01-01")
def test_shared_tracer_nested_run_multithreaded() -> None:
"""Test shared tracer on a nested run."""
tracer = FakeSharedTracer()
tracer.remove_runs()
tracer.new_session()
threads = []
num_threads = 10
for _ in range(num_threads):
thread = threading.Thread(target=_perform_nested_run, args=(tracer,))
thread.start()
threads.append(thread)
for thread in threads:
thread.join()
assert tracer.runs == [_get_compare_run()] * num_threads