Skip to content

armgit / dspy

Public Imported from GitHub
0

DSPy: The framework for programming—not prompting—language models

Python

perf(dspy): defer anyio, openai and jiter to their use sites (#10381)

Adriaan Moorsb04a07335bb8
diff --git a/dspy/clients/engines/lifecycle.py b/dspy/clients/engines/lifecycle.py
index 45c543692..c966850cb 100644
--- a/dspy/clients/engines/lifecycle.py
+++ b/dspy/clients/engines/lifecycle.py
@@ -3,7 +3,9 @@
import inspect
from contextlib import asynccontextmanager, contextmanager
-import anyio
+from dspy.utils.lazy_import import require
+
+anyio = require("anyio")
def _secondary(primary, cleanup):
diff --git a/dspy/clients/execution.py b/dspy/clients/execution.py
index b1dc8c342..2b8759b8f 100644
--- a/dspy/clients/execution.py
+++ b/dspy/clients/execution.py
@@ -7,7 +7,6 @@ from contextlib import nullcontext
from dataclasses import dataclass, replace
from typing import Any
-import anyio
import pydantic
from dspy._vendor.lm15.result import StreamAccumulator
@@ -36,6 +35,9 @@ from dspy.lm15 import (
request_from_openai_chat,
)
from dspy.utils.exceptions import LMUnsupportedFeatureError, is_retryable_lm_error
+from dspy.utils.lazy_import import require
+
+anyio = require("anyio")
IGNORED_CACHE_KEYS = ["api_key", "api_base", "base_url"]
diff --git a/dspy/clients/lm.py b/dspy/clients/lm.py
index 0d0487b08..a3b30b1d2 100644
--- a/dspy/clients/lm.py
+++ b/dspy/clients/lm.py
@@ -4,10 +4,7 @@ import os
import re
import threading
import warnings
-from typing import Any, Literal, cast
-
-import anyio.from_thread
-from anyio.streams.memory import MemoryObjectSendStream
+from typing import TYPE_CHECKING, Any, Literal, cast
import dspy
from dspy.clients._litellm import get_litellm
@@ -21,9 +18,15 @@ from dspy.clients.utils_finetune import TrainDataFormat
from dspy.lm15 import CacheConfig
from dspy.utils.callback import BaseCallback
from dspy.utils.exceptions import LMConfigurationError, LMError, LMUnsupportedFeatureError
+from dspy.utils.lazy_import import require
from .base_lm import BaseLM
+anyio = require("anyio")
+
+if TYPE_CHECKING:
+ from anyio.streams.memory import MemoryObjectSendStream
+
logger = logging.getLogger(__name__)
@@ -452,7 +455,7 @@ def _get_stream_completion_fn(
return None
# The stream is already opened, and will be closed by the caller.
- stream = cast(MemoryObjectSendStream, stream)
+ stream = cast("MemoryObjectSendStream", stream)
caller_predict_id = id(caller_predict) if caller_predict else None
if dspy.settings.track_usage:
diff --git a/dspy/clients/openai.py b/dspy/clients/openai.py
index f687249d7..cd7c195c5 100644
--- a/dspy/clients/openai.py
+++ b/dspy/clients/openai.py
@@ -2,10 +2,11 @@ import time
from datetime import datetime
from typing import Any
-import openai
-
from dspy.clients.provider import Provider, TrainingJob
from dspy.clients.utils_finetune import TrainDataFormat, TrainingStatus, save_data
+from dspy.utils.lazy_import import require
+
+openai = require("openai")
class TrainingJobOpenAI(TrainingJob):
diff --git a/dspy/streaming/messages.py b/dspy/streaming/messages.py
index 0c58e19ec..19e1f915d 100644
--- a/dspy/streaming/messages.py
+++ b/dspy/streaming/messages.py
@@ -3,10 +3,11 @@ import concurrent.futures
from dataclasses import dataclass
from typing import Any
-import anyio.from_thread
-
from dspy.dsp.utils.settings import settings
from dspy.utils.callback import BaseCallback
+from dspy.utils.lazy_import import require
+
+anyio = require("anyio")
@dataclass
diff --git a/dspy/streaming/streamify.py b/dspy/streaming/streamify.py
index 5fd3ea436..b3a64e517 100644
--- a/dspy/streaming/streamify.py
+++ b/dspy/streaming/streamify.py
@@ -7,14 +7,15 @@ from queue import Queue
from typing import TYPE_CHECKING, Any, AsyncGenerator, Awaitable, Callable, Generator
import orjson
-from anyio import create_memory_object_stream, create_task_group
-from anyio.streams.memory import MemoryObjectSendStream
from dspy.dsp.utils.settings import settings
from dspy.primitives.prediction import Prediction
from dspy.streaming.messages import StatusMessage, StatusMessageProvider, StatusStreamingCallback
from dspy.streaming.streaming_listener import StreamListener, find_predictor_for_stream_listeners
from dspy.utils.asyncify import asyncify
+from dspy.utils.lazy_import import require
+
+anyio = require("anyio")
logger = logging.getLogger(__name__)
@@ -29,6 +30,8 @@ def _is_litellm_model_response_stream(value: Any) -> bool:
if TYPE_CHECKING:
+ from anyio.streams.memory import MemoryObjectSendStream
+
from dspy.primitives.module import Module
@@ -174,15 +177,15 @@ def streamify(
if not any(isinstance(c, StatusStreamingCallback) for c in callbacks):
callbacks.append(status_streaming_callback)
- async def generator(args, kwargs, stream: MemoryObjectSendStream):
+ async def generator(args, kwargs, stream: "MemoryObjectSendStream"):
with settings.context(send_stream=stream, callbacks=callbacks, stream_listeners=stream_listeners):
prediction = await program(*args, **kwargs)
await stream.send(prediction)
async def async_streamer(*args, **kwargs):
- send_stream, receive_stream = create_memory_object_stream(16)
- async with create_task_group() as tg, send_stream, receive_stream:
+ send_stream, receive_stream = anyio.create_memory_object_stream(16)
+ async with anyio.create_task_group() as tg, send_stream, receive_stream:
tg.start_soon(generator, args, kwargs, send_stream)
async for value in receive_stream:
diff --git a/dspy/streaming/streaming_listener.py b/dspy/streaming/streaming_listener.py
index d38112ab6..32e9f9eb5 100644
--- a/dspy/streaming/streaming_listener.py
+++ b/dspy/streaming/streaming_listener.py
@@ -6,14 +6,15 @@ from collections import defaultdict
from queue import Queue
from typing import TYPE_CHECKING, Any
-import jiter
-
from dspy.adapters.chat_adapter import ChatAdapter
from dspy.adapters.json_adapter import JSONAdapter
from dspy.adapters.types import Type
from dspy.adapters.xml_adapter import XMLAdapter
from dspy.dsp.utils.settings import settings
from dspy.streaming.messages import StreamResponse
+from dspy.utils.lazy_import import require
+
+jiter = require("jiter")
if TYPE_CHECKING:
from litellm import ModelResponseStream
diff --git a/dspy/utils/asyncify.py b/dspy/utils/asyncify.py
index 7f9259af6..9bc3ae435 100644
--- a/dspy/utils/asyncify.py
+++ b/dspy/utils/asyncify.py
@@ -1,8 +1,9 @@
import functools
from typing import TYPE_CHECKING, Any, Awaitable, Callable
-import anyio
-from anyio import CapacityLimiter
+from dspy.utils.lazy_import import require
+
+anyio = require("anyio")
if TYPE_CHECKING:
from dspy.primitives.module import Module
@@ -21,7 +22,7 @@ def get_limiter():
global _limiter
if _limiter is None:
- _limiter = CapacityLimiter(async_max_workers)
+ _limiter = anyio.CapacityLimiter(async_max_workers)
elif _limiter.total_tokens != async_max_workers:
_limiter.total_tokens = async_max_workers