Hello everyone,
We have a use case in our company where we want to use AI. It’s about the following, we have about 5000 different topics for blog posts or magazines like:
“Food & Beverage - All about sugar”
“Food & Beverage - Hail the potato!”
"Health - Foot Care for Diabetes
“Home & Garden - Garden: Preserving the harvest”
…and so on…
Our goal is to create search tags for each topic, which we can then store in each article. Our previous approach was to create a file with about 2000 tags, e.g. single words like “sugar”, “wheat christmas”, “tomatoes” and so on with about 2000 tags in it. We then went and created embeddings for our headings like “Food & Beverage - All about sugar” and used cosine similarity to calculate these against the embeddings of the 2000 tags like “sugar”, “christmas”, “tomatoes” and then had the 10 most similar ones output. This actually works quite well, but sometimes there are some outliers in the generated top 10 tags due to embeddings.
That’s why we now want to send our 5000 different headings to the Chat Completion Api so that it gives us exact tags for each heading. This worked best in our tests and the tags were very good.
Now my question is, is there a way how I can send 5000 headlines at once to the chat completion to get the generated tags from the AI? Because if I send them one by one then I am making 5000 requests to the API which would take forever. I also tried aysnc but again I can only send a maximum of 80 at a time which doesn’t really get me anywhere. How can this be solved?
My Code for async:
# Read the Excel file
df = pd.read_excel(file_path)
df['Query'] = df.apply(lambda row: f"{row['Branche']} - {row['Thema']}", axis=1)
queries = df['Query'].tolist()
print(queries)
test_queries = queries[100:1500]
task_description = """
Task: You need to provide the top 10 single-word tags in German for the given subject.
Instructions:
Deliver exactly 10 tags.
Each tag should be a single word.
The tags should be highly relevant and specific to the subject.
Example:
If the subject is "Künstliche Intelligenz":
KI
Maschinelleslernen
Algorithmen
Datenanalyse
Automatisierung
NeuronaleNetze
DeepLearning
Technologie
Innovation
Zukunft
Query: [Your subject here]
"""
async def get_completion(query):
async with aiohttp.ClientSession() as session:
async with session.post('https://api.openai.com/v1/chat/completions', json={
'model': 'gpt-4',
'messages': [
{'role': 'system', 'content': task_description},
{'role': 'user', 'content': query}
],
}, headers={
'Authorization': f'Bearer {openai.api_key}',
'Content-Type': 'application/json'
}) as response:
result = await response.json()
return result['choices'][0]['message']['content']
async def main():
start_time = time.time() # Start time
tasks = [get_completion(query) for query in test_queries]
responses = await asyncio.gather(*tasks)
end_time = time.time() # End time
print(f"Total time taken: {end_time - start_time:.2f} seconds")
with open("output_keywords.txt", "w", encoding="utf-8") as file:
for i, response in enumerate(responses):
file.write(f"{response}\n")
asyncio.run(main())
