feat(groq): openai-oss (#32411)

use new openai-oss for integration tests, set module-level testing model
names and improve robustness of tool tests
This commit is contained in:
Mason Daugherty 2025-08-05 14:18:56 -04:00 committed by GitHub
parent 4011257c25
commit 419c173225
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194

View File

@ -24,7 +24,8 @@ from tests.unit_tests.fake.callbacks import (
FakeCallbackHandlerWithChatStart, FakeCallbackHandlerWithChatStart,
) )
MODEL_NAME = "llama-3.3-70b-versatile" DEFAULT_MODEL_NAME = "openai/gpt-oss-20b"
REASONING_MODEL_NAME = "deepseek-r1-distill-llama-70b"
# #
@ -34,7 +35,7 @@ MODEL_NAME = "llama-3.3-70b-versatile"
def test_invoke() -> None: def test_invoke() -> None:
"""Test Chat wrapper.""" """Test Chat wrapper."""
chat = ChatGroq( chat = ChatGroq(
model=MODEL_NAME, model=DEFAULT_MODEL_NAME,
temperature=0.7, temperature=0.7,
base_url=None, base_url=None,
groq_proxy=None, groq_proxy=None,
@ -55,7 +56,7 @@ def test_invoke() -> None:
@pytest.mark.scheduled @pytest.mark.scheduled
async def test_ainvoke() -> None: async def test_ainvoke() -> None:
"""Test ainvoke tokens from ChatGroq.""" """Test ainvoke tokens from ChatGroq."""
chat = ChatGroq(model=MODEL_NAME, max_tokens=10) chat = ChatGroq(model=DEFAULT_MODEL_NAME, max_tokens=10)
result = await chat.ainvoke("Welcome to the Groqetship!", config={"tags": ["foo"]}) result = await chat.ainvoke("Welcome to the Groqetship!", config={"tags": ["foo"]})
assert isinstance(result, BaseMessage) assert isinstance(result, BaseMessage)
@ -65,7 +66,7 @@ async def test_ainvoke() -> None:
@pytest.mark.scheduled @pytest.mark.scheduled
def test_batch() -> None: def test_batch() -> None:
"""Test batch tokens from ChatGroq.""" """Test batch tokens from ChatGroq."""
chat = ChatGroq(model=MODEL_NAME, max_tokens=10) chat = ChatGroq(model=DEFAULT_MODEL_NAME, max_tokens=10)
result = chat.batch(["Hello!", "Welcome to the Groqetship!"]) result = chat.batch(["Hello!", "Welcome to the Groqetship!"])
for token in result: for token in result:
@ -76,7 +77,7 @@ def test_batch() -> None:
@pytest.mark.scheduled @pytest.mark.scheduled
async def test_abatch() -> None: async def test_abatch() -> None:
"""Test abatch tokens from ChatGroq.""" """Test abatch tokens from ChatGroq."""
chat = ChatGroq(model=MODEL_NAME, max_tokens=10) chat = ChatGroq(model=DEFAULT_MODEL_NAME, max_tokens=10)
result = await chat.abatch(["Hello!", "Welcome to the Groqetship!"]) result = await chat.abatch(["Hello!", "Welcome to the Groqetship!"])
for token in result: for token in result:
@ -87,7 +88,7 @@ async def test_abatch() -> None:
@pytest.mark.scheduled @pytest.mark.scheduled
async def test_stream() -> None: async def test_stream() -> None:
"""Test streaming tokens from Groq.""" """Test streaming tokens from Groq."""
chat = ChatGroq(model=MODEL_NAME, max_tokens=10) chat = ChatGroq(model=DEFAULT_MODEL_NAME, max_tokens=10)
for token in chat.stream("Welcome to the Groqetship!"): for token in chat.stream("Welcome to the Groqetship!"):
assert isinstance(token, BaseMessageChunk) assert isinstance(token, BaseMessageChunk)
@ -97,7 +98,7 @@ async def test_stream() -> None:
@pytest.mark.scheduled @pytest.mark.scheduled
async def test_astream() -> None: async def test_astream() -> None:
"""Test streaming tokens from Groq.""" """Test streaming tokens from Groq."""
chat = ChatGroq(model=MODEL_NAME, max_tokens=10) chat = ChatGroq(model=DEFAULT_MODEL_NAME, max_tokens=10)
full: Optional[BaseMessageChunk] = None full: Optional[BaseMessageChunk] = None
chunks_with_token_counts = 0 chunks_with_token_counts = 0
@ -136,7 +137,7 @@ async def test_astream() -> None:
def test_generate() -> None: def test_generate() -> None:
"""Test sync generate.""" """Test sync generate."""
n = 1 n = 1
chat = ChatGroq(model=MODEL_NAME, max_tokens=10) chat = ChatGroq(model=DEFAULT_MODEL_NAME, max_tokens=10)
message = HumanMessage(content="Hello", n=1) message = HumanMessage(content="Hello", n=1)
response = chat.generate([[message], [message]]) response = chat.generate([[message], [message]])
assert isinstance(response, LLMResult) assert isinstance(response, LLMResult)
@ -155,7 +156,7 @@ def test_generate() -> None:
async def test_agenerate() -> None: async def test_agenerate() -> None:
"""Test async generation.""" """Test async generation."""
n = 1 n = 1
chat = ChatGroq(model=MODEL_NAME, max_tokens=10, n=1) chat = ChatGroq(model=DEFAULT_MODEL_NAME, max_tokens=10, n=1)
message = HumanMessage(content="Hello") message = HumanMessage(content="Hello")
response = await chat.agenerate([[message], [message]]) response = await chat.agenerate([[message], [message]])
assert isinstance(response, LLMResult) assert isinstance(response, LLMResult)
@ -178,7 +179,7 @@ def test_invoke_streaming() -> None:
"""Test that streaming correctly invokes on_llm_new_token callback.""" """Test that streaming correctly invokes on_llm_new_token callback."""
callback_handler = FakeCallbackHandler() callback_handler = FakeCallbackHandler()
chat = ChatGroq( chat = ChatGroq(
model=MODEL_NAME, model=DEFAULT_MODEL_NAME,
max_tokens=2, max_tokens=2,
streaming=True, streaming=True,
temperature=0, temperature=0,
@ -195,7 +196,7 @@ async def test_agenerate_streaming() -> None:
"""Test that streaming correctly invokes on_llm_new_token callback.""" """Test that streaming correctly invokes on_llm_new_token callback."""
callback_handler = FakeCallbackHandlerWithChatStart() callback_handler = FakeCallbackHandlerWithChatStart()
chat = ChatGroq( chat = ChatGroq(
model=MODEL_NAME, model=DEFAULT_MODEL_NAME,
max_tokens=10, max_tokens=10,
streaming=True, streaming=True,
temperature=0, temperature=0,
@ -222,7 +223,7 @@ async def test_agenerate_streaming() -> None:
def test_reasoning_output_invoke() -> None: def test_reasoning_output_invoke() -> None:
"""Test reasoning output from ChatGroq with invoke.""" """Test reasoning output from ChatGroq with invoke."""
chat = ChatGroq( chat = ChatGroq(
model="deepseek-r1-distill-llama-70b", model=REASONING_MODEL_NAME,
reasoning_format="parsed", reasoning_format="parsed",
) )
message = [ message = [
@ -241,7 +242,7 @@ def test_reasoning_output_invoke() -> None:
def test_reasoning_output_stream() -> None: def test_reasoning_output_stream() -> None:
"""Test reasoning output from ChatGroq with stream.""" """Test reasoning output from ChatGroq with stream."""
chat = ChatGroq( chat = ChatGroq(
model="deepseek-r1-distill-llama-70b", model=REASONING_MODEL_NAME,
reasoning_format="parsed", reasoning_format="parsed",
) )
message = [ message = [
@ -300,7 +301,7 @@ def test_streaming_generation_info() -> None:
callback = _FakeCallback() callback = _FakeCallback()
chat = ChatGroq( chat = ChatGroq(
model=MODEL_NAME, model="llama-3.1-8b-instant", # Use a model that properly streams content
max_tokens=2, max_tokens=2,
temperature=0, temperature=0,
callbacks=[callback], callbacks=[callback],
@ -314,7 +315,7 @@ def test_streaming_generation_info() -> None:
def test_system_message() -> None: def test_system_message() -> None:
"""Test ChatGroq wrapper with system message.""" """Test ChatGroq wrapper with system message."""
chat = ChatGroq(model=MODEL_NAME, max_tokens=10) chat = ChatGroq(model=DEFAULT_MODEL_NAME, max_tokens=10)
system_message = SystemMessage(content="You are to chat with the user.") system_message = SystemMessage(content="You are to chat with the user.")
human_message = HumanMessage(content="Hello") human_message = HumanMessage(content="Hello")
response = chat.invoke([system_message, human_message]) response = chat.invoke([system_message, human_message])
@ -324,7 +325,7 @@ def test_system_message() -> None:
def test_tool_choice() -> None: def test_tool_choice() -> None:
"""Test that tool choice is respected.""" """Test that tool choice is respected."""
llm = ChatGroq(model=MODEL_NAME) llm = ChatGroq(model=DEFAULT_MODEL_NAME)
class MyTool(BaseModel): class MyTool(BaseModel):
name: str name: str
@ -332,7 +333,7 @@ def test_tool_choice() -> None:
with_tool = llm.bind_tools([MyTool], tool_choice="MyTool") with_tool = llm.bind_tools([MyTool], tool_choice="MyTool")
resp = with_tool.invoke("Who was the 27 year old named Erick?") resp = with_tool.invoke("Who was the 27 year old named Erick? Use the tool.")
assert isinstance(resp, AIMessage) assert isinstance(resp, AIMessage)
assert resp.content == "" # should just be tool call assert resp.content == "" # should just be tool call
tool_calls = resp.additional_kwargs["tool_calls"] tool_calls = resp.additional_kwargs["tool_calls"]
@ -354,7 +355,7 @@ def test_tool_choice() -> None:
def test_tool_choice_bool() -> None: def test_tool_choice_bool() -> None:
"""Test that tool choice is respected just passing in True.""" """Test that tool choice is respected just passing in True."""
llm = ChatGroq(model=MODEL_NAME) llm = ChatGroq(model=DEFAULT_MODEL_NAME)
class MyTool(BaseModel): class MyTool(BaseModel):
name: str name: str
@ -362,7 +363,7 @@ def test_tool_choice_bool() -> None:
with_tool = llm.bind_tools([MyTool], tool_choice=True) with_tool = llm.bind_tools([MyTool], tool_choice=True)
resp = with_tool.invoke("Who was the 27 year old named Erick?") resp = with_tool.invoke("Who was the 27 year old named Erick? Use the tool.")
assert isinstance(resp, AIMessage) assert isinstance(resp, AIMessage)
assert resp.content == "" # should just be tool call assert resp.content == "" # should just be tool call
tool_calls = resp.additional_kwargs["tool_calls"] tool_calls = resp.additional_kwargs["tool_calls"]
@ -379,7 +380,7 @@ def test_tool_choice_bool() -> None:
@pytest.mark.xfail(reason="Groq tool_choice doesn't currently force a tool call") @pytest.mark.xfail(reason="Groq tool_choice doesn't currently force a tool call")
def test_streaming_tool_call() -> None: def test_streaming_tool_call() -> None:
"""Test that tool choice is respected.""" """Test that tool choice is respected."""
llm = ChatGroq(model=MODEL_NAME) llm = ChatGroq(model=DEFAULT_MODEL_NAME)
class MyTool(BaseModel): class MyTool(BaseModel):
name: str name: str
@ -417,7 +418,7 @@ def test_streaming_tool_call() -> None:
@pytest.mark.xfail(reason="Groq tool_choice doesn't currently force a tool call") @pytest.mark.xfail(reason="Groq tool_choice doesn't currently force a tool call")
async def test_astreaming_tool_call() -> None: async def test_astreaming_tool_call() -> None:
"""Test that tool choice is respected.""" """Test that tool choice is respected."""
llm = ChatGroq(model=MODEL_NAME) llm = ChatGroq(model=DEFAULT_MODEL_NAME)
class MyTool(BaseModel): class MyTool(BaseModel):
name: str name: str
@ -462,7 +463,9 @@ def test_json_mode_structured_output() -> None:
setup: str = Field(description="question to set up a joke") setup: str = Field(description="question to set up a joke")
punchline: str = Field(description="answer to resolve the joke") punchline: str = Field(description="answer to resolve the joke")
chat = ChatGroq(model=MODEL_NAME).with_structured_output(Joke, method="json_mode") chat = ChatGroq(model=DEFAULT_MODEL_NAME).with_structured_output(
Joke, method="json_mode"
)
result = chat.invoke( result = chat.invoke(
"Tell me a joke about cats, respond in JSON with `setup` and `punchline` keys" "Tell me a joke about cats, respond in JSON with `setup` and `punchline` keys"
) )
@ -476,38 +479,38 @@ def test_setting_service_tier_class() -> None:
message = HumanMessage(content="Welcome to the Groqetship") message = HumanMessage(content="Welcome to the Groqetship")
# Initialization # Initialization
chat = ChatGroq(model=MODEL_NAME, service_tier="auto") chat = ChatGroq(model=DEFAULT_MODEL_NAME, service_tier="auto")
assert chat.service_tier == "auto" assert chat.service_tier == "auto"
response = chat.invoke([message]) response = chat.invoke([message])
assert isinstance(response, BaseMessage) assert isinstance(response, BaseMessage)
assert isinstance(response.content, str) assert isinstance(response.content, str)
assert response.response_metadata.get("service_tier") == "auto" assert response.response_metadata.get("service_tier") == "auto"
chat = ChatGroq(model=MODEL_NAME, service_tier="flex") chat = ChatGroq(model=DEFAULT_MODEL_NAME, service_tier="flex")
assert chat.service_tier == "flex" assert chat.service_tier == "flex"
response = chat.invoke([message]) response = chat.invoke([message])
assert response.response_metadata.get("service_tier") == "flex" assert response.response_metadata.get("service_tier") == "flex"
chat = ChatGroq(model=MODEL_NAME, service_tier="on_demand") chat = ChatGroq(model=DEFAULT_MODEL_NAME, service_tier="on_demand")
assert chat.service_tier == "on_demand" assert chat.service_tier == "on_demand"
response = chat.invoke([message]) response = chat.invoke([message])
assert response.response_metadata.get("service_tier") == "on_demand" assert response.response_metadata.get("service_tier") == "on_demand"
chat = ChatGroq(model=MODEL_NAME) chat = ChatGroq(model=DEFAULT_MODEL_NAME)
assert chat.service_tier == "on_demand" assert chat.service_tier == "on_demand"
response = chat.invoke([message]) response = chat.invoke([message])
assert response.response_metadata.get("service_tier") == "on_demand" assert response.response_metadata.get("service_tier") == "on_demand"
with pytest.raises(ValueError): with pytest.raises(ValueError):
ChatGroq(model=MODEL_NAME, service_tier=None) # type: ignore[arg-type] ChatGroq(model=DEFAULT_MODEL_NAME, service_tier=None) # type: ignore[arg-type]
with pytest.raises(ValueError): with pytest.raises(ValueError):
ChatGroq(model=MODEL_NAME, service_tier="invalid") # type: ignore[arg-type] ChatGroq(model=DEFAULT_MODEL_NAME, service_tier="invalid") # type: ignore[arg-type]
def test_setting_service_tier_request() -> None: def test_setting_service_tier_request() -> None:
"""Test setting service tier defined at request level.""" """Test setting service tier defined at request level."""
message = HumanMessage(content="Welcome to the Groqetship") message = HumanMessage(content="Welcome to the Groqetship")
chat = ChatGroq(model=MODEL_NAME) chat = ChatGroq(model=DEFAULT_MODEL_NAME)
response = chat.invoke( response = chat.invoke(
[message], [message],
@ -537,7 +540,7 @@ def test_setting_service_tier_request() -> None:
# If an `invoke` call is made with no service tier, we fall back to the class level # If an `invoke` call is made with no service tier, we fall back to the class level
# setting # setting
chat = ChatGroq(model=MODEL_NAME, service_tier="auto") chat = ChatGroq(model=DEFAULT_MODEL_NAME, service_tier="auto")
response = chat.invoke( response = chat.invoke(
[message], [message],
) )
@ -564,7 +567,7 @@ def test_setting_service_tier_request() -> None:
def test_setting_service_tier_streaming() -> None: def test_setting_service_tier_streaming() -> None:
"""Test service tier settings for streaming calls.""" """Test service tier settings for streaming calls."""
chat = ChatGroq(model=MODEL_NAME, service_tier="flex") chat = ChatGroq(model=DEFAULT_MODEL_NAME, service_tier="flex")
chunks = list(chat.stream("Why is the sky blue?", service_tier="auto")) chunks = list(chat.stream("Why is the sky blue?", service_tier="auto"))
assert chunks[-1].response_metadata.get("service_tier") == "auto" assert chunks[-1].response_metadata.get("service_tier") == "auto"
@ -572,7 +575,7 @@ def test_setting_service_tier_streaming() -> None:
async def test_setting_service_tier_request_async() -> None: async def test_setting_service_tier_request_async() -> None:
"""Test async setting of service tier at the request level.""" """Test async setting of service tier at the request level."""
chat = ChatGroq(model=MODEL_NAME, service_tier="flex") chat = ChatGroq(model=DEFAULT_MODEL_NAME, service_tier="flex")
response = await chat.ainvoke("Hello!", service_tier="on_demand") response = await chat.ainvoke("Hello!", service_tier="on_demand")
assert response.response_metadata.get("service_tier") == "on_demand" assert response.response_metadata.get("service_tier") == "on_demand"