nimble minnow
I'm using the OpenAI NPM package (https://npmjs.com/package/openai), with the configurations that I'll attach as images.
I'm currently having it connected to a discord bot, however, when I try to request any messages, its spitting out random text, and the text that it does spit out gets truncated.
(It's not the filter.clean method doing that, its the raw data)
dusky mauve
hello!
why don´t you complete the completion object?
maybe something as follows...
$curlbody = [
"prompt" => $this->sourceoftruth . $this->prompt . $history_string . $this->message . "\n" . $this->assistantname . ':',
"temperature" => (float) $temperature,
"max_tokens" => (int) $maxlength,
"top_p" => (float) $topp,
"frequency_penalty" => (float) $frequency,
"presence_penalty" => (float) $presence,
"stop" => $this->username . ":"
];
and I think it is better to use one of these engines:
'gpt-3.5-turbo-0301' => 'chat'
'gpt-3.5-turbo' => 'chat'
I hope it helps.
Regards!
nimble minnow
hello!
why don´t you complete the completion object?
maybe something as follo...
I have no idea what that is, is that a different langauge than Node?
I tried selefting gpt-3.5-turbo but it said that “this is a chat completion”
dusky mauve
ok ok... try something like this:
const promptText = ${sourceOfTruth}${prompt}${historyString}${message}\n${assistantName}:;
const completion = await openai.createCompletion({
model: 'gpt-3.5-turbo',
prompt: promptText,
temperature: parseFloat(temperature),
max_tokens: parseInt(maxlength, 10),
top_p: parseFloat(topp),
frequency_penalty: parseFloat(frequency),
presence_penalty: parseFloat(presence),
stop: [${username}:]
});
sourceOfTruth is not really needed
but I think you should give more context creating $prompt and $historyString