feat(groq): losen restrictions on reasoning_effort, inject effort in meta, update tests (#32415)

Author: mdrxy

libs/partners/groq/langchain_groq/chat_models.py

@@ -6,15 +6,7 @@
 import warnings
 from collections.abc import AsyncIterator, Iterator, Mapping, Sequence
 from operator import itemgetter
-from typing import (
-    Any,
-    Callable,
-    Literal,
-    Optional,
-    TypedDict,
-    Union,
-    cast,
-)
+from typing import Any, Callable, Literal, Optional, TypedDict, Union, cast
 
 from langchain_core._api import deprecated
 from langchain_core.callbacks import (
@@ -46,10 +38,7 @@
     ToolMessage,
     ToolMessageChunk,
 )
-from langchain_core.output_parsers import (
-    JsonOutputParser,
-    PydanticOutputParser,
-)
+from langchain_core.output_parsers import JsonOutputParser, PydanticOutputParser
 from langchain_core.output_parsers.base import OutputParserLike
 from langchain_core.output_parsers.openai_tools import (
     JsonOutputKeyToolsParser,
@@ -60,23 +49,13 @@
 from langchain_core.outputs import ChatGeneration, ChatGenerationChunk, ChatResult
 from langchain_core.runnables import Runnable, RunnableMap, RunnablePassthrough
 from langchain_core.tools import BaseTool
-from langchain_core.utils import (
-    from_env,
-    get_pydantic_field_names,
-    secret_from_env,
-)
+from langchain_core.utils import from_env, get_pydantic_field_names, secret_from_env
 from langchain_core.utils.function_calling import (
     convert_to_openai_function,
     convert_to_openai_tool,
 )
 from langchain_core.utils.pydantic import is_basemodel_subclass
-from pydantic import (
-    BaseModel,
-    ConfigDict,
-    Field,
-    SecretStr,
-    model_validator,
-)
+from pydantic import BaseModel, ConfigDict, Field, SecretStr, model_validator
 from typing_extensions import Self
 
 from langchain_groq.version import __version__
@@ -122,7 +101,7 @@ class ChatGroq(BaseChatModel):
 
             See the `Groq documentation
             <https://console.groq.com/docs/reasoning#reasoning>`__ for more
-            details and a list of supported reasoning models.
+            details and a list of supported models.
         model_kwargs: Dict[str, Any]
             Holds any model parameters valid for create call not
             explicitly specified.
@@ -328,20 +307,15 @@ class Joke(BaseModel):
       overridden in ``reasoning_effort``.
 
     See the `Groq documentation <https://console.groq.com/docs/reasoning#reasoning>`__
-    for more details and a list of supported reasoning models.
+    for more details and a list of supported models.
     """
-    reasoning_effort: Optional[Literal["none", "default"]] = Field(default=None)
+    reasoning_effort: Optional[str] = Field(default=None)
     """The level of effort the model will put into reasoning. Groq will default to
-    enabling reasoning if left undefined. If set to ``none``, ``reasoning_format`` will
-    not apply and ``reasoning_content`` will not be returned.
-
-    - ``'none'``: Disable reasoning. The model will not use any reasoning tokens when
-      generating a response.
-    - ``'default'``: Enable reasoning.
+    enabling reasoning if left undefined.
 
     See the `Groq documentation
     <https://console.groq.com/docs/reasoning#options-for-reasoning-effort>`__ for more
-    details and a list of models that support setting a reasoning effort.
+    details and a list of options and models that support setting a reasoning effort.
     """
     model_kwargs: dict[str, Any] = Field(default_factory=dict)
     """Holds any model parameters valid for `create` call not explicitly specified."""
@@ -601,6 +575,11 @@ def _stream(
                     generation_info["system_fingerprint"] = system_fingerprint
                 service_tier = params.get("service_tier") or self.service_tier
                 generation_info["service_tier"] = service_tier
+                reasoning_effort = (
+                    params.get("reasoning_effort") or self.reasoning_effort
+                )
+                if reasoning_effort:
+                    generation_info["reasoning_effort"] = reasoning_effort
             logprobs = choice.get("logprobs")
             if logprobs:
                 generation_info["logprobs"] = logprobs
@@ -644,6 +623,11 @@ async def _astream(
                     generation_info["system_fingerprint"] = system_fingerprint
                 service_tier = params.get("service_tier") or self.service_tier
                 generation_info["service_tier"] = service_tier
+                reasoning_effort = (
+                    params.get("reasoning_effort") or self.reasoning_effort
+                )
+                if reasoning_effort:
+                    generation_info["reasoning_effort"] = reasoning_effort
             logprobs = choice.get("logprobs")
             if logprobs:
                 generation_info["logprobs"] = logprobs
@@ -714,6 +698,9 @@ def _create_chat_result(
             "system_fingerprint": response.get("system_fingerprint", ""),
         }
         llm_output["service_tier"] = params.get("service_tier") or self.service_tier
+        reasoning_effort = params.get("reasoning_effort") or self.reasoning_effort
+        if reasoning_effort:
+            llm_output["reasoning_effort"] = reasoning_effort
         return ChatResult(generations=generations, llm_output=llm_output)
 
     def _create_message_dicts(

libs/partners/groq/tests/integration_tests/test_chat_models.py

@@ -25,6 +25,8 @@
 )
 
 DEFAULT_MODEL_NAME = "openai/gpt-oss-20b"
+
+# gpt-oss doesn't support `reasoning_effort`
 REASONING_MODEL_NAME = "deepseek-r1-distill-llama-70b"
 
 
@@ -272,7 +274,7 @@ def test_reasoning_output_stream() -> None:
 def test_reasoning_effort_none() -> None:
     """Test that no reasoning output is returned if effort is set to none."""
     chat = ChatGroq(
-        model="qwen/qwen3-32b",  # Only qwen3 currently supports reasoning_effort
+        model="qwen/qwen3-32b",  # Only qwen3 currently supports reasoning_effort = none
         reasoning_effort="none",
     )
     message = HumanMessage(content="What is the capital of France?")
@@ -282,6 +284,79 @@ def test_reasoning_effort_none() -> None:
     assert "<think>" not in response.content and "<think/>" not in response.content
 
 
+@pytest.mark.parametrize("effort", ["low", "medium", "high"])
+def test_reasoning_effort_levels(effort: str) -> None:
+    """Test reasoning effort options for different levels."""
+    # As of now, only the new gpt-oss models support `'low'`, `'medium'`, and `'high'`
+    chat = ChatGroq(
+        model=DEFAULT_MODEL_NAME,
+        reasoning_effort=effort,
+    )
+    message = HumanMessage(content="What is the capital of France?")
+    response = chat.invoke([message])
+    assert isinstance(response, AIMessage)
+    assert isinstance(response.content, str)
+    assert len(response.content) > 0
+    assert response.response_metadata.get("reasoning_effort") == effort
+
+
+@pytest.mark.parametrize("effort", ["low", "medium", "high"])
+def test_reasoning_effort_invoke_override(effort: str) -> None:
+    """Test that reasoning_effort in invoke() overrides class-level setting."""
+    # Create chat with no reasoning effort at class level
+    chat = ChatGroq(
+        model=DEFAULT_MODEL_NAME,
+    )
+    message = HumanMessage(content="What is the capital of France?")
+
+    # Override reasoning_effort in invoke()
+    response = chat.invoke([message], reasoning_effort=effort)
+    assert isinstance(response, AIMessage)
+    assert isinstance(response.content, str)
+    assert len(response.content) > 0
+    assert response.response_metadata.get("reasoning_effort") == effort
+
+
+def test_reasoning_effort_invoke_override_different_level() -> None:
+    """Test that reasoning_effort in invoke() overrides class-level setting."""
+    # Create chat with reasoning effort at class level
+    chat = ChatGroq(
+        model=DEFAULT_MODEL_NAME,  # openai/gpt-oss-20b supports reasoning_effort
+        reasoning_effort="high",
+    )
+    message = HumanMessage(content="What is the capital of France?")
+
+    # Override reasoning_effort to 'low' in invoke()
+    response = chat.invoke([message], reasoning_effort="low")
+    assert isinstance(response, AIMessage)
+    assert isinstance(response.content, str)
+    assert len(response.content) > 0
+    # Should reflect the overridden value, not the class-level setting
+    assert response.response_metadata.get("reasoning_effort") == "low"
+
+
+def test_reasoning_effort_streaming() -> None:
+    """Test that reasoning_effort is captured in streaming response metadata."""
+    chat = ChatGroq(
+        model=DEFAULT_MODEL_NAME,
+        reasoning_effort="medium",
+    )
+    message = HumanMessage(content="What is the capital of France?")
+
+    chunks = list(chat.stream([message]))
+    assert len(chunks) > 0
+
+    # Find the final chunk with finish_reason
+    final_chunk = None
+    for chunk in chunks:
+        if chunk.response_metadata.get("finish_reason"):
+            final_chunk = chunk
+            break
+
+    assert final_chunk is not None
+    assert final_chunk.response_metadata.get("reasoning_effort") == "medium"
+
+
 #
 # Misc tests
 #