core: Cache serialized manifest for llms, chat models and prompts (#26404)

libs/core/langchain_core/language_models/chat_models.py

@@ -6,6 +6,7 @@ import json
 import uuid
 import warnings
 from abc import ABC, abstractmethod
+from functools import cached_property
 from operator import itemgetter
 from typing import (
     TYPE_CHECKING,
@@ -247,6 +248,10 @@ class BaseChatModel(BaseLanguageModel[BaseMessage], ABC):
         arbitrary_types_allowed=True,
     )
 
+    @cached_property
+    def _serialized(self) -> dict[str, Any]:
+        return dumpd(self)
+
     # --- Runnable methods ---
 
     @property
@@ -378,7 +383,7 @@ class BaseChatModel(BaseLanguageModel[BaseMessage], ABC):
                 self.metadata,
             )
             (run_manager,) = callback_manager.on_chat_model_start(
-                dumpd(self),
+                self._serialized,
                 [messages],
                 invocation_params=params,
                 options=options,
@@ -450,7 +455,7 @@ class BaseChatModel(BaseLanguageModel[BaseMessage], ABC):
             self.metadata,
         )
         (run_manager,) = await callback_manager.on_chat_model_start(
-            dumpd(self),
+            self._serialized,
             [messages],
             invocation_params=params,
             options=options,
@@ -551,7 +556,7 @@ class BaseChatModel(BaseLanguageModel[BaseMessage], ABC):
             param_string = str(sorted([(k, v) for k, v in params.items()]))
             # This code is not super efficient as it goes back and forth between
             # json and dict.
-            serialized_repr = dumpd(self)
+            serialized_repr = self._serialized
             _cleanup_llm_representation(serialized_repr, 1)
             llm_string = json.dumps(serialized_repr, sort_keys=True)
             return llm_string + "---" + param_string
@@ -613,7 +618,7 @@ class BaseChatModel(BaseLanguageModel[BaseMessage], ABC):
             self.metadata,
         )
         run_managers = callback_manager.on_chat_model_start(
-            dumpd(self),
+            self._serialized,
             messages,
             invocation_params=params,
             options=options,
@@ -705,7 +710,7 @@ class BaseChatModel(BaseLanguageModel[BaseMessage], ABC):
         )
 
         run_managers = await callback_manager.on_chat_model_start(
-            dumpd(self),
+            self._serialized,
             messages,
             invocation_params=params,
             options=options,

libs/core/langchain_core/language_models/llms.py

@@ -317,6 +317,10 @@ class BaseLLM(BaseLanguageModel[str], ABC):
             values["callbacks"] = values.pop("callback_manager", None)
         return values
 
+    @functools.cached_property
+    def _serialized(self) -> dict[str, Any]:
+        return dumpd(self)
+
     # --- Runnable methods ---
 
     @property
@@ -544,7 +548,7 @@ class BaseLLM(BaseLanguageModel[str], ABC):
                 self.metadata,
             )
             (run_manager,) = callback_manager.on_llm_start(
-                dumpd(self),
+                self._serialized,
                 [prompt],
                 invocation_params=params,
                 options=options,
@@ -609,7 +613,7 @@ class BaseLLM(BaseLanguageModel[str], ABC):
             self.metadata,
         )
         (run_manager,) = await callback_manager.on_llm_start(
-            dumpd(self),
+            self._serialized,
             [prompt],
             invocation_params=params,
             options=options,
@@ -931,7 +935,7 @@ class BaseLLM(BaseLanguageModel[str], ABC):
         if (self.cache is None and get_llm_cache() is None) or self.cache is False:
             run_managers = [
                 callback_manager.on_llm_start(
-                    dumpd(self),
+                    self._serialized,
                     [prompt],
                     invocation_params=params,
                     options=options,
@@ -950,7 +954,7 @@ class BaseLLM(BaseLanguageModel[str], ABC):
         if len(missing_prompts) > 0:
             run_managers = [
                 callback_managers[idx].on_llm_start(
-                    dumpd(self),
+                    self._serialized,
                     [prompts[idx]],
                     invocation_params=params,
                     options=options,
@@ -1168,7 +1172,7 @@ class BaseLLM(BaseLanguageModel[str], ABC):
             run_managers = await asyncio.gather(
                 *[
                     callback_manager.on_llm_start(
-                        dumpd(self),
+                        self._serialized,
                         [prompt],
                         invocation_params=params,
                         options=options,
@@ -1194,7 +1198,7 @@ class BaseLLM(BaseLanguageModel[str], ABC):
             run_managers = await asyncio.gather(
                 *[
                     callback_managers[idx].on_llm_start(
-                        dumpd(self),
+                        self._serialized,
                         [prompts[idx]],
                         invocation_params=params,
                         options=options,

libs/core/langchain_core/prompts/base.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import json
 from abc import ABC, abstractmethod
+from functools import cached_property
 from pathlib import Path
 from typing import (
     TYPE_CHECKING,
@@ -103,6 +104,10 @@ class BasePromptTemplate(
         arbitrary_types_allowed=True,
     )
 
+    @cached_property
+    def _serialized(self) -> dict[str, Any]:
+        return dumpd(self)
+
     @property
     def OutputType(self) -> Any:
         """Return the output type of the prompt."""
@@ -190,7 +195,7 @@ class BasePromptTemplate(
             input,
             config,
             run_type="prompt",
-            serialized=dumpd(self),
+            serialized=self._serialized,
         )
 
     async def ainvoke(
@@ -215,7 +220,7 @@ class BasePromptTemplate(
             input,
             config,
             run_type="prompt",
-            serialized=dumpd(self),
+            serialized=self._serialized,
         )
 
     @abstractmethod