Skip to content

vertexai: add frequency_penalty and presence_penalty #889

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 6 commits into from
Apr 30, 2025
Merged

vertexai: add frequency_penalty and presence_penalty #889

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
7cdff75
vertexai: add frequency and presence penalty parameters to model and …
thomas-corcoran Apr 24, 2025
2055de9
vertexai: enhance frequency and presence penalty descriptions in _Ver…
thomas-corcoran Apr 24, 2025
7a81018
vertexai: add frequency and presence penalty to model parameters
thomas-corcoran Apr 28, 2025
73bbeb7
Merge branch 'main' into missing-generation-config-params
thomas-corcoran Apr 28, 2025
7550333
vertexai: remove frequency and presence penalty parameters from model…
thomas-corcoran Apr 29, 2025
5ff5791
Merge branch 'main' into missing-generation-config-params
thomas-corcoran Apr 29, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 6 additions & 0 deletions libs/vertexai/langchain_google_vertexai/_base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -202,6 +202,12 @@ class _VertexAICommon(_VertexAIBase):
"Underlying model name."
temperature: Optional[float] = None
"Sampling temperature, it controls the degree of randomness in token selection."
frequency_penalty: Optional[float] = None
Copy link
Collaborator

@lkuligin lkuligin Apr 25, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

you need to extend _allow_model_args too, otherwise args sent to the invoke would be ignored

Copy link
Contributor Author

@thomas-corcoran thomas-corcoran Apr 28, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks. I added them to the gemma models -- that's the only place I see allowed_model_args used.

Copy link
Collaborator

@lkuligin lkuligin Apr 29, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I dont' think Gemma supports these params.

I'm sorry I wasn't clear, that's the place that should be modified:

langchain-google/libs/vertexai/langchain_google_vertexai/chat_models.py

Line 155 in aa25c69

_allowed_params = [

Copy link
Contributor Author

@thomas-corcoran thomas-corcoran Apr 29, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sorry. I'm clearly missing something. frequency_penalty and presence_penalty are already there, no?

"Positive values penalize tokens that repeatedly appear in the generated text, "
"decreasing the probability of repeating content."
presence_penalty: Optional[float] = None
"Positive values penalize tokens that already appear in the generated text, "
"increasing the probability of generating more diverse content."
max_output_tokens: Optional[int] = Field(default=None, alias="max_tokens")
"Token limit determines the maximum amount of text output from one prompt."
top_p: Optional[float] = None
Expand Down
16 changes: 14 additions & 2 deletions libs/vertexai/tests/unit_tests/test_chat_models.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -971,15 +971,27 @@ class Multiply(BaseModel):


def test_generation_config_gemini() -> None:
model = ChatVertexAI(model_name="gemini-pro", temperature=0.2, top_k=3)
model = ChatVertexAI(
model_name="gemini-pro",
temperature=0.2,
top_k=3,
frequency_penalty=0.2,
presence_penalty=0.6,
)
generation_config = model._generation_config_gemini(
temperature=0.3, stop=["stop"], candidate_count=2
temperature=0.3,
stop=["stop"],
candidate_count=2,
frequency_penalty=0.9,
presence_penalty=0.8,
)
expected = GenerationConfig(
stop_sequences=["stop"],
temperature=0.3,
top_k=3,
candidate_count=2,
frequency_penalty=0.9,
presence_penalty=0.8,
)
assert generation_config == expected

Expand Down
14 changes: 12 additions & 2 deletions libs/vertexai/tests/unit_tests/test_llm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,6 +62,8 @@ def test_vertexai_args_passed() -> None:
"temperature": 0,
"top_k": 10,
"top_p": 0.5,
"frequency_penalty": 0.2,
"presence_penalty": 0.3,
}

# Mock the library to ensure the args are passed correctly
Expand All @@ -78,7 +80,9 @@ def test_vertexai_args_passed() -> None:
mock_prediction_service.return_value.generate_content = mock_generate_content

llm = VertexAI(model_name="gemini-pro", **prompt_params)
response = llm.invoke(user_prompt, temperature=0.5)
response = llm.invoke(
user_prompt, temperature=0.5, frequency_penalty=0.5, presence_penalty=0.5
)
assert response == response_text
mock_generate_content.assert_called_once()

Expand All @@ -90,7 +94,13 @@ def test_vertexai_args_passed() -> None:
== "Hello"
)
expected = GenerationConfig(
candidate_count=1, temperature=0.5, top_p=0.5, top_k=10, max_output_tokens=1
candidate_count=1,
temperature=0.5,
top_p=0.5,
top_k=10,
max_output_tokens=1,
frequency_penalty=0.5,
presence_penalty=0.5,
)
assert (
mock_generate_content.call_args.kwargs["request"].generation_config
Expand Down
Loading