added top_p for google and BaseLLMClientApi

EyalPazz · Jul 1, 2023 · 6bbaf86 · 6bbaf86
1 parent 432b155
commit 6bbaf86
Show file tree

Hide file tree

Showing 3 changed files with 6 additions and 5 deletions.
diff --git a/llm_client/llm_api_client/base_llm_api_client.py b/llm_client/llm_api_client/base_llm_api_client.py
@@ -30,7 +30,7 @@ def __init__(self, config: LLMAPIClientConfig):
 
     @abstractmethod
     async def text_completion(self, prompt: str, model: Optional[str] = None, max_tokens: Optional[int] = None,
-                              temperature: Optional[float] = None, **kwargs) -> list[str]:
+                              temperature: Optional[float] = None,top_p :  Optional[float] = None, **kwargs) -> list[str]:
         raise NotImplementedError()
 
     async def embedding(self, text: str, model: Optional[str] = None, **kwargs) -> list[float]:

diff --git a/llm_client/llm_api_client/google_client.py b/llm_client/llm_api_client/google_client.py
@@ -33,10 +33,11 @@ def __init__(self, config: LLMAPIClientConfig):
         self._params = {AUTH_PARAM: self._api_key}
 
     async def text_completion(self, prompt: str, model: Optional[str] = None, max_tokens: Optional[int] = 64,
-                              temperature: Optional[float] = None, **kwargs) -> list[str]:
+                              temperature: Optional[float] = None,top_p: float = 0.95, **kwargs) -> list[str]:
         model = model or self._default_model
         kwargs[PROMPT_KEY] = {TEXT_KEY: prompt}
         kwargs[MAX_TOKENS_KEY] = kwargs.pop(MAX_TOKENS_KEY, max_tokens)
+        kwargs["topP"] = kwargs.pop("topP", top_p)
         kwargs["temperature"] = kwargs.pop("temperature", temperature)
         response = await self._session.post(self._base_url + model + ":" + COMPLETE_PATH,
                                             params=self._params,

diff --git a/tests/llm_api_client/google_client/test_google_client.py b/tests/llm_api_client/google_client/test_google_client.py
@@ -31,7 +31,7 @@ async def test_text_completion__sanity(mock_aioresponse, llm_client, params):
                       'Once upon a time, there was a young boy named Billy...']
     mock_aioresponse.assert_called_once_with(url, method='POST', params={AUTH_PARAM: llm_client._api_key},
                                              json={PROMPT_KEY: {TEXT_KEY: 'These are a few of my favorite'},
-                                                   MAX_TOKENS_KEY: 64,
+                                                   MAX_TOKENS_KEY: 64,"topP" : 0.95,
                                                    'temperature': None},
                                              headers=llm_client._headers,
                                              raise_for_status=True,
@@ -53,7 +53,7 @@ async def test_text_completion__override_model(mock_aioresponse, llm_client, par
                       'Once upon a time, there was a young boy named Billy...']
     mock_aioresponse.assert_called_once_with(url, method='POST', params={AUTH_PARAM: llm_client._api_key},
                                              json={PROMPT_KEY: {TEXT_KEY: 'These are a few of my favorite'},
-                                                   MAX_TOKENS_KEY: 64,
+                                                   MAX_TOKENS_KEY: 64,"topP" : 0.95,
                                                    'temperature': None},
                                              headers=llm_client._headers,
                                              raise_for_status=True,
@@ -76,7 +76,7 @@ async def test_text_completion__with_kwargs(mock_aioresponse, llm_client, params
                                              json={PROMPT_KEY: {TEXT_KEY: 'These are a few of my favorite'},
                                                    MAX_TOKENS_KEY: 10,
                                                    'temperature': None,
-                                                   'blabla': 'aaa'},
+                                                   'blabla': 'aaa',"topP" : 0.95},
                                              headers=llm_client._headers,
                                              raise_for_status=True,
                                              )