Selecting gpt 3.5 16k?

[w0010]

Is it possible to select the new gpt-3.5-turbo-16k? The drop-down menu only has gpt-3.5-turbo and gpt-4.

[brianpetro]

@w0010 I just pushed an update containing the new model.

Let me know how it works for you.

[w0010]

@w0010

Sounds like I might've missed on some token calculation.

In the developer console you should be able to some logs that contain something like "max tokens" when you send a message.

Let me know if you see any patterns in those logs when you get the error.

Thanks for your help. 🌴

Perhaps this is what you are looking for:

{
  "error": {
    "message": "This model's maximum context length is 16385 tokens. However, you requested 18982 tokens (8998 in the messages, 9984 in the completion). Please reduce the length of the messages or completion.",
    "type": "invalid_request_error",
    "param": "messages",
    "code": "context_length_exceeded"
  }
}

[S-Wil]

I have the same issue - in fact no matter which model I select I get an error saying I have submitted more tokens than the model can handle:
"message": "This model's maximum context length is 16385 tokens. However, you requested 19897 tokens (9978 in the messages, 9919 in the completion). Please reduce the length of the messages or completion.",
"message": "This model's maximum context length is 4097 tokens. However, you requested 5137 tokens (2691 in the messages, 2446 in the completion). Please reduce the length of the messages or completion.",

[brianpetro]

@w0010 @S-Wil thanks for your help in solving this.

Are your notes in English? What about use of other "special characters" in your notes?

I'll also be reviewing the token estimate calculations today to see if I overlooked something.

I'll probably also add another setting that scales down the number of max tokens, which will be necessary in the case of non-english languages and special character usage.

Thanks again for your help
Brian 🌴

[S-Wil]

All English for me. To provide some more context in case it's relevant: my notes are mostly all text. Minimal formatting, minimal code blocks, etc. About the only thing I can think of is that I do embed images (sometimes url links but mostly images locally stored in the vault).

[brianpetro]

@w0010 @S-Wil I just fixed a bug in the way the tokens are calculated.

This should resolve the issue.

Please do let me know if it keeps happening.

Thanks for your help in solving this
🌴

[S-Wil]

hmm now I am getting a different error:

[brianpetro]

Does it keep happening?

I traced this, and it looks like a prior request, get_context_hyde, failed to produce the expected output.

Thanks for your help in figuring this out
Brian 🌴

[S-Wil]

I have uninstalled your plugin, restarted obsidian several times, replaced the API key, created new notes, etc. and yes it keeps happening. I haven't rebooted yet but I'll get to that too. This message pops up in the UI when I send:

[brianpetro]

Can you give me an example message you are sending?

Have you tried changing the Smart Chat Model in the settings?

Thanks
🌴

[S-Wil]

Something probably unrelated but there's a bug in the text entry area where it shows this sometimes:

If I click on the tickbox it changes to the send button and then I can send it.

[brianpetro]

That's the "stop streaming response" button, similar to what's available in ChatGPT.

It might be getting stuck because of the errors, though.

[S-Wil]

The sample message is the one in the screenshot above. Weird - I changed models and I received one response. Then with the next prompt another error:

[brianpetro]

@S-Wil are you updated to 1.6.28?

I tried to address this issue in that update, however there could still be edge cases.

[S-Wil]

Doh - I updated about an hour ago but yes when I look at the version it was still on 1.6.27. I have updated now and it seems to be working. Sorry for the confusion. Thanks for your help with this. Great work with the plugin!

[brianpetro]

Thanks for following up!
🌴