OpenAI Model chatgpt-4o-latest not available to select

[Neoony]

I once asked similar thing a while ago, when I could not see some OpenAI model to choose from the dropdown. (but then it appeared, or I just didnt see it)

Seems now I have the same kind of thing with chatgpt-4o-latest model

On latest 0.8.3

When I check the models API (using the python import OpenAI)

pip install openai

from openai import OpenAI

# Set your OpenAI API key directly
api_key = 'apikey'

# Initialize the client with the API key
client = OpenAI(api_key=api_key)

def list_chat_completion_models():
    try:
        # Use the new method to list models
        models = client.models.list()

        # Iterate over the models in the SyncPage object
        chat_completion_models = [model for model in models if 'gpt' in model.id]

        if chat_completion_models:
            print("Available Chat Completion Models:")
            for model in chat_completion_models:
                print(f"- {model.id}")
        else:
            print("No chat completion models found.")

    except Exception as e:
        print(f"Error fetching models: {e}")

if __name__ == "__main__":
    list_chat_completion_models()

But gpt-subtrans

Any idea why its missing? I can see that it is querying the models api

That model just seems much better in everything else I use it. Would love to use it here.

(Again why I would rather prefer to be able to type custom model for such situations, as alternative 😄 )

[Neoony]

I also updated my api key in subtrans to those newer project API keys (I still had the old user api key there) But also no change

[machinewrapped]

The model list is filtered by ids that start with "gpt" to exclude models like DALL-E, Whisper and the old Davinci/Babbage models. I can include "chatgpt-" too, but it looks like Open AI advise against using it for production use 🤔

[machinewrapped]

As a workaround you could use the Local Server provider and specify Server Address https://api.openai.com/ and Endpoint v1/chat/completions - you can specify the model name manually then😉

[Neoony]

The model list is filtered by ids that start with "gpt" to exclude models like DALL-E, Whisper and the old Davinci/Babbage models. I can include "chatgpt-" too, but it looks like Open AI advise against using it for production use 🤔

Yeah, Likely also because it keeps getting updated (it keeps changing) and probably also priority is not high so might be slower. But I have not noticed any issues at all using it.

https://lmarena.ai/ (the o1 works very differently, so not interested in that one for this use)

As a workaround you could use the Local Server provider and specify Server Address https://api.openai.com/ and Endpoint v1/chat/completions - you can specify the model name manually then😉

Ah right, in that case, thats good enough for me. Just tried and it works. However keep getting not matching lines / fuzzy translations (while some batches translate fine)

It was working almost perfectly with model gpt-4o-2024-08-06 and I didnt need any changes for my stuff using GPT to use this new model, but indeed this needs a lot more specific format from GPT, which I am not doing anywhere myself

Not sure if its the model or that something works differently when using Local Server (I see it also has a prompt template) Well I also used subtitles for a TV Series that I didnt translate yet, might also be the subtitles.

Gonna have to take a better look when I have more time. (I will also try my previous model on the Local Server method and on some subtitles I translated before to compare, just need to find the time)

But anyways, Thanks!

[Neoony]

Ok I think its just the model. It seems to like to stop earlier than the gpt-4o-2024-08-06 If I use local server with gpt-4o-2024-08-06, it seems to be ok

I was actually using quite huge batches (min 100 max 400), since these models can all output 16k For gpt-4o-2024-08-06 it worked quite perfectly, only sometimes it would fail (I know that then it has to retry much bigger part and some other disadvantages of big batches)

But seems like chatgpt-4o-latest just likes to stop early before translating the whole thing (just a reason stop) Even though it still could have used more output tokens. Seems to work a lot better with min 100 and max 200 batches Maybe would need some instructions tweaking (I might play around with it)

Its a lot slower though

Anyways, I just like to experiment on how the new models behave, whether they are intended for production or not :P

Anyways, closing this. Thanks!