Merge pull request #929 from parea-ai/new-cookbook

new cookbook, set openai.api_type

Author: jalexanderII

parea/cookbook/evals_and_experiments/modify_dataset_before_experiment.py

@@ -0,0 +1,52 @@
+import os
+
+from dotenv import load_dotenv
+from openai import OpenAI
+
+from parea import Parea, trace
+from parea.evals.rag import context_query_relevancy_factory
+from parea.schemas import TestCase
+
+load_dotenv()
+
+client = OpenAI(api_key=os.environ.get("OPENAI_API_KEY"))
+p = Parea(api_key=os.getenv("PAREA_API_KEY"))
+p.wrap_openai_client(client)
+
+context_query_relevancy = context_query_relevancy_factory(model="gpt-4o", context_fields=["context"])
+
+
+@trace(eval_funcs=[context_query_relevancy])
+def run_experiment(question: str, context: str) -> str:
+    return (
+        client.chat.completions.create(
+            model="gpt-4o",
+            temperature=0,
+            messages=[{"role": "user", "content": f"Answer question using context. Context: {context}. Question: {question}"}],
+        )
+        .choices[0]
+        .message.content
+    )
+
+
+# You can fetch a dataset directly and then modify it to meet our needs before passing it to p.experiment.
+def rename_information_to_context(num_samples: int = 3):
+    dataset = p.get_collection("Example_Dataset_Name")
+    if dataset:
+        testcases: list[TestCase] = list(dataset.test_cases.values())
+        # Assume dataset looks like this:
+        # [
+        #     inputs={"information": "Some long document", "question": "What is X?"}, target="X is Y" ...
+        # ]
+        return [{"context": case.inputs["information"], "question": case.inputs["question"], "target": case.target} for case in testcases[:num_samples]]
+    return []
+
+
+def main():
+    data = rename_information_to_context()
+    experiment = p.experiment("My_Experiment_Name", func=run_experiment, data=data)
+    experiment.run()
+
+
+if __name__ == "__main__":
+    main()

parea/evals/utils.py

@@ -42,9 +42,12 @@ def safe_json_loads(s) -> dict:
 def call_openai(
     messages, model, temperature=1.0, max_tokens=None, top_p=1.0, frequency_penalty=0.0, presence_penalty=0.0, response_format=None, n=1, is_azure=False
 ) -> Union[str, List[str]]:
+    openai.api_type = "openai"
     if is_azure:
         from openai.lib.azure import AzureOpenAI
 
+        openai.api_type = "azure"
+
         completion = AzureOpenAI().chat.completions.create(
             model=model,
             messages=messages,
@@ -94,9 +97,11 @@ def call_openai(
 
 
 def embed(model, input, is_azure=False) -> List[float]:
+    openai.api_type = "openai"
     if is_azure:
         from openai.lib.azure import AzureOpenAI
 
+        openai.api_type = "azure"
         return AzureOpenAI().embeddings.create(model=model, input=input, encoding_format="float").data[0].embedding
     if openai_version.startswith("0."):
         return openai.Embedding.create(model=model, input=input, encoding_format="float").data[0]["embedding"]

pyproject.toml

@@ -6,7 +6,7 @@ build-backend = "poetry.core.masonry.api"
 [tool.poetry]
 name = "parea-ai"
 packages = [{ include = "parea" }]
-version = "0.2.171"
+version = "0.2.172"
 description = "Parea python sdk"
 readme = "README.md"
 authors = ["joel-parea-ai <[email protected]>"]