openai v1.2

Author: iryna-kondr

pyproject.toml

@@ -6,12 +6,12 @@ build-backend = "setuptools.build_meta"
 dependencies = [
   "scikit-learn>=1.1.0",
   "pandas>=1.5.0",
-  "openai>=0.27.9",
+  "openai>=1.2.0",
   "tqdm>=4.60.0",
   "google-cloud-aiplatform>=1.27.0"
 ]
 name = "scikit-llm"
-version = "0.4.1"
+version = "1.0.0a1"
 authors = [
   { name="Oleg Kostromin", email="[email protected]" },
   { name="Iryna Kondrashchenko", email="[email protected]" },

skllm/llm/gpt/clients/openai/completion.py

@@ -1,4 +1,5 @@
 import openai
+from openai import OpenAI
 from skllm.llm.gpt.clients.openai.credentials import (
     set_azure_credentials,
     set_credentials,
@@ -36,15 +37,14 @@ def get_chat_completion(
     completion : dict
     """
     if api == "openai":
-        set_credentials(key, org)
-        model_dict = {"model": model}
+        client = set_credentials(key, org)
+        model_dict = {"model": model, "response_format": {"type": "json_object"}}
     elif api == "azure":
-        set_azure_credentials(key, org)
-        model_dict = {"engine": model}
+        client = set_azure_credentials(key, org)
+        model_dict = {"model": model}
     else:
         raise ValueError("Invalid API")
-
-    completion = openai.ChatCompletion.create(
+    completion = client.chat.completions.create(
         temperature=0.0, messages=messages, **model_dict
     )
     return completion

skllm/llm/gpt/clients/openai/credentials.py

@@ -1,62 +1,7 @@
 import openai
 from skllm.config import SKLLMConfig as _Config
 from time import sleep
-
-
-def get_chat_completion(
-    messages: dict,
-    key: str,
-    org: str,
-    model: str = "gpt-3.5-turbo",
-    max_retries: int = 3,
-    api="openai",
-):
-    """Gets a chat completion from the OpenAI API.
-
-    Parameters
-    ----------
-    messages : dict
-        input messages to use.
-    key : str
-        The OPEN AI key to use.
-    org : str
-        The OPEN AI organization ID to use.
-    model : str, optional
-        The OPEN AI model to use. Defaults to "gpt-3.5-turbo".
-    max_retries : int, optional
-        The maximum number of retries to use. Defaults to 3.
-    api : str
-        The API to use. Must be one of "openai" or "azure". Defaults to "openai".
-
-    Returns
-    -------
-    completion : dict
-    """
-    if api == "openai":
-        set_credentials(key, org)
-        model_dict = {"model": model}
-    elif api == "azure":
-        set_azure_credentials(key, org)
-        model_dict = {"engine": model}
-    else:
-        raise ValueError("Invalid API")
-    error_msg = None
-    error_type = None
-    for _ in range(max_retries):
-        try:
-            completion = openai.ChatCompletion.create(
-                temperature=0.0, messages=messages, **model_dict
-            )
-            return completion
-        except Exception as e:
-            error_msg = str(e)
-            error_type = type(e).__name__
-            sleep(3)
-    print(
-        f"Could not obtain the completion after {max_retries} retries: `{error_type} ::"
-        f" {error_msg}`"
-    )
-
+from openai import OpenAI, AzureOpenAI
 
 def set_credentials(key: str, org: str) -> None:
     """Set the OpenAI key and organization.
@@ -68,12 +13,8 @@ def set_credentials(key: str, org: str) -> None:
     org : str
         The OPEN AI organization ID to use.
     """
-    openai.api_key = key
-    openai.organization = org
-    openai.api_type = "open_ai"
-    openai.api_version = None
-    openai.api_base = "https://api.openai.com/v1"
-
+    client = OpenAI(api_key=key, organization=org)
+    return client
 
 def set_azure_credentials(key: str, org: str) -> None:
     """Sets OpenAI credentials for Azure.
@@ -85,9 +26,6 @@ def set_azure_credentials(key: str, org: str) -> None:
     org : str
         The OpenAI (Azure) organization ID to use.
     """
-    if not openai.api_type or not openai.api_type.startswith("azure"):
-        openai.api_type = "azure"
-    openai.api_key = key
-    openai.organization = org
-    openai.api_base = _Config.get_azure_api_base()
-    openai.api_version = _Config.get_azure_api_version()
+    client = AzureOpenAI(api_key=key, organization=org, api_version=_Config.get_azure_api_version(), azure_endpoint = _Config.get_azure_api_base())
+    return client
+    

skllm/llm/gpt/clients/openai/embedding.py

@@ -1,6 +1,7 @@
 from skllm.llm.gpt.clients.openai.credentials import set_credentials
 from skllm.utils import retry
 import openai
+from openai import OpenAI
 
 
 @retry(max_retries=3)
@@ -31,10 +32,11 @@ def get_embedding(
     emb : list
         The GPT embedding for the string.
     """
-    set_credentials(key, org)
+    client = OpenAI()
+    set_credentials(client, key, org)
     text = [str(t).replace("\n", " ") for t in text]
     embeddings = []
-    emb = openai.Embedding.create(input=text, model=model)
+    emb = client.embeddings.create(input=text, model=model)
     for i in range(len(emb["data"])):
         e = emb["data"][i]["embedding"]
         if not isinstance(e, list):

skllm/llm/gpt/mixin.py

@@ -151,9 +151,12 @@ def _extract_out_label(self, completion: Mapping[str, Any], **kwargs) -> Any:
         label : str
         """
         try:
-            label = extract_json_key(
-                completion["choices"][0]["message"]["content"], "label"
-            )
+            if hasattr(completion, "__getitem__"):
+                label = extract_json_key(
+                    completion["choices"][0]["message"]["content"], "label"
+                )
+            else:
+                label = extract_json_key(completion.choices[0].message.content, "label")
         except Exception as e:
             print(completion)
             print(f"Could not extract the label from the completion: {str(e)}")

skllm/models/_base/classifier.py

@@ -135,6 +135,9 @@ def _extract_labels(self, y) -> List[str]:
 
 
 class BaseClassifier(ABC, _SklBaseEstimator, _SklClassifierMixin):
+
+    system_msg = "You are a text classifier."
+
     def __init__(
         self,
         model: Optional[str],  # model can initially be None for tunable estimators
@@ -263,7 +266,7 @@ def _get_prompt(self, x: str) -> dict:
                 self.max_labels,
                 template=self._get_prompt_template(),
             )
-        return {"messages": prompt, "system_message": "You are a text classifier."}
+        return {"messages": prompt, "system_message": self.system_msg}
 
 
 class BaseFewShotClassifier(BaseClassifier):