Async Compare GPT Responses to Dataframe

compare_gpt_responses_async.py

import numpy as np
import pandas as pd
import asyncio
import openai
from sentence_transformers import SentenceTransformer

openai.api_key = "YOUR-API-KEY"

model = SentenceTransformer('all-MiniLM-L6-v2')

async def fetch_gpt_responses(prompts: list[str]) -> list[str]:
    tasks = []
    for prompt in prompts:
        task = openai.ChatCompletion.acreate(
            engine="gpt-35-turbo",
            messages = [
                {"role":"user","content":f"{prompt}"},
            ],
            temperature=0.7,
            max_tokens=800,
            top_p=0.95,
            frequency_penalty=0,
            presence_penalty=0,
            stop=None
        )
        tasks.append(task)
    responses = await asyncio.gather(*tasks)
    return [response.choices[0].message['content'].strip() for response in responses]

def generate_embeddings(texts: list[str]) -> np.ndarray:
    return model.encode(texts, show_progress_bar=False)

def cosine_similarity(vec_a: np.ndarray, vec_b: np.ndarray) -> float:
    return np.dot(vec_a, vec_b) / (np.linalg.norm(vec_a) * np.linalg.norm(vec_b))

async def compare_responses_async(dataframe: pd.DataFrame) -> pd.DataFrame:
    prompts = dataframe['prompts'].tolist()
    original_responses = dataframe['responses'].tolist()
    gpt_responses = await fetch_gpt_responses(prompts)
    embeddings = generate_embeddings(original_responses + gpt_responses)
    original_embeddings, gpt_embeddings = np.split(embeddings, 2)
    similarity_scores = [cosine_similarity(orig_emb, gpt_emb) for orig_emb, gpt_emb in zip(original_embeddings, gpt_embeddings)]
    dataframe['gpt_responses'] = gpt_responses
    dataframe['similarity_scores'] = similarity_scores
    return dataframe

async def main():
    # Sample DataFrame for demonstration
    df = pd.DataFrame({
        'prompts': ['What is the capital of France?', 'What is the leading AutoML platform?'],
        'responses': ['Paris', 'DataRobot']
    })

    updated_df = await compare_responses_async(df)
    print(updated_df)

await main()