#someone tried the new way to use the api

1 messages · Page 1 of 1 (latest)

tacit crag
#

I have ever tried, and it is absolutely better.

#

it is amazing

ripe glacier
#

u tried the old method as well? what is better in it?

tacit crag
ripe glacier
#

do you know why is that? cuz actually they use the same models and it looks like just easier code but not neccerly better

#

do you know if they made the create_context function better? the one that use cosine similarrity checks

tacit crag
ripe glacier
#

what was the intstructions u puted and the question u asked? im curious cuz i try build a chatbot for custom uses case.

tacit crag
ripe glacier
#

what is it mean?

tacit crag
#

Using that model, I have tested JSON creating task.

#

Using gpt-4 model, sometimes, it could not create json

#

but using gpt-4 turbo, it always created json

#

I felt improvement there so

ripe glacier
#

intresting..

tacit crag
#

yes,

ripe glacier
#

can i share with u piece of my code and u tell me what do u think about it and if better for me to move to the new api way?

tacit crag
ripe glacier
#

def generate_answer(
df,
question,
messages,
model="gpt-4",
max_len=500,
size="ada",
max_tokens=150,
stop_sequence=None,
):
try:
context,most_useful_link = create_context(question, df, max_len=max_len, size=size)
# Continue with further processing using 'context' and 'useful_link'
except Exception as e:
# Handle the error or exception as per your requirement
print(f"An error occurred: {e}")
# You can also log the error, raise a more specific exception, or take other appropriate actions

try:
    combined_message_content = f"You are the AI assistant, designed to provide helpful information and guidance on various topics. Please offer informative and concise responses based on the context provided in this conversation. If you encounter a question for which there is no specific guideline available, kindly respond with 'I don't know.'\n\nContext: {context}\n\nQuestion: {question}"
    messages.append({"role": "user", "content": combined_message_content})
    response = openai.ChatCompletion.create(
        model=model,
        messages=messages,
        temperature=0.1,
        max_tokens=max_tokens,
        stream=True
    )

    # Break the response into chunks and yield them
    for chunk in response:
        #print(chunk.choices[0].delta.get("content", ""),
        #end = "",
        #flush = True)
        yield chunk.choices[0].delta.get("content", "")
    # Append the link to the final chunk and yield it
    print(combined_message_content)
    final_chunk =f'\nRelated link: {most_useful_link}'
    yield final_chunk
tacit crag
#
  1. can you just replace model="gpt-4" to model="gpt-4-1106-preview"?
ripe glacier
#

ill check.

#

nah cant