llm fix

This reverts commit 1f364dcd81.

.github/scripts/check_diff.py

@@ -95,6 +95,11 @@ def _get_configs_for_single_dir(job: str, dir_: str) -> List[Dict[str, str]]:
         # declare deps in funny way
         max_python = "3.11"
 
+    if dir_ in ["libs/community", "libs/langchain"] and job == "extended-tests":
+        # community extended test resolution in 3.12 is slow
+        # even in uv
+        max_python = "3.11"
+
     return [
         {"working-directory": dir_, "python-version": min_python},
         {"working-directory": dir_, "python-version": max_python},

libs/core/langchain_core/rate_limiters.py

@@ -105,23 +105,31 @@ class InMemoryRateLimiter(BaseRateLimiter):
 
         .. code-block:: python
 
-            from langchain_core import InMemoryRateLimiter
+            import time
 
-            from langchain_core.runnables import RunnableLambda, InMemoryRateLimiter
+            from langchain_core.rate_limiters import InMemoryRateLimiter
 
             rate_limiter = InMemoryRateLimiter(
-                requests_per_second=100, check_every_n_seconds=0.1, max_bucket_size=10
+                requests_per_second=0.1,  # <-- Can only make a request once every 10 seconds!!
+                check_every_n_seconds=0.1,  # Wake up every 100 ms to check whether allowed to make a request,
+                max_bucket_size=10,  # Controls the maximum burst size.
             )
 
-            def foo(x: int) -> int:
-                return x
+            from langchain_anthropic import ChatAnthropic
+            model = ChatAnthropic(
+                model_name="claude-3-opus-20240229",
+                rate_limiter=rate_limiter
+            )
+
+            for _ in range(5):
+                tic = time.time()
+                model.invoke("hello")
+                toc = time.time()
+                print(toc - tic)
 
-            foo_ = RunnableLambda(foo)
-            chain = rate_limiter | foo_
-            assert chain.invoke(1) == 1
 
     .. versionadded:: 0.2.24
-    """
+    """  # noqa: E501
 
     def __init__(
         self,

libs/langchain/langchain/output_parsers/fix.py

@@ -3,10 +3,9 @@ from __future__ import annotations
 from typing import Any, TypeVar, Union
 
 from langchain_core.exceptions import OutputParserException
-from langchain_core.language_models import BaseLanguageModel
-from langchain_core.output_parsers import BaseOutputParser
+from langchain_core.output_parsers import BaseOutputParser, StrOutputParser
 from langchain_core.prompts import BasePromptTemplate
-from langchain_core.runnables import RunnableSerializable
+from langchain_core.runnables import Runnable, RunnableSerializable
 from typing_extensions import TypedDict
 
 from langchain.output_parsers.prompts import NAIVE_FIX_PROMPT
@@ -42,7 +41,7 @@ class OutputFixingParser(BaseOutputParser[T]):
     @classmethod
     def from_llm(
         cls,
-        llm: BaseLanguageModel,
+        llm: Runnable,
         parser: BaseOutputParser[T],
         prompt: BasePromptTemplate = NAIVE_FIX_PROMPT,
         max_retries: int = 1,
@@ -58,7 +57,7 @@ class OutputFixingParser(BaseOutputParser[T]):
         Returns:
             OutputFixingParser
         """
-        chain = prompt | llm
+        chain = prompt | llm | StrOutputParser()
         return cls(parser=parser, retry_chain=chain, max_retries=max_retries)
 
     def parse(self, completion: str) -> T:

libs/langchain/langchain/output_parsers/retry.py

@@ -4,7 +4,7 @@ from typing import Any, TypeVar, Union
 
 from langchain_core.exceptions import OutputParserException
 from langchain_core.language_models import BaseLanguageModel
-from langchain_core.output_parsers import BaseOutputParser
+from langchain_core.output_parsers import BaseOutputParser, StrOutputParser
 from langchain_core.prompt_values import PromptValue
 from langchain_core.prompts import BasePromptTemplate, PromptTemplate
 from langchain_core.runnables import RunnableSerializable
@@ -82,7 +82,7 @@ class RetryOutputParser(BaseOutputParser[T]):
         Returns:
             RetryOutputParser
         """
-        chain = prompt | llm
+        chain = prompt | llm | StrOutputParser()
         return cls(parser=parser, retry_chain=chain, max_retries=max_retries)
 
     def parse_with_prompt(self, completion: str, prompt_value: PromptValue) -> T:

libs/langchain/tests/unit_tests/output_parsers/test_fix.py

@@ -3,6 +3,7 @@ from typing import Any, Callable, Dict, Optional, TypeVar
 
 import pytest
 from langchain_core.exceptions import OutputParserException
+from langchain_core.messages import AIMessage
 from langchain_core.prompts.prompt import PromptTemplate
 from langchain_core.runnables import Runnable, RunnableLambda, RunnablePassthrough
 from pytest_mock import MockerFixture
@@ -63,6 +64,22 @@ def test_output_fixing_parser_parse(
     # TODO: test whether "instructions" is passed to the retry_chain
 
 
+def test_output_fixing_parser_from_llm() -> None:
+    def fake_llm(prompt: str) -> AIMessage:
+        return AIMessage("2024-07-08T00:00:00.000000Z")
+
+    llm = RunnableLambda(fake_llm)
+
+    n = 1
+    parser = OutputFixingParser.from_llm(
+        llm=llm,
+        parser=DatetimeOutputParser(),
+        max_retries=n,
+    )
+
+    assert parser.parse("not a date")
+
+
 @pytest.mark.parametrize(
     "base_parser",
     [

libs/partners/ollama/langchain_ollama/llms.py

@@ -305,9 +305,7 @@ class OllamaLLM(BaseLLM):
             if not isinstance(stream_resp, str):
                 chunk = GenerationChunk(
                     text=(
-                        stream_resp["message"]["content"]
-                        if "message" in stream_resp
-                        else ""
+                        stream_resp.get("response","")
                     ),
                     generation_info=(
                         dict(stream_resp) if stream_resp.get("done") is True else None
@@ -331,9 +329,7 @@ class OllamaLLM(BaseLLM):
             if not isinstance(stream_resp, str):
                 chunk = GenerationChunk(
                     text=(
-                        stream_resp["message"]["content"]
-                        if "message" in stream_resp
-                        else ""
+                        stream_resp.get("response","")
                     ),
                     generation_info=(
                         dict(stream_resp) if stream_resp.get("done") is True else None