Hi,
I have been trying to use Open AI APIs for text classification. From the playground, I am only able to get text-davinci-003 for completion but the result from this API are no where close to the chatGPT bot (chat.openai.com website). So I am trying to switch the model to gpt-3.5-turbo in the below code. But I get this error: InvalidRequestError: This is a chat model and not supported in the v1/completions endpoint. Did you mean to use v1/chat/completions?
Any guidance on how to use gpt-3.5-turbo? Basically, I want to replicate the chatGPT bot through API calls.
response = openai.Completion.create(
model="text-davinci-003",
prompt= prompt,
temperature=0.7,
max_tokens=400,
top_p=1,
frequency_penalty=0,
presence_penalty=0
)
Regards,
dbeings
same here would be great to use the turbo model on the completion api.
Yes, sure.
Go to the main OpenAI API reference page:
In your browser search bar, paste this:
Create chat completion
Then read the API does for the chat
completion and it’s easy.
It is important to read the docs.
HTH
Wes
March 14, 2023, 12:42am
4
Here’s sample code of my sentiment analysis function.
def sentiment_analysis(text):
MODEL = "gpt-3.5-turbo"
response = openai.ChatCompletion.create(
model = MODEL,
messages=[
{"role": "system", "content": "You are performing text analysis."},
{"role": "user", "content": "YOUR PROMPT GOES HERE: " + text},
],
temperature = 0,
max_tokens = 250,
)
message = response["choices"][0]["message"]["content"].strip().lower().translate(str.maketrans('', '', string.punctuation))
sentiment = None
if "yes" in message:
sentiment = "1"
elif "no" in message:
sentiment = "2"
else:
sentiment = "0"
return sentiment
This has essentially been my replacement to move over from the engine system, and functions essentially identically.
try to change end point to:
v1/chat/completions
This is the reference link OpenAI API
1 Like
Thanks. There were two issues at my end:
Was using Completion instead of ChatCompletion
ChatCompletion wasnt working for me as I was an earlier version of openai package. Updating it fixed the issues.
Regards,
dbeings
@databeings that worked for me.
Also useful to note that for NodeJS the method is createChatCompletion
.
And the parameters of the function have changed as per the api documentation and introduction guide .
Most notably is that it no longer uses prompt
but instead messages
which have a specific “chat format” .